Context-Adaptive Image Generation of Intangible Cultural Heritage Furniture for Architectural Interiors: A ComfyUI-Based AIGC Virtual Studio

Abstract

To address the challenge of efficiently and cost-effectively generating images of intangible cultural heritage (ICH) furniture that can adapt to diverse modern spatial contexts for visual communication, this paper proposes and constructs an Generative Artificial Intelligence (AIGC) virtual studio system based on ComfyUI. The system is designed for ICH furniture designers, cultural communicators, and digital preservation practitioners, aiming to overcome the bottlenecks of scene switching encountered in traditional photography and 3D modeling. First, furniture images and user scene descriptions are collected, and a dual lexicon consisting of AI prompts and user prompts is constructed. The analytic hierarchy process (AHP) is then applied to weight and filter prompt combinations, forming a quantifiable and integrated prompt system. Second, a visual workflow incorporating ControlNet and IPAdapter nodes is built in ComfyUI to enable the transfer of ICH furniture images to various preset spatial scenes. Finally, a Likert-scale comparison is conducted between the experimental group (using AHP-weighted prompts) and the control group (using unweighted prompts). The results show that the experimental group achieves significant improvements in image realism, style consistency, and cultural communication effectiveness. The images generated by this system can be directly used for digital display, e-commerce product pages, design proposals, and cultural archives of ICH furniture. The method is applicable to the context-aware AIGC generation of traditional furniture and home products, provided that a certain amount of image data and a ComfyUI environment are available. This study provides a reusable technical pathway for the modern visual presentation of ICH furniture and offers methodological support and empirical evidence for the integration of AIGC into environmental design.

1. Introduction

Intangible Cultural Heritage (ICH) furniture, as an important carrier of traditional Chinese arts and crafts, embodies unique historical memories and cultural symbols [1]. In digital consumption environments, ICH furniture requires high-quality contextualized images (e.g., display images in modern homes or commercial spaces) to reach younger audiences, thereby enhancing cultural identity and market conversion rates [2]. However, traditional physical photography and 3D modeling approaches suffer from high costs, long production cycles, and limited scene variety, making it difficult to meet the demand for high-frequency, diverse content creation.

Generative Artificial Intelligence (AIGC) technologies offer new possibilities for addressing the above problems. Image generation technologies, particularly diffusion models, are capable of producing high-quality scene images given specific prompts or reference images [3,4]. Nevertheless, directly applying general-purpose diffusion models to ICH furniture image generation presents two key challenges. First, prompts struggle to precisely control the morphological fidelity and stylistic consistency of the furniture, often resulting in detail distortion or cultural symbol misalignment [5]. Second, there is a lack of systematic adaptation mechanisms for diverse modern spatial contexts, leading to strong randomness in the generated outputs and insufficient controllability [6,7]. These challenges constitute the core research gap of this study.

Based on the above research gap, this study proposes the following three research questions (RQs):

RQ1 (Prompt construction problem): How to construct an integrated prompt framework that fuses AI prompts and user natural language prompts, and optimizes the weights using the Analytic Hierarchy Process (AHP), so as to achieve precise control over the cultural semantics in ICH furniture image generation [8,9]?

RQ2 (Workflow implementation problem): How to design a visual generation workflow based on the ComfyUI v1.6 platform that integrates “prompt control—structure preservation—style transfer”, enabling ICH furniture images to be efficiently transferred to diverse modern spatial contexts (e.g., new Chinese style living rooms, wabi-sabi tea rooms, light luxury studies) while maintaining morphological fidelity?

RQ3 (Effectiveness validation problem): Compared with pure AI prompts and pure user prompts, do images generated by AHP-weighted integrated prompts exhibit significant advantages in cultural expressiveness and aesthetic adaptability?

Addressing the above research questions, this study proposes the following core hypotheses (Hs):

H1. 

By constructing a tripartite AIGC workflow integrating “prompt control—structure preservation—style transfer”, it is possible to effectively achieve the contextual migration of ICH furniture from traditional static display to multi-scene embedded expression within indoor spaces (corresponding to RQ1 and RQ2).

H2. 

Images generated by integrated prompts weighted through the AHP are significantly superior to unweighted control groups in terms of cultural expressiveness and aesthetic adaptability (corresponding to RQ3).

Based on the above research questions and hypotheses(see

Table 1. Overview of research gap, research questions, hypotheses, and objectives.

Research Gap	Research Questions (RQs)	Core Hypotheses (Hs)	Research Objectives
Insufficient control over morphological fidelity and cultural semantics Lack of systematic adaptation to diverse modern spatial contexts	RQ1: How to build an AHP-weighted prompt framework combining AI and user language to precisely control cultural semantics? RQ2: How to design a ComfyUI-based workflow integrating prompt control, structure preservation, and style transfer while maintaining morphological fidelity? RQ3: Do AHP-weighted integrated prompts produce images with better cultural expressiveness and aesthetic adaptability than pure AI or user prompts?	H1: A tripartite workflow (prompt control—structure preservation—style transfer) effectively transfers ICH furniture to multi-scene contexts. H2: AHP-weighted integrated prompts generate images significantly superior to unweighted controls in cultural expressiveness and aesthetic adaptability.	Construct an AHP-weighted integrated prompt system Build a ComfyUI work-flow integrating CLIP en-coding, ControlNet, and IPAdapter Validate effectiveness via Likert-scale comparative experiments

), the research objectives of this study are: (1) to construct a quantifiable integrated prompt system that fuses AI and user language and is weighted through AHP; (2) to build a visual workflow based on ComfyUI that integrates CLIP encoding, ControlNet for structural control, and IPAdapter for style transfer [10]; (3) to validate the effectiveness and superiority of the proposed system through comparative experiments using Likert scales.

2. Core Issues and Theoretical Foundation

2.1. The Value and the “Heritage, Yet Contemporary” Dilemma of ICH Furniture in Architectural Interiors

Intangible cultural heritage (ICH) furniture refers to furniture categories rooted in traditional woodworking, lacquering, carving, and related ICH techniques, which simultaneously embody practical functions and cultural symbolic attributes. Typical examples include lacquered furniture, Su-style rosewood furniture, and Cantonese-style carved furniture (see

Table 2. Categories of ICH Furniture.

Name	Lacquer Furniture	Mortise-and-Tenon Furniture	Carving Craft Furniture	Weaving Structure Furniture	Inlay Craft Furniture	Craft Composite Furniture
Model
Class	Intangible cultural heritage crafts	Intangible cultural heritage crafts	Intangible cultural heritage crafts	Intangible cultural heritage crafts	Intangible cultural heritage crafts	Intangible cultural heritage crafts
Feature	The lacquered surface is as smooth as jade	requiring no nails or rivets, with a rigorous structure	It features diverse themes and luxurious styles	crafted through hand weaving for a light and delicate form	Embedded in wooden objects, it creates an exquisite effect	Integrating multiple techniques, it is extremely intricate.
Main Display Mode	Staged photographs, traditional static exhibitions	Demo video and technical diagram	Physical exhibits in the gallery, close-up photos	Natural Wind Home Space Coordination	Detail GIFs, Macro Space Renderings	Digital exhibitions with high-definition details
Name	Suzhou-style Furniture	Jin-style Furniture	Guang-style Furniture	Jing-style Furniture	Desks	Chairs
Model
Class	Regional cultural schools	Regional cultural schools	Regional cultural schools	Regional cultural schools	Function and Purpose	Function and Purpose
Feature	elegant style and well-proportioned	Sturdy and unadorned in form	Southern School Style, intricate carvings	Courtly style, dignified structure	Scholar’s desk furniture	Emphasizing posture and status symbol
Main Display Mode	Filming in an antique-style setting	Rural scene photography	Exhibition of court style and dynamic video presentation	The exhibition in the Palace Museum displays both scene and real scene	Complementing the Four Treasures of the Study decor	Chinese-style space rendering
Name	Bedding	Cabinet	Ritual and Religious Furniture	Traditional Replica Furniture	Innovative Design Furniture	Cross-border Fusion Furniture
Model
Class	Function and Purpose	Function and Purpose	Function and Purpose	Degree of Inheritance and Innovation	Degree of Inheritance and Innovation	Degree of Inheritance and Innovation
Feature	Exquisitely carved with rich layers	Pay attention to the layout and visual beauty	Functional, dignified in design, and meticulously crafted in materials	Highly faithful to ancient traditions, emphasizing historical authenticity	Preserving traditional elements through modern techniques	China-fashion, IP collaborations, and youthful forms
Main Display Mode	static physical display	Scene images and feature detail videos	Cultural and Folklore Performances, Physical Exhibitions	Photo of the object with a white background, along with a cultural commentary video	Modern home scene	Social media short videos

). The value of ICH furniture lies not only in the sophistication of materials and craftsmanship but also in the design philosophy that “utensils embody the Way”—i.e., through form, ornamentation, and structure, they convey Eastern aesthetics of living, ritual logic, and intergenerational spiritual heritage [11]. Consequently, ICH furniture is widely regarded as a living embodiment of traditional Chinese culture in contemporary life.

However, when ICH furniture is presented within modern consumer contexts where digital images serve as the primary medium, its cultural communication encounters a structural contradiction. Although consumers show significantly increased interest in products with Eastern cultural connotations, existing visual presentations of ICH furniture often fail to evoke aesthetic appreciation or purchase intention [12,13]. Recent advances in consumer aesthetic research provide theoretical support for understanding this contradiction. Buschgens, Figueiredo, and Blijlevens [14] proposed the aesthetic cultural precept of “heritage, yet contemporary” (i.e., rooted in tradition yet integrated with contemporaneity), which reveals a specific consumer aesthetic preference for product visual design: consumers prefer hybrid visual designs that achieve an optimal balance between “traditional heritage elements of the ancestral land” and “contemporary visual elements of the place of residence” [14].

Empirical evidence from social media platforms supports this observation. Using the QianGua Data v2.8.0 platform, this study collected Xiaohongshu App v9.29 posts published between 1 January 2023 and 31 December 2024, using the following keywords: “ICH furniture”, “traditional furniture”, “Chinese style furniture”, “vintage furniture”, “ICH styling”, “furniture styling”, and “new Chinese style furniture”. After deduplication and manual verification to ensure that the content involved ICH furniture, and excluding pure video posts and commercial promotions, a valid sample of 15,628 posts was obtained. Taking the Xiaohongshu platform as an example, analysis of the 15,628 ICH-furniture-related posts from 2023 to 2024 shows that user discussion engagement on the topic of “ICH furniture styling” increased by 214% year-on-year, yet the collection (save) rate was significantly lower than the like rate. High-frequency comments focused on visual context issues such as “the background looks dated”, “I don’t know how to place it in my home”, and “it doesn’t match modern decoration”. These comments indicate that consumers are not rejecting ICH furniture per se; rather, they struggle to form spatial imagination and aesthetic empathy due to the lack of appropriate contextualized images showing the furniture within plausible architectural interiors.

Theoretically, this phenomenon can be understood through the lens of semiotics and reception aesthetics. The “signifiers” of ICH furniture—their shapes, materials, and decorations—clearly point to traditional cultural “signifieds” in their original historical contexts. However, in modern living spaces, the original cultural context has largely dissolved. Without a new visual context to re-anchor their meanings, these signifiers become suspended in unfamiliar environments, leading to cognitive barriers or aesthetic rejection. From the perspective of reception aesthetics, the audience’s “horizon of expectation” is shaped by everyday exposure to contemporary interior design imagery [15,16]. When ICH furniture is presented without a coherent spatial setting, it fails to meet these expectations, resulting in a perception of the furniture as a “museum specimen” rather than a usable everyday object.

Therefore, for ICH furniture to transition from heritage artifacts to integrated elements of modern living environments, a systematic transfer and reconstruction of their image context within architectural interiors is required, following the “heritage, yet contemporary” principle. The core problem reduces to generating a set of high-quality, highly adaptable spatial images that enable viewers to re-recognize and embrace the cultural value of ICH furniture within familiar visual environments. This problem orientation provides a clear rationale and methodological basis for the AIGC virtual studio system introduced in subsequent sections [17,18,19].

2.2. Mechanisms of Communication Failure in ICH Furniture

2.2.1. Efficiency Bottleneck: Traditional Image Production Cannot Meet Market Demands for Diversity and Frequency

Currently, the visual communication of ICH furniture relies primarily on two technical approaches: physical photography and 3D modeling. These approaches lead to communication failures at two distinct levels—production efficiency and semantic communication—each of which is examined below. Physical photography requires on-site scene construction, with a single shoot costing between 800 and 2000 RMB (including venue, props, and labor). From lighting setup to final image output, the average time consumption ranges from 2 to 5 h, resulting in low production efficiency. More critically, due to cost constraints, approximately 60–80% of the output consists of simple white-background or single traditional scenes, which cannot meet the market demand for diverse image styles. Taking Sanfu Arts & Crafts Co., Ltd. (Xianyou, China) in Xianyou, Fujian Province as an example, its “Ming-style armchair” has been included in the provincial intangible cultural heritage (ICH) list. Nevertheless, in the company’s product image library for 2023, about 75% were warehouse white-background images, only 15% were set in traditional scenes, and the remaining 10% were close-up detail shots. This stylistic homogeneity has led to the loss of younger customer groups. To produce a set of scene images that integrate modern minimalist style with rosewood furniture, the company needs to invest an additional 3000–6000 RMB to build a temporary physical set, and the final images are still confined to a single “new Chinese style” template, making it difficult to adapt to diverse home decoration styles such as light luxury or Nordic styles. As a result, its online sales and social media promotion remain persistently constrained.

Although 3D modeling technology can enhance visual quality, the workflow is even more complex [20]. Taking a Qing-style carved marble-inlaid seat as an example, completing its high-precision modeling requires 12 core processes, including 3D scanning, point cloud processing, topology reconstruction, UV unwrapping, and material texture calibration. The total time typically ranges from 30 to 50 working hours. The restoration of details such as high-relief carvings and marble veining demands very high polygon counts and texture map resolutions. The workflow is illustrated in Figure 1. When a scene change is required—for instance, switching from a traditional hall to a modern space—the global illumination, reflectivity, and material texture parameters must be readjusted, adding approximately 10 to 20 additional working hours. This makes modeling approaches unable to meet the material update frequency required by e-commerce platforms, which typically expect new visuals every 7 to 10 days.

The above cost and time range data are derived from field research conducted by our research team in 2025 in ICH furniture industrial clusters such as Xianyou (Fujian Province) and Dongyang (Zhejiang Province). The data were obtained through semi-structured interviews with production managers and e-commerce operation staff from 12 enterprises, including Sanfu Arts & Crafts Co., Ltd. and Shuangyang Hongmu. Industry, (Dongyang, Zhejiang, China) benchmark data refer to the China Rosewood Furniture Annual Market and Outlook (2023) published by Wu Bingliang, Vice Chairman of the China National Furniture Association, as well as relevant industry analysis reports [21]. The data are presented as ranges to reflect reasonable fluctuations across different company sizes, regions, and technical conditions.

2.2.2. Semantic Failure: Lack of Spatial Context Disconnects Cultural Value from Audience Perception

A deeper issue is that even when high-quality furniture images are obtained through the above techniques, they generally lack effective integration with modern living scenarios. Nearly two-thirds of existing images focus solely on the physical structure of the furniture without placing it within a perceptible residential, commercial, or exhibition space. This “isolated presentation” prevents the cultural symbols of ICH furniture from establishing meaningful connections with the everyday spatial experiences of contemporary consumers.

From the perspective of communication effectiveness, when consumers search for cross-style keywords such as “ICH furniture + LOFT apartment” or “rosewood furniture + modern minimalism” on digital platforms, less than 20% of the content provides suitable visual solutions. This means that the abstract cultural values embodied by ICH furniture—such as “Eastern aesthetics” and “craftsmanship spirit”—cannot be easily perceived and embraced by younger audiences due to the lack of a concrete architectural interior context as an anchor. This phenomenon can be termed semantic failure: the image, as a communication medium, fails to successfully “translate” the cultural meaning of the product from its traditional context to a modern reception context [22].

From a theoretical perspective, the phenomenon of semantic failure can be more profoundly explained through the intersection of architectural language theory and structuralist methodology. Since the mid-20th century, the generative grammar theory represented by Noam Chomsky has provided important methodological insights for understanding the formal generative mechanisms of architecture. In his 1957 work Syntactic Structures, Chomsky proposed that language has a dual organizational hierarchy of deep structure and surface structure, and that a finite system of rules can generate an infinite number of legitimate expressions. This notion of “generativity” has profoundly influenced architecture. The Italian architectural theorist Franco Purini further introduced this line of thinking into architectural creation. In his foundational studies Una ipotesi di architettura (1966–1968), Purini explicitly regarded architecture as an autonomous language system based on grammatical and syntactic rules, noting that his linguistic conception is rooted in Chomsky’s generative grammar and the rationalist tradition. Purini argued that architectural drawing is not merely a tool for representation but a form of architectural thought and theoretical production—the autonomy and generativity of architectural grammar allow it to be systematically reconstructed in different contexts [23]. During the same period, Signs, Symbols, and Architecture (1980), co-edited by Geoffrey Broadbent, Charles Jencks, and Umberto Eco, systematically explored the theoretical framework of architectural semiotics, proposing that the formal and spatial elements of architecture can be regarded as a “sign system”. The generation of meaning depends on the viewer’s ability to decode cultural codes; when the cultural codes used for encoding and decoding do not match, a rupture of meaning occurs.

From this perspective, the “semantic failure” encountered by ICH furniture in digital communication is essentially a “cultural encoding-decoding mismatch” [24]. The “signifiers” of ICH furniture (form, pattern, material) point to clear cultural “signifieds” (ritual propriety, elegance, craftsmanship spirit) in the traditional context, forming a complete semiotic system [25]. However, when these pieces of furniture are placed in a modern consumption context mediated by digital images, the original cultural context has already been dissolved. Without a new visual context to re-anchor their meaning, the signifiers float in an unfamiliar environment, leading to cognitive obstacles or aesthetic rejection. In other words, for ICH furniture to achieve a cognitive transformation from “museum specimen” to “modern everyday object”, a systematic migration and reconstruction of image context must be accomplished. This is precisely the theoretical starting point of the tripartite workflow—“prompt control—structure preservation—style transfer”—proposed in this study: precise encoding of cultural semantics through AHP-weighted prompts, faithful restoration of furniture form through ControlNet structural control, and semiotic alignment with modern spatial contexts through IPAdapter style transfer, thereby systematically addressing the semantic failure across the three dimensions of architectural linguistics: “syntax—semantics—pragmatics”.

In summary, the root cause of the communication dilemma of ICH furniture lies in the failure to achieve the optimal balance between traditional visual symbols and contemporary interior spatial contexts, as required by the “heritage, yet contemporary” aesthetic precept. From an interior design perspective, a Ming-style armchair that possesses perfect ambience in a traditional hall often appears incongruous when placed in a modern minimalist living room. The key to solving this problem is to generate high-quality interior scene images for designers, so that ICH furniture can present clear scale relationships and material combinations in specific spaces (e.g., new Chinese style living rooms, wabi-sabi tea rooms, light luxury studies), thereby transforming from an “isolated cultural artefact” into a “spatial element that can be integrated into the discourse of contemporary interior design”.

2.3. Generative AI and a New Design Pathway via ComfyUI

In recent years, rapid advances in generative artificial intelligence (AIGC) have begun to reshape the production of visual content for design communication. In visually intensive sectors such as furniture and interior design, AIGC image generation systems have demonstrated significant advantages in efficiency, multi-style output, and cost reduction [26], thereby overcoming many constraints of physical photography and physical set construction.

Industry examples illustrate this trend. Since 2021, IKEA has increasingly adopted AI-generated furniture catalogue images to replace traditional studio photography, achieving greater product style diversity and environmental adaptability while substantially reducing production costs. Similarly, the Chinese home design platform Kujiale has launched an “AI Studio Shoot” product that rapidly generates e-commerce-ready furniture display images, shortening the cycle from product to visual content. However, these applications primarily target standardized, industrially produced modern furniture. Their training datasets and aesthetic frameworks are largely built upon Western modern design paradigms, and they exhibit limited understanding of traditional Eastern craftsmanship and the unique cultural semantics of intangible cultural heritage (ICH) furniture.

Specifically, existing AIGC solutions struggle to: (1) accurately represent the material textures and cultural connotations of mortise-and-tenon joinery, carving, and lacquerwork; (2) comprehend traditional Chinese spatial scales and ritualistic furniture arrangements within architectural interiors; and (3) convey the deeper philosophical notion that “utensils embody the Way” beyond superficial formal imitation. These limitations mean that while current AIGC tools are suitable for mass-market furniture, they cannot meet the higher demands of ICH furniture communication, which require cultural authenticity, aesthetic precision, and spatial contextualization [27].

Against this backdrop, ComfyUI—an open-source visual workflow platform—offers a new pathway for the context-aware visualization of ICH furniture within architectural interiors. Its node-based architecture enables users to flexibly construct an end-to-end workflow: from prompt encoding and structural control to style transfer and image enhancement. This forms a closed-loop production process of “design–generation–evaluation–redesign,” as illustrated in Figure 2. Compared to conventional black-box AIGC tools, ComfyUI provides greater process transparency, reproducibility, and fine-grained control. It can accurately respond to cultural and spatial semantic requirements, facilitating the evolution of ICH furniture presentation from static object display to contextualized scene presentation within diverse interior environments (e.g., historic residences, modern lofts, museum galleries).

By integrating structural preservation nodes (ControlNet) and style transfer nodes (IPAdapter), the proposed ComfyUI-based system overcomes the spatial, cost, and efficiency constraints of traditional photography and 3D modeling. It offers an efficient, controllable, and user-friendly technical pathway for the digital revitalization of cultural heritage, while directly supporting architectural and interior design practices that seek to incorporate ICH furniture into contemporary built environments [28].

3. Research Methods and Process

3.1. Design Concept and Generation Framework Analysis

This study is based on the open-source platform ComfyUI, aiming to construct a design pathway for virtual studio shooting scenes of ICH furniture by developing an automated scene migration system for such furniture. By integrating methods such as prompt engineering, structural control, and style transfer, it seeks to achieve the visual reconstruction and cultural contextual re-creation of traditional furniture within various modern spatial scenes. The overall process consists of the following six steps: (1) Data Collection: Gathering a substantial number of image materials related to ICH furniture from authoritative databases to establish a multi-source image-text corpus, providing foundational support for prompt construction and generation control [29]. (2) Constructing the Prompt System: Integrating AI prompt corpora and user perceptual language, and using the Analytic Hierarchy Process (AHP) for weight prioritization to form a structured and controllable prompt combination strategy [30]. (3) Building the Workflow: Based on prompt modules and the requirements for image structure control, establishing a standardized image generation workflow [31]. Simultaneously, introducing control models such as ControlNet, IPAdapter, and T2I-Adapter to achieve bidirectional control over furniture structure preservation and target space style transfer [32]. (4) Scene Migration: Employing AI prompts, user prompts, and integrated prompts within the workflow to generate three sets of scene migration samples. (5) Effect Evaluation and Scheme Validation: Conducting a comparative analysis of the three sets of generated images regarding quality, cultural expressiveness, and aesthetic suitability through expert evaluation and user questionnaires. This step validates the effectiveness of the prompt system and generation strategy, enabling multi-dimensional analysis and reverse optimization of the generation results. Consequently, it forms an operable and transferable image generation process for ICH furniture, as shown in Figure 3.

3.2. Constructing the Prompt System

In the process of generative AI-based image generation, the prompt is not only the core input that controls the semantic generation of the image but also largely determines the expressive direction and stylistic dimensions of the image content. Traditional AIGC prompts are mostly derived from model training corpora, offering high generality and strong stability. However, in culture-specific generation tasks such as ICH furniture, they often suffer from problems such as “semantic inaccuracy”, “cultural disconnection”, and “style misalignment with audience expectations” [33]. Meanwhile, although user natural language expressions are emotionally rich and culturally close, they often contain vagueness, redundancy, and structural ambiguity, making them difficult for models to interpret directly.

The essence of this dilemma can be understood as a kind of “encoding-decoding mismatch”. Any cultural sign system relies on specific grammatical rules and semantic hierarchies to achieve effective meaning transmission. Architectural language theory has long pointed out that spatial and formal elements constitute an autonomous sign system, and the generation of meaning depends on the decoder’s ability to interpret cultural codes—when the “cultural code” used for encoding does not match that used for decoding, a rupture of meaning occurs. Inspired by this insight, this study proposes the adoption of an “integrated prompt framework” that fuses AI prompts and user prompts, and employs scientific methods for screening and optimization, so as to improve the efficiency, semantic relevance, and structural rationality of prompt generation.

To construct a systematic, quantifiable, and culturally adaptive prompt system, this study introduces the Analytic Hierarchy Process (AHP) to provide decision support and weight optimization for the prompt schemes [34]. AHP is a multi-criteria decision-making method suitable for decomposing complex problems into hierarchical levels, prioritizing the importance of various factors through expert judgment and pairwise comparisons, thereby enabling quantitative analysis and optimal scheme selection. This methodology establishes an intrinsic correspondence with the aforementioned language theories. The decomposition of design decisions into a hierarchical structure in AHP echoes the generative mechanism from “deep structure to surface structure” in language—both approach the structured modeling of complex systems through finite rules. The “autonomy and generativity of formal language” emphasized by architectural grammar theory is embodied in this paper as the systematic modeling of prompt weights via AHP, transforming prompt design from empirical parameter tuning into evidence-based scientific decision-making. Meanwhile, the matching mechanism of “cultural codes” revealed by semiotics is ensured by the consistency check function of AHP (CR < 0.1)—the judgment matrix formed by experts through pairwise comparisons essentially provides a formal validation of the encoding rules of cultural semantics [35].

Based on the above theoretical framework, the construction process of the AHP prompt system consists of the following four key steps:

(1)

Construction of the Hierarchical Structure Model for ICH Furniture: The decision-making problem is divided into three levels—the goal level, the criterion level, and the alternative level—to form a clear hierarchical decision structure.

(2)

Construction of the Judgment Matrix: An expert panel from relevant fields such as AIGC design, ICH cultural communication, and user experience is invited to perform pairwise comparisons of various factors within the same level, establishing judgment matrices based on the standard AHP 1–9 scale method. The comparison scales for elements in the judgment matrix are shown in

Table 3. Comparison Scale of Elements in the Judgment Matrix.

Scale	Meaning
1	Equally important
3	Slightly important
5	Significantly important
7	Highly important
9	Extremely important
2, 4, 6, 8	The median of the two adjacent judgments
Count backwards	When comparing A and B on a 3-point scale, the ratio is 1/3.

.

(3)

Weight Calculation and Consistency Check: After normalizing the judgment matrix by columns and averaging the results, the weight of each judgment criterion is obtained. The formula for weight calculation is shown in Equation (1).

$${\mathrm{W}}_{\mathrm{i}}={\displaystyle \frac{{\left({\prod }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{a}}_{\mathrm{i}\mathrm{j}}\right)}^{\frac{1}{\mathrm{n}}}}{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\left({\prod }_{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{a}}_{\mathrm{i}\mathrm{j}}\right)}^{\frac{1}{\mathrm{n}}}}},\hspace{1em}\mathrm{i}=1,2,3,\dots ,\mathrm{n}$$

(1)

Subsequently, the Consistency Index (CI) and the Consistency Ratio (CR) are calculated. The Consistency Ratio (CR) is obtained by dividing the Consistency Index (CI) by the Random Consistency Index (RI). This ensures the judgment matrix meets the consistency requirement (CR < 0.1), guaranteeing the rationality of expert scores and the reliability of the results. If the CR value reaches or exceeds 0.1, experts must re-evaluate and modify the judgment matrix until the CR value falls below 0.1. The formula for CR is shown in Equation (2).

$$\mathrm{C}\mathrm{R}=\mathrm{C}\mathrm{I}/\mathrm{R}\mathrm{I}=({\mathsf{\lambda }}_{\mathrm{m}\mathrm{a}\mathrm{x}}-\mathrm{n})/\left(\right(\mathrm{n}-1\left)\mathrm{R}\mathrm{I}\right)<0.1$$

(2)

The formula for CI is shown in Equation (3).

$$\mathrm{C}\mathrm{I}=({\mathsf{\lambda }}_{\mathrm{m}\mathrm{a}\mathrm{x}}-\mathrm{n})/(\mathrm{n}-1)$$

(3)

(4)

Integrated Prompt Weighted Scoring and Output: Each prompt scheme is scored based on its performance under the criteria level. These scores are then combined with the respective weights for weighted calculation, yielding a comprehensive score for each scheme. The prompt combination with the highest score is identified as the optimal integrated prompt scheme, which serves as the standard input for the subsequent image generation workflow.

3.3. Construction of the ComfyUI Scene Migration System Based on Key Nodes: Prompt Control, Structure Stability, and Style Transfer

To transform the visual representation of ICH furniture from traditional static displays to multi-scenario embedded expressions, this study constructed an image generation system based on the ComfyUI platform, with a core pathway of “prompt control—structure preservation—style transfer.” The system’s design logic centers on semantic guidance, achieving controllability of image content, adaptability of spatial context, and high stability of generation results through the synergistic interaction of prompt vector encoding, structure control models, and style transfer modules. The entire system comprises five main modules: prompt input, structure stabilization, style transfer, image enhancement, and result output, with the first three constituting the generative core.

In the prompt control section, the system employs the CLIP Text Encode [36] node to encode the AHP-optimized integrated prompts and negative exclusion prompts into semantic vectors. These vectors guide the main content and stylistic direction of the generated images. Positive prompts cover furniture type, craft details, and target context, while negative prompts are used to avoid issues such as ambiguity, deformation, and background interference Meanwhile, the Empty Latent Image node is used to generate a latent space, providing a newly created blank background for image sampling. This mechanism ensures the accuracy of semantic expression of ICH furniture during the generation process—i.e., “what piece of furniture, in what space it is presented”—thereby laying the foundation for subsequent spatial context adaptation.

To ensure image structure stability and subsequent multi-scenario adaptation capability, the structure stabilization module comprises two major control workflows: a matting workflow and a stable composition workflow. First, the matting workflow aims for image structure isolation and portability. The system introduces a ControlNet control mechanism, loading pre-processing models such as the Canny edge detection node and the Depth node to perform edge extraction and depth analysis on the original furniture image. This ensures the accuracy and stability of the generated results in terms of structural outlines and spatial layout. Concurrently, it utilizes nodes like the Segment Anything Model (SAM) for semantic masking [37], background replacement, and transparent channel output to achieve high-precision separation of the furniture main body from its background, significantly enhancing the flexibility and visual effect of image material during scene migration. Second, to address issues such as unstable image composition and main subject position shift, the research further introduced a stable composition workflow based on the Stable Diffusion XL (SDXL) model. By combining modules like ControlNet and T2I-Adapter, precise control over the spatial position, perspective direction, and geometric proportions of the furniture main body is achieved. It also incorporates composition sketches, reference images, or structured prompts for visual focal point alignment, ensuring consistency, completeness, and compositional balance in batch scheme generation. This module resolves common spatial unrealistic issues in traditional image transfer, such as “floating furniture” and “scale distortion”, enabling the generated images to be naturally integrated into different interior scenes such as living rooms, studies, and tea rooms. Consequently, it enhances the visual credibility and spatial adaptability of ICH furniture within real built environments.

To achieve the natural integration and stylistic translation of ICH furniture images within various cultural contexts, the study constructed a scene style transfer module based on reference image guidance within the system. This module primarily relies on the IPAdapter and T2I-Adapter nodes within the ComfyUI platform [38,39]. By introducing spatial reference images, it extracts features such as spatial lighting, color tone atmosphere, and scene style, transferring these characteristics to the background space of the main generated image. This process allows ICH furniture to be naturally embedded into diversified contexts, enhancing the image’s cultural resonance and sense of realism. The images generated by this module enable ICH furniture to be naturally integrated into various interior style scenarios, with coordinated and harmonious relationships of scale, light, shadow, and color between the furniture and the spatial background. This effectively avoids the common issues of “furniture appearing isolated from the space” or “excessive collage-like feel” found in traditional images. As a result, it provides interior designers with more realistic and referable previews of spatial matching effects, reducing the cognitive cost in solution communication and product selection decisions.

After image generation is complete, the system enhances image quality through modules like VAE Decode [40], Latent Upscale, or ESRGAN super-resolution, achieving display-level clarity and detail restoration. Auxiliary nodes such as Color Adjust or Inpaint can also be optionally used for post-processing to optimize color, lighting, and local error correction. Finally, images are exported via the Save Image node, supporting batch saving and preview comparison, facilitating subsequent evaluation and selection.

In summary, through the tripartite working mechanism of “prompt semantic control—structural stability guarantee—scene style transfer,” this system constructs an AIGC scene migration generation pathway suitable for the digital dissemination of ICH furniture, as shown in Figure 4. This system features a clear workflow, flexible modules, and adjustable parameters. It not only achieves the contextual reconstruction of ICH image content but also provides a practical paradigm for the automated generation of cultural visual content.

4. Scene Migration System and Design Generation for ICH Furniture

This study, focusing on ICH furniture as the research subject, constructed a scene migration generation system based on the ComfyUI platform, aiming to verify the control capability and cultural expressive power of prompt combinations during the image generation process. The core innovation of this system lies in the introduction of a menu mechanism for prompts with adjustable weights. This enables designers and users to freely select and combine prompt elements based on semantic weights, flexibly steering the generation direction, thereby realizing an image generation workflow with clear cultural adaptability and style specificity.

4.1. Data Collection

To construct the ComfyUI-based scene migration system for ICH furniture, this study systematically collected images of ICH furniture encompassing various regional schools and craft types by integrating multi-source channels, including museum digital collections, catalogs from ICH protection centers, academic publications, and on-site photography. This effort established a comprehensive multi-source image-text corpus. The collection work also simultaneously acquired corresponding textual description data for the images, including structured information such as craft features, form semantics, and cultural symbolism. All images were uniformly processed into standard dimensions of 512 × 512 or 768 × 768 pixels. The semantic texts underwent cleaning, clustering, and structuring to form standardized corpus resources suitable for prompt engineering, providing a reliable data foundation for subsequent generation experiments.

4.2. Study on the Integrated Prompt Framework

4.2.1. Acquisition of AI Prompts

Regarding the acquisition of AI prompts, the study employs a three-stage dialogue strategy based on ChatGPT 4o, as shown in

Table 4. ChatGPT Questioning Interface.

Dialogue	Answer Questions with ChatGPT Messages
Dialogue 1	Q: You are a senior intangible cultural heritage (ICH) furniture designer. Please elaborate on the characteristics of ICH furniture.
	A: As a senior ICH furniture designer, I will systematically elaborate on the distinctive characteristics of ICH furniture from four dimensions: craftsmanship, materials, cultural connotation, and functional form. I. Craftsmanship Characteristics 1. Mortise and tenon joinery as the core. 2. Hand-carving and embellishment. 3. Unique surface treatment techniques. II. Material Characteristics 1. Selection of precious and fine woods. 2. The material philosophy of “working in harmony with the wood’s natural properties.” III. Cultural and Aesthetic Characteristics 1. Embodiment of Eastern philosophical thought. 2. Carrier of the core cultural essence of ICH. 3. Auspicious symbolism and emphasis on symmetry. IV. Functional Form Characteristics 1. Early application of ergonomic principles. 2. Rigorous proportions and scales. 3. Unity of function and form.
Dialogue 2	Q: Please naturally integrate the characteristics of the intangible cultural heritage furniture mentioned in the previous step into modern spatial settings and provide supplementary details.
	A: Certainly. The following is supplementary content that further adapts the aforementioned characteristic prompts of ICH furniture to modern-style spaces. It unfolds across four aspects: the integration of craftsmanship, material contrasts, the translation of form and function, and the creation of an artistic atmosphere reflecting cultural connotation. This aims to embody its design expressiveness and Eastern sensibility within contemporary settings: I. Integration of Craftsmanship 1. Structural Aestheticization of Mortise and Tenon: Featuring deconstructed or suspended mortise and tenon joints as design elements. 2. Implantation of the “Warmth” from Handcrafted Traces: Preserving marks of hand-planing, carving tool textures, and the unique patterns formed by the natural flow of lacquer. II. Material Contrasts 1. Fusion of Precious Wood and Modern Materials: Combining a warm sandalwood or rosewood tea table with sleek, copper-plated stainless steel legs; or placing a substantial solid wood slab atop an acrylic or floating metal base. 2. Contrast in Texture and Color: Using neutral-colored background spaces to accentuate ICH furniture in natural wood tones or lacquer red/black hues. Employing soft furnishings made of cotton, linen, or wool to echo the natural feel of wood, while incorporating small accents of vibrant color to activate the space. III. Translation of Form and Function 1. “Invisible Inheritance” of Ergonomics: The inheritance lies not in rigid external forms but in the inherent(scientific nature) and comfort at its core. 2. Flexible Aesthetics of “One Object, Multiple Uses”: Exploring the “modular” and “multi-functional” potential inherent in ICH furniture. IV. Creation of an Artistic Atmosphere Reflecting Cultural Connotation 1. Suggestive Rather than Literal: By “translating” traditional cultural symbols into modern design language, creating an “Eastern artistic conception” that can be sensed rather than explicitly described. 2. Creating “Spiritual Nooks”: Cultivating tranquil, introspective corners that provide inhabitants with a spiritual haven to detach from the hustle and bustle and return to their inner selves.
Dialogue 3	Q: Please integrate the prompts from the above two stages into precise ComfyUI prompts.
	A: Below are the precise ComfyUI prompts resulting from the integration of the two stages: “Characteristics of ICH Furniture Itself” and “Descriptions of Modern Spatial Adaptation Scenarios.” The content is divided into three parts: a prompt format template, prompt categories, and an example. Prompt Format Template: [Subject + Craftsmanship + Color & Material + Spatial Type + Lighting & Atmosphere + Composition & Shot + Scene Details + Style Orientation + Output Quality] Prompt Categories: 1. Furniture Type & Form: Ming-style round-back armchair, Official’s hat armchair, Zen meditation chair, Lacquer altar table, Rectangular table (Tiao’an), Closed-door cupboard (Menhu chu), Elegant and simple form, Sculptural silhouette, Flowing lines, Massive and imposing structure. 2. Core Craftsmanship & Details: Natural raw lacquer, Gold tracing, Mother-of-pearl inlay, Gold coating (Zhaojin), Lacquer relief (Duilun), Polished sheen, Traditional lacquer decoration techniques, Mortise and tenon structure, Floating panel structure, Interlocking joinery, Hand-carved cloud and floral motifs, Fine openwork carving, Plain, unadorned surface, Beeswax polish highlighting wood grain, Traditional lacquer gloss, Warm patina from years of use. 3. Spatial Type: Modern minimalist space, Contemporary Oriental interior, New Chinese-style living room, Grey-toned bedroom, Wabi-sabi tea room, Loft dining area, Art space, Serene study, Modern loft, Tea room nook, Hotel-style lobby, Wabi-sabi aesthetic. 4. Lighting Environment: Soft diffused light, Linear lighting, Natural light streaming in, Dappled light and shadow, Warm grey background, Concrete wall, Localized spotlighting, Tranquil Eastern ambiance. 5. Color & Texture: Vermilion red, Profound black, Ochre, Gold trim, Low-saturation reddish brown, Warm-toned rosewood, Amber hue of huanghuali wood, Near-black rosewood, Earthy tone of aged oak, Satin-smooth wood surface, Tactile carved texture, Semi-matte finish, Mirror-like lacquer sheen, Stone texture, Linen fabric, Metal lines, Blank white walls, Raw wood backdrop. 6. Display Method: Freestanding placement, Artistic display, Geometric symmetry, Asymmetrical arrangement, Spatial emptiness (negative space), Focused arrangement. 7. Additional Scene Elements: Floor-to-ceiling windows, Modern pendant lamp, Ink wash painting, Ceramic flower vase, Indoor plants, Sheer curtains, Linear LED light strips. Example: Prompt: “(masterpiece, best quality, 8 k, photorealistic, detailed interior photography), A single Ming-style official’s hat armchair, elegant and minimalist form, (pronounced mortise and tenon joinery: 1.3), (smooth, undecorated surface: 1.1), (natural wood grain emphasized by beeswax polish: 1.4), warm, rich tones of rosewood, in a (serene study with wabi-sabi aesthetic: 1.3), with a textured plaster wall, placed on a natural jute rug, (a polished concrete side table: 1.1) nearby, (placed as a solitary focal point: 1.5) in the center of the room, eye-level shot, asymmetrical balance, (soft morning light streaming through a paper lantern: 1.5), creating gentle shadows that accentuate the chair’s form, a sense of serene tranquility and mindful contemplation, a stack of ancient-style books on the side table, a pot of delicate orchid on the floor, shallow depth of field.” Negative: “plastic-looking, cartoon style, low detail, sci-fi interior, messy composition”

, to construct a prompt system for ICH furniture across three levels: structural semantics, stylistic context, and combination optimization. First, by inputting “You are a senior ICH furniture designer. Please describe the characteristics of ICH furniture.”, structural prompts such as “lacquered wood texture, Ming-style cabinet, mother-of-pearl inlay” are obtained to highlight the furniture’s appearance and material features. Second, the model is guided to incorporate spatial contexts like “modern Oriental” or “wabi-sabi style,” generating supplementary stylistic prompts such as “placed in minimalist interior, soft warm light, ink-painting wall” to achieve contextual embedding. Finally, prompts from the above two stages are integrated into precise ComfyUI prompts, and negative prompts such as “blurry, deformed, oversaturated” are introduced to control image clarity and composition stability. Through multiple rounds of dialogue experiments and semantic screening, an AI prompt library covering four major dimensions—craft, style, and context—is ultimately constructed, providing foundational control language for subsequent image generation [41].

4.2.2. Acquisition of User Prompts

To obtain prompt expressions that reflect both cultural perception and user preference characteristics, this study adopted a combination of random sampling and snowball sampling to select 20 users who have long-term interest in ICH furniture or related spatial aesthetics. The specific sampling process was as follows. First, eight users were randomly selected as the initial “seed” sample, including two product operation or visual planning staff members, three design practitioners (one interior designer, one product designer, and one home furnishing buyer), and three general consumers. Subsequently, using the snowball sampling method, these eight respondents were asked to recommend two to three eligible friends or peers with similar interests. After two rounds of referrals, a total of 20 respondents were obtained. The demographic characteristics of the sample are as follows: ages ranged from 25 to 45 years, with 65% aged 25–35 and 35% aged 36–45; the gender ratio was approximately 1:1 (11 females, 9 males). In terms of professional backgrounds, there were five product operation or visual planning staff members (with more than three years of experience), eight design practitioners (including interior designers, product designers, and home furnishing buyers), and seven general consumers (with experience in home decoration or furniture purchasing). The interview content revolved around core topics including “perception of tradition,” “sense of spatial integration,” and “acceptance of modern aesthetics”, as shown in

Table 5. Outline of User Interviews.

Number	Interview Questions
1	What keywords or images come to mind when you hear about intangible cultural heritage furniture?
2	What kind of space do you think is best suited for intangible cultural heritage furniture? Please describe the style and atmosphere of this space in detail.
3	If an image of an intangible cultural heritage (ICH) furniture product appears in social media, live streaming backgrounds, or digital exhibition halls, what visual features do you think would make it more appealing?
4	Do you prefer scenes that are “embedded in living spaces” over cold white backgrounds? Why?
5	Here are some reference images. Which one do you think best represents your expectation of “combining modernity with tradition”? Please explain why.

. The collected data were ultimately summarized and organized using the KJ method (affinity diagram method) to extract commonly used descriptive terms and preferred imagery expressions by the users. Although this language carries subjective characteristics, it provides culturally descriptive materials rich in perceptual features, thereby offering an intuitive foundation for prompt integration.

By integrating user-need prompts with AI prompts, key criteria for the AHP process were obtained, as shown in

Table 6. Hierarchical Structure Model.

Target Layer (A)	Criterion Layer (B)	Scheme Layer (C)	Vocabulary Explanation
Creating High-quality Images of Intangible Cultural Heritage Furniture and Modern Space	Type and Shape	Techniques and Processes	The traditional handcrafts used in intangible cultural heritage furniture, such as mortise and tenon, carving and polishing, reflect the precision of craftsmanship and the spirit of craftsmanship.
		Regional School	The furniture style system formed in different regions, such as Suzhou, Guangzhou and Beijing, has its own unique structure layout and decoration style.
		Material Composition	The composition of materials used in furniture, such as wood, metal, lacquer, and wicker, is designed to reflect the overall texture and visual tone.
		Functional Form	The functional use and structural design of furniture, such as the size ratio, usage method and functional adaptability of chairs, desks, tables and cabinets.
	Process and Details	Wood Construction	Traditional wooden construction methods, including mortise-and-tenon joints, frame connections, and panel assembly, demonstrate both structural rationality and artistic craftsmanship.
		Surface Decoration	Decorative treatments on furniture surfaces, such as varnishing, polishing, topcoating, and waxing, are employed to enhance color, texture, and protective properties.
		Engraving and Inlay	The decorative details, including relief, openwork, inlay, and metal ornaments, enhance the artistic and cultural value of the furniture.
		Detail Modification	The artistic treatment of the details such as the edge, leg, foot, drawer, etc.
	Color and Material	Natural Wood Texture	The natural wood grain, color, and texture convey a warm, natural visual and tactile experience.
		Vermicelli Color System	The traditional color expression system with lacquer as the core includes black lacquer, vermilion, purple sauce, eggshell white and other Oriental visual languages.
		Metal Decorative Material	The furniture incorporates localized metal accents—such as copper fittings, wrought iron, silver inlays, and corner cladding—to heighten its decorative appeal and material contrast.
		Composite Material Comparison	The layered expression of multi-material combinations and their visual contrast effects, such as wood + stone, wood + lacquer, and wood + metal.
	Design Style	Minimalist Fusion Style	By abstracting traditional elements and integrating them with modern minimalist spatial design, the approach emphasizes negative space, balance, and structural simplification.
		Wabi-sabi Zen style	The design style influenced by Japanese and Eastern philosophy emphasizes natural, simple, quiet and sense of time.
		New Chinese Luxury Style	The design style blends traditional Chinese elements with modern premium materials, highlighting the fusion of cultural essence and spatial texture.
		Eclectic Industrial Style	The industrial components (metal, concrete, rivets, etc.) are fused with traditional furniture in a cross-stylistic manner, highlighting the raw and conflicting aesthetic.
	Composition and Atmosphere	Center Focal Composition	Place the main furniture in the center of the picture to form a strong visual focus and highlight the core content.
		Golden Section Composition	The image layout is based on the “golden ratio” to enhance the stability and aesthetic harmony of the image.
		Diagonal Dynamic Composition	The furniture layout or background structure presents a distinct diagonal orientation, emphasizing visual tension and dynamic momentum.
		Frame Composition	The visual “boundary” is formed by door frame, window frame, corridor column and other elements, which highlights the context atmosphere of the main furniture.
	Add Scene Elements	Core Cultural Ornaments	Small decorative objects such as incense burners, vases, and pen holders, which echo the theme of intangible cultural heritage furniture, serve as cultural highlights.
		Lively Natural Elements	Natural symbols such as plants, stone, water patterns, and light effects are used to enhance the spatial ambiance and embody the Eastern aesthetic of nature.
		Modern Functional Decoration	Functional or decorative elements that meet modern lifestyle needs, such as light strips, storage compartments, and power outlets.
		Traces of Humanistic Life	Furniture spaces reflect elements of daily use and life atmosphere, such as books, tea sets, cushions, and other scene fragments.

.

4.2.3. Expert Scoring

To ensure the scientific rigor of constructing the prompt system for ICH furniture and the consistency of the judgment process, this study invited five experts with relevant professional backgrounds. The expert selection criteria were set as follows: (1) having at least five years of professional or research experience in the fields of ICH design, furniture aesthetics, or AI image generation; (2) being familiar with the pairwise comparison scoring mechanism of the Analytic Hierarchy Process (AHP). The five experts finally invited were: one professor of ICH art research (20 years of experience, member of the China Artists Association); one professor of design (15 years of experience); one senior interior designer from an architectural design and research institute (16 years of experience, specializing in cultural space and Chinese aesthetic design); one AI image generation algorithm engineer (6 years of experience, specializing in the application of diffusion models to cultural content generation); and one product operations staff member from a home furnishing e-commerce platform (6 years of experience, responsible for visual material planning and product selection for furniture categories). These five types of experts covered the entire chain of “design—production—technology—application”, ensuring multi-perspective construction of the judgment matrix and the comprehensive validity of group decision-making. AHP scoring was then conducted on the prompt words at the criterion level and the alternative level. The expert scoring employed the standard pairwise comparison method to construct judgment matrices, supporting subsequent weight calculation and consistency checks. The questionnaire content centered on the importance of prompts in generating ICH furniture images as the core evaluation criterion, divided into six major indicator dimensions: Type and Form, Craftsmanship and Details, Color and Material, Design Style, Composition and Atmosphere, and Additional Contextual Elements. Each category included several specific prompts, and experts performed pairwise importance comparisons among prompts within the same category using a 1–9 scale, completing the judgment matrices [42]. To more intuitively display the importance of each element, the questionnaire results were visualized as a heatmap, where the warmth or coolness of the color represents the weight level. Cooler colors indicate lower weights, meaning the feature is less important in the design. Warmer colors indicate higher weights, meaning the feature holds greater relative importance in the design. The following displays the questionnaire results filled out by one of the experts, used to illustrate the practical application of AHP in determining prompt weights, as shown in Figure 5.

4.2.4. Calculation Results

The weights were calculated through steps including normalization, eigenvalue solving, and consistency checks. The final judgment matrices and the results of the consistency checks for the four criteria are presented in

Table 7. Judgment Matrix and Consistency Check for Prompt Weights: Generating High-Quality Images Integrating ICH Furniture and Modern Spaces.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Type and Shape	0.4796	0.3706	6.1573	0.0315	1.26	0.025 < 0.1 The consistency checkpasses
Process and Details	1.4357	0.2393
Color and Material	0.8924	0.1487
Design Style	0.6613	0.1102
Composition and Atmosphere	0.4796	0.0799
Add Scene Elements	0.3073	0.0512

,

Table 8. Judgment Matrix and Consistency Check: Type and Form.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Techniques and Processes	1.8633	0.4658	4.031	0.0103	0.89	0.0116 < 0.1 The consistency checkpasses
Regional School	0.6443	0.1611
Material Composition	0.3839	0.096
Functional Form	1.1086	0.2771

,

Table 9. Judgment Matrix and Consistency Check: Craftsmanship and Details.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Wood Construction	1.8204	0.4551	4.088	0.0293	0.89	0.033 < 0.1 The consistency checkpasses
Surface Decoration	0.7012	0.1753
Engraving and Inlay	1.1558	0.2889
Detail Modification	0.3226	0.0807

,

Table 10. Judgment Matrix and Consistency Check: Color and Material.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Natural Wood Texture	1.8473	0.4618	4.1574	0.0525	0.89	0.059 < 0.1 The consistency checkpasses
Vermicelli Color System	1.1038	0.276
Metal Decorative Material	0.4742	0.1186
Composite Material Comparison	0.5746	0.1436

,

Table 11. Judgment Matrix and Consistency Check: Design Style.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Minimalist Fusion Style	1.0575	0.2644	4.1189	0.0396	0.89	0.0445 < 0.1 The consistency checkpasses
Wabi-sabi Zen style	0.762	0.1905
New Chinese Luxury Style	1.8782	0.4695
Eclectic Industrial Style	0.3022	0.0756

,

Table 12. Judgment Matrix and Consistency Check: Composition and Atmosphere.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Center Focal Composition	1.1231	0.2808	4.1185	0.0395	0.89	0.0444 < 0.1 The consistency checkpasses
Golden Section Composition	1.8462	0.4615
Diagonal Dynamic Composition	0.6025	0.1506
Frame Composition	0.4282	0.1071

and

Table 13. Judgment Matrix and Consistency Check: Additional Contextual Elements.

Sublayer Name	Eigenvector	Weight	λmax	CI	RI	CR
Core Cultural Ornaments	1.8391	0.4598	4.0884	0.0295	0.89	0.0331 < 0.1 The consistency checkpasses
Lively Natural Elements	1.0894	0.2723
Modern Functional Decoration	0.3505	0.0876
Traces of Humanistic Life	0.721	0.1803

.

Based on the results of the judgment matrix consistency tests, all outcomes met the requirement of CR < 0.1, indicating that the consistency tests were passed. Finally, local judgment matrices for each alternative were constructed under the four criteria respectively, and their local weights and composite weights were further calculated, as shown in

Table 14. Composite Weights and Rankings of Prompts for Generating High-Quality Images Integrating ICH Furniture and Modern Spaces.

Target Layer (A)	Criterion Layer (B)	Weight	Scheme Layer (C)	Weight	Composite Weight	CR	Sort
Creating High-quality Images of Intangible Cultural Heritage Furniture and Modern Space	Type and Shape	0.3691	Material Composition	0.102	0.0376	0.0824	9
			Regional School	0.1638	0.0605		6
			Techniques and Processes	0.4953	0.1828		1
			Functional Form	0.2388	0.0881		3
	Process and Details	0.24	Surface Decoration	0.1749	0.0420	0.0296	8
			Engraving and Inlay	0.2411	0.0579		7
			Wood Construction	0.5048	0.1212		2
			Detail Modification	0.0793	0.0190		17
	Color and Material	0.1343	Vermicelli Color System	0.2402	0.0323	0.0668	10
			Composite Material Comparison	0.1421	0.0191		16
			Metal Decorative Material	0.098	0.0132		18
			Natural Wood Texture	0.5197	0.0698		4
	Design Style	0.1294	Wabi-sabi Zen style	0.1986	0.0257	0.0464	14
			Minimalist Fusion Style	0.2176	0.0282		12
			New Chinese Luxury Style	0.5132	0.0664		5
			Eclectic Industrial Style	0.0706	0.0091		22
	Composition and Atmosphere	0.0775	Diagonal Dynamic Composition	0.1611	0.0125	0.0251	19
			Golden Section Composition	0.4038	0.0313		11
			Frame Composition	0.0922	0.0071		23
			Center Focal Composition	0.3429	0.0266		13
	Add Scene Elements	0.0497	Core Cultural Ornaments	0.499	0.0248	0.0218	15
			Lively Natural Elements	0.1967	0.0098		21
			Traces of Humanistic Life	0.2209	0.0110		20
			Modern Functional Decoration	0.0834	0.0041		24

.

The AHP calculation results indicate that within the criterion layer, the indicators “Type and Form” (0.3691) and “Craftsmanship and Details” (0.2400) dominate, suggesting that in the image generation of ICH furniture, both expert judgment and user preferences place a high emphasis on the compositional logic and manufacturing techniques of traditional furniture [43]. This prioritization of “form” and “technique” reflects a strong reliance on core cultural identifiability in the visual communication of ICH. Within the alternative layer, prompts such as “Craft Technique” (0.1828), “Woodwork Structure” (0.1212), and “Functional Form” (0.0881) rank highest in weight, demonstrating their crucial guiding role in semantic control and the image generation process. In contrast, modern or auxiliary elements like “Modern Functional Decoration” and “Eclectic Industrial Style” have relatively lower weights, serving more as stylistic supplements rather than dominant prompts.

On this basis, a sensitivity analysis was conducted to perform perturbation tests on the criterion layer weights [44]. Specifically, the weights of “Type and Form” and “Craftsmanship and Details” were each fluctuated by ±10%, and the resulting changes in the ranking of the alternative-level prompts were observed (see

Table 15. Sensitivity analysis results (criterion layer weight ± 10% perturbation).

Scheme Layer Indicator	Type and Shape + 10%	Rank	Type and Shape − 10%	Rank	Craftsmanship and Details + 10%	Rank	Craftsmanship and Details − 10%	Rank
Craft Technique	0.2011	1	0.1645	1	0.1828	1	0.1828	1
Wood Construction	0.1141	2	0.1284	2	0.1333	2	0.1091	2
Functional Form	0.0969	3	0.0793	3	0.0881	3	0.0881	3
Natural Wood Texture	0.0657	4	0.0739	4	0.0698	4	0.0698	4
New Chinese Luxury Style	0.0625	5	0.0703	5	0.0664	5	0.0664	5
Regional School	0.0666	6	0.0544	6	0.0605	6	0.0605	6
Engraving and Inlay	0.0545	7	0.0613	7	0.0637	7	0.0521	7
Surface Decoration	0.0395	8	0.0445	8	0.0462	8	0.0378	8
Material Composition	0.0414	9	0.0338	9	0.0376	9	0.0376	9
Vermicelli Color System	0.0304	10	0.0342	10	0.0323	10	0.0323	10
Golden Section Composition	0.0295	11	0.0332	11	0.0313	11	0.0313	11

). The results showed that when the two dominant weights varied within the range of ±10%, the ranking of the top five alternative-level prompts (craftsmanship techniques, joinery construction, functional form, motif subject matter, color attributes) remained unchanged. Only when the weight change exceeded ±20% did minor adjustments in the ranking occur (e.g., “functional form” and “motif subject matter” swapping positions). This finding indicates that the weight ranking obtained in this study is highly robust, and the identification of dominant factors is not affected by small weight fluctuations. Consequently, the “menu-style prompt structure” constructed based on these weights can maintain a relatively stable semantic guidance capability across different application scenarios.

Based on the above weight ranking results, the research further organized the final integrated prompt set into a “menu-style prompt structure” with adjustable weights, integrating it into the ComfyUI image generation workflow. This menu is automatically generated in a “keyword: weight value” format, for example: “lacquer gloss: 1.4, background harmony: 1.2, modern furniture: 1.0”, where the weights are normalized and scaled based on the AHP output. Users can dynamically fine-tune the weights of each prompt according to project requirements (e.g., emphasizing craft details more or spatial atmosphere), achieving more targeted image generation control. This mechanism enhances the expressive precision of prompt control while also increasing the flexibility of human–computer collaboration within the ComfyUI system [45].

In summary, the introduction of AHP analysis transformed prompt strategy selection from an experience-based judgment into a structured, quantitative decision-making process. This effectively enhances the scientific rigor and reusability of the prompt system, providing a structurally sound, semantically accurate, and culturally adaptive control foundation for prompt input in the ICH furniture scene migration system. The ultimately selected “Integrated Prompt Group” will serve as the core control input in the ComfyUI scene migration system for the main variable input in subsequent image generation and comparative experiments, ensuring the system’s balanced and excellent performance in cultural transmission and visual generation.

4.3. ComfyUI Scene Migration System for Lacquer Furniture

The study selected a representative lacquer furniture case to validate the workflow. First, the WD14 Tagger reverse prompt node was used to identify key information within the image (as shown in Figure 6) and reverse-engineer prompts, while the Janus Image Understanding node was added for in-depth image comprehension. Subsequently, positive and negative prompts were incorporated via the prompt menu, and the output was directed globally [46]. In this workflow, the prompt menu serves as the input parameter for image generation control, allowing for dynamic adjustment within the workflow and providing feedback on the generated results. Manufacturers or designers can fine-tune the weight ratios of various prompts in the menu via the interface to quickly obtain multiple differentiated versions of image output. This facilitates project decision-making, style comparison, or aesthetic filtering.

Following this, the Segment Anything Ultra V2 node was used with “sofa” input in the prompt section to perform matting on the lacquer furniture product image. The Remove Alpha node was then employed to assign a white background to the image, as shown in Figure 7. Concurrently, Mask Expand and Mask Invert nodes were used to obtain the image mask, as shown in Figure 8. Through this process, a white-background image of the sofa was obtained, providing flexible space for subsequent image compositing and virtual scene rendering.

Subsequently, depth information was processed through the Depth ControlNet node, as shown in Figure 9, and the product line art was obtained via the Canny ControlNet node, as shown in Figure 10. Combined with the T2I-Adapter node, the processed image was input into the KSampler node for image processing. This enabled precise control over the furniture object’s spatial position, scale, and perspective direction within the scene, ensuring the complete presentation of the furniture object within the visual focal area. The Color Overlay V2 node was then applied to restore the furniture’s inherent colors and textures. Through this process, a notable enhancement in image quality and detail was achieved, with the final output image shown in Figure 11.

4.4. Experimental Design

The experimental procedure of this study consisted of three stages: prompt system construction, scene migration, and result evaluation. In the prompt system construction stage, three comparative prompt schemes were established: the integrated prompts (Group A), pure AI prompts (Group B), and pure user prompts (Group C). All three prompt groups cover four dimensions of prompting—type and form, craftsmanship and details, style and aesthetic, and spatial atmosphere—with a consistent number of keywords. To ensure the rigor and reproducibility of the experiment, the following key parameters were fixed throughout all generation processes: the random seed was uniformly set to 42 to maintain a consistent initial noise state; the sampler was set to Euler a, with the number of sampling steps fixed at 30; the CFG Scale (prompt guidance strength) was fixed at 7.5; the output image size was uniformly set to 1728 × 2304 pixels; and the weight coefficients of the ControlNet and IPAdapter modules were fixed at 0.8 and 0.7, respectively. All generations were based on the SDXL 1.0 base model without any fine-tuning or LoRA weights.

During the image generation stage, the unified input was lacquer furniture image material. The three sets of prompts were respectively imported as text inputs into the image generation workflow on the ComfyUI platform. The target scenes were set as two typical spatial types: “Song Dynasty Study Room” and “Modern Tea Room,” used to test the adaptive performance of ICH furniture images in both traditional and contemporary contexts. In the specific generation process, taking the Song Dynasty Study Room scene as an example, experimental Group A input utilized the weighted prompt structure optimized by AHP, resulting in the optimal prompts and weight ratios as follows: “lacquer furniture, butterfly backrest: 1.8, symmetrical structure: 1.5, dark wood tone: 1.2, Song style tea room: 1.0, warm lighting: 0.8”. The generated images demonstrated a high degree of coordination in material representation, structural proportion, and scene integration, exhibiting high composition stability and strong cultural style consistency. Control Groups B and C were respectively input with “traditional Chinese lacquer chair in ancient study room, carved frame, red backdrop, elegant shadows” and “a quiet wooden chair that fits in a minimalist tea room, not too dark, elegant feeling” based on their respective acquisition results. A total of 20 migrated image samples were generated for each of the three prompt groups (A/B/C), resulting in 60 images for this scene. Cumulatively, 120 images were output. All image outputs were uniformly sized at 1728 × 2304 pixels. The system generation workflow uniformly employed CLIP Text Encode for encoding, ControlNet for structural control modules, IPAdapter for style transfer modules, and was based on the SDXL model for high-quality image generation.

The final image generation results for the three groups are shown in Figure 12, Figure 13 and Figure 14. In terms of visual performance, Group A (integrated prompts) excelled most notably in image structural clarity, background harmony, and cultural atmosphere creation. As shown in Figure 12, the yellow-circled areas in Group A images exhibit natural and realistic light-shadow transitions—this benefits from the IPAdapter style transfer module’s precise extraction of lighting characteristics from reference images, enabling a coherent transition between the furniture’s gloss and the spatial illumination, thereby achieving visual adaptation on the “contemporary” dimension. The blue-circled lacquer surface texture is clearly distinguishable—this stems from the high weight assignments of the AHP weight system to “heritage” dimension indicators such as “craftsmanship techniques” (0.1828) and “joinery construction” (0.1212), allowing the prompts to accurately encode the material details of lacquer furniture. The red-circled elements of Song-style scenes are integrated naturally into the environment, with an overall harmonious visual perception—this directly reflects the optimal balance achieved by AHP-weighted prompts across both the “heritage” and “contemporary” dimensions, successfully realizing the organic integration of lacquer furniture with the spatial scene. Furthermore, the ControlNet structural control module, through edge detection and depth analysis, ensures the stability of the furniture’s main contour and spatial position, avoiding the common scale distortion issues observed in Group C.

In contrast, Group B (AI prompts) images maintain good stability in composition and clarity, making them suitable for structural display tasks. However, their prompts originate from general model training corpora and are essentially a form of “generic semantic encoding”, where weight allocation is determined by statistical patterns from the model’s pre-training phase rather than by a value ranking specific to ICH culture. Consequently, although the generated results show no obvious visual errors, they tend toward templatization: for example, the model by default associates “Chinese furniture” with dark red tones, heavy carvings, and traditional backgrounds, while failing to capture more culturally distinctive details such as “lacquer layering” or “mortise-and-tenon precision”. Thus, Group B images lack sufficient expression of cultural symbols and exhibit relatively monotonous spatial atmosphere, making it difficult to fully convey the traditional semantics carried by the furniture—a flattening of the “heritage” dimension that fails to meet consumers’ aesthetic expectations for a balance between tradition and modernity.

Group C (user prompts), although more closely aligned with users’ emotional cognition and reflecting a certain life-oriented context, often suffer from compositional disorder and element scale imbalance during image generation due to the inherent “semantic vagueness” and “structural deficiency” of natural language. As shown in Figure 13, the green-circled area shows an unreasonable chair-to-table scale and a large deviation in screen height—this occurs because user prompts (e.g., “a quiet wooden chair”, “an elegant feeling”) rely on implicit context rather than precise visual parameters, lacking explicit constraints on key variables such as furniture form, spatial scale, and light-shadow relationships, thereby increasing randomness in the diffusion model’s latent space sampling and resulting in structural distortion. The orange-circled elements remain at the level of superficial symbolic overlay of Chinese style—users tend to use generalized tags such as “Chinese style” or “traditional” rather than craft-detail terms like “lacquer” or “mortise-and-tenon”, causing the generated outputs to remain at the level of symbolic accumulation without reaching the deep semantic structure of ICH culture. Therefore, Group C images exhibit considerable quality fluctuation and lack stability. Although they show slight closeness to “contemporary” perception, they sacrifice cultural precision on the “heritage” dimension.

In summary, the differences among the three groups essentially reflect three distinct mechanisms of “semantic encoding—structural constraint—contextual adaptation”. Group A achieves explicit encoding of cultural value ranking through AHP weights, together with ControlNet and IPAdapter for structure preservation and style transfer, attaining an optimal balance on the “heritage, yet contemporary” dual dimensions. Group B relies on the model’s general statistical patterns, which are stable but lack cultural specificity. Group C depends on users’ subjective and vague expressions, which are emotionally close but lack structural constraints. This comparison validates the theoretical rationality and practical effectiveness of the integrated prompt system and the ComfyUI workflow constructed in this study.

5. Discussion

5.1. Results Evaluation

To comprehensively evaluate the performance of different prompt strategies in image generation, this study adopted a dual assessment mechanism combining expert subjective evaluation and user perception scoring. The focus was on rating the experimental images across three dimensions: semantic accuracy, style integration, and cultural expressiveness.

Expert evaluation. Seven experts in the fields of intangible cultural heritage (ICH) design, furniture aesthetics, and AI image generation were invited to rate the generated images on a 5-point Likert scale across three dimensions: “semantic accuracy”, “style integration”, and “cultural expressiveness”. The mean rating scores are presented in

Table 16. Mean Values of Likert Scale Scores for the Experimental Group (Group A) and Control Groups (Groups B and C).

Software	Prompt Type	Mean Semantic Accuracy	Average Degree of Style Fusion	Average Value of Cultural Expressiveness	Composite Average
ComfyUI	Integrated Prompts (Group A)	3.6	4.1	3.8	3.8
	AI Prompts (Groups B)	3.1	3.2	2.9	3.1
	User Prompts (Group C)	3.3	3.1	3.4	3.3

. Given the small sample size of the expert ratings and the violation of the normality assumption, the Friedman test (non-parametric repeated-measures ANOVA) was employed to test for differences among the three prompt groups. The results showed significant differences among the three prompt groups in semantic accuracy (χ² = 8.86, p = 0.012), style integration (χ² = 10.57, p = 0.005), and cultural expressiveness (χ² = 7.14, p = 0.028). Subsequently, the Wilcoxon signed-rank test was used for pairwise comparisons, with Bonferroni correction (adjusted significance level α = 0.0167). The results indicated that Group A (integrated prompts) significantly outperformed Group B (AI prompts) and Group C (user prompts) in all three dimensions (p < 0.01), while no significant difference was found between Group B and Group C (p > 0.05), as shown in

Table 17. Statistical test results of image ratings under different prompt strategies.

Evaluation Type	Dimension	Test Method	Test Statistic	p-Value	Effect Size	Post Hoc Comparison (Bonferroni Corrected)
Expert evaluation (n = 7)	Semantic accuracy	Friedman test	χ2 = 8.86	0.012	—	A > B, C (p < 0.01); B vs. C n.s.
	Style fusion	Friedman test	χ2 = 10.57	0.005	—	A > B, C (p < 0.01); B vs. C n.s.
	Cultural expressiveness	Friedman test	χ2 = 7.14	0.028	—	A > B, C (p < 0.01); B vs. C n.s.
User evaluation (n = 30)	Overall rating	One-way repeated measures ANOVA	F (2, 58) = 15.32	<0.001	partial η2 = 0.35	A > B, C (p < 0.01); B vs. C n.s.

.

User evaluation. A user study was conducted with 30 target users aged 25–40 years. The user evaluation was positioned as an exploratory analysis aimed at preliminarily investigating user perception differences in the generated images under different prompt strategies, providing directional references for future large-scale empirical studies. The sample composition was as follows: 6 representatives from furniture manufacturers (all in product operation or visual planning roles with more than three years of experience), 12 design practitioners (including interior designers, product designers, and home furnishing buyers), and 12 general consumers (with experience in home decoration or furniture purchasing). These three groups covered the perspectives of “supply side—professional end—consumer side”, enabling exploratory insights from diverse viewpoints. The invited users rated the images in terms of aesthetic acceptability, cultural recognizability, and spatial fit. A one-way repeated-measures ANOVA was conducted on the user rating data. The results showed a significant difference among the three prompt groups in user ratings (F (2, 58) = 15.32, p < 0.001, partial η² = 0.35). Post hoc comparisons (LSD correction) further confirmed that images from Group A were significantly higher than those from Group B and Group C in aesthetic acceptability, cultural recognizability, and spatial fit (p < 0.01). Group A performed particularly well in the dimensions of “cultural recognizability” and “spatial integration”. Most users provided feedback stating that this group of images “possessed traditional beauty while also being suitable for modern spatial settings.” In contrast, Group B images were perceived as “tending towards a templatized style,” and Group C images were seen as “subjective feeling is strong but compositionally unstable.”

Integrating the expert and user scoring results, it is evident that the Group A prompt strategy achieved the most ideal comprehensive effect in guiding image generation quality and cultural expression. This validates the necessity and effectiveness of constructing an integrated prompt system and introducing a weight control mechanism. It also provides a practically valuable optimization pathway for the input structure in the AI scene migration of ICH furniture. These results offer quantitative evidence and theoretical support for subsequent prompt iteration, generation mechanism optimization, and human–computer collaborative system design.

5.2. Theoretical Explanation and Mechanistic Analysis of the Results

The experimental results show that Group A (AHP-weighted integrated prompts) significantly outperforms Group B (pure AI prompts) and Group C (pure user prompts) in semantic accuracy, style integration, and cultural expressiveness. The underlying mechanism can be explained from three interconnected aspects: semantic encoding, structural constraint, and contextual adaptation.

First, AHP transforms implicit cultural values—such as “craftsmanship techniques” and “joinery construction”—into a quantifiable weight system through hierarchical decomposition and pairwise expert comparisons. This imposes a strongly constrained direction in the latent space for the prompts, thereby avoiding the flattening of cultural symbols that occurs in Group B due to reliance on statistical co-occurrence in the model (e.g., the stereotypical output of dark red, heavily carved images), and also remedies the semantic defocus caused by vague user expressions in Group C (e.g., “a quiet wooden chair”). In essence, this process converts fuzzy design decisions into a logically consistent weighted encoding, achieving a transition from “statistical correlation” and “subjective perception” to “computable weights”.

Second, the ControlNet structural control module adds geometric anchors to the generation process through edge detection and depth estimation. Contour and depth information extracted from the original furniture image locks the furniture’s boundaries and spatial hierarchy, ensuring that the diffusion model sampling always follows the preset geometric constraints. This effectively suppresses the common issues of scale distortion and floating subjects observed in Group C. From the perspective of architectural language theory, this is equivalent to making the “deep structure” (geometric essence) of the furniture explicit, thereby maintaining morphological fidelity during style transfer and providing a material basis for cultural recognizability.

Finally, the IPAdapter style transfer module performs the function of “context reconstruction”. By extracting low-level visual features such as lighting, color tone, and material ambiance from reference images and transferring them to the background space of the main generated image, ICH furniture can be naturally embedded into different interior styles, including modern minimalist, wabi-sabi, and light luxury. This mechanism compensates for the insufficient constraint of pure text control on low-level visual features, avoiding the homogenization of the “Chinese furniture” statistical style in Group B and resolving the “collage-like” feel in Group C caused by the lack of specific stylistic references. From a semiotic perspective, IPAdapter, through paradigmatic anchoring, reassembles a comprehensible “contemporary signified” for the “signifier” of the furniture, enabling a semantic alignment between traditional symbols and modern spaces.

Taken together, AHP-weighted prompts ensure that the core cultural features of the “heritage” dimension receive high-weight constraints during generation; ControlNet guarantees morphological fidelity and spatial logic; and IPAdapter achieves natural integration with modern contexts. The synergy of these three components precisely addresses the requirement of the “heritage, yet contemporary” aesthetic principle—i.e., the optimal balance between traditional heritage elements and contemporary visual elements—providing a computable and reproducible technical pathway for transforming ICH furniture from an “isolated cultural artefact” into a “spatial element that can be integrated into the discourse of contemporary interior design”. This mechanistic analysis not only validates the core hypothesis of this study but also offers a theoretical basis for controllable design in the contextual generation of cultural heritage using AIGC.

6. Conclusions

6.1. Principal Findings

This study addresses the three research questions (RQs) proposed in the introduction through theoretical construction, system development, and experimental validation, reaching the following conclusions:

RQ1 (Prompt construction problem): This study constructs an integrated prompt framework that fuses AI prompts and user natural language, and introduces the Analytic Hierarchy Process (AHP) for weight optimization. By decomposing the ICH furniture image generation task into a goal layer, a criterion layer (six dimensions), and an alternative layer (24 indicators), and employing pairwise comparisons and consistency checks (CR < 0.1) by five experts, the study successfully transforms implicit cultural values—such as “craftsmanship techniques”, “joinery construction”, and “functional form”—into a quantifiable weight system, achieving precise control over the cultural semantics of ICH furniture.

RQ2 (Workflow implementation problem): This study designs and implements a visual generation workflow based on the ComfyUI platform that integrates “prompt control—structure preservation—style transfer”, incorporating CLIP text encoding, ControlNet structural control (Canny edge detection + Depth depth estimation), and the IPAdapter style transfer module. Experiments demonstrate that this workflow can efficiently transfer ICH furniture images into diverse modern spatial contexts—such as new Chinese style living rooms, wabi-sabi tea rooms, and light luxury studies—while maintaining morphological fidelity (clear structure and proportional harmony), with a single-group image generation cycle reduced to the minute level.

RQ3 (Effectiveness validation problem): Through a dual evaluation mechanism involving expert scoring (7 experts) and user scoring (30 users), along with statistical tests (Friedman test, repeated-measures ANOVA), the results show that images generated by AHP-weighted integrated prompts are significantly superior to those generated by pure AI prompts and pure user prompts across multiple dimensions: semantic accuracy (3.6 vs. 3.1/3.3), style integration (4.1 vs. 3.2/3.1), cultural expressiveness (3.8 vs. 2.9/3.4), as well as user-side measures of aesthetic acceptability, cultural recognisability, and spatial fit (p < 0.01). This validates the core hypotheses H1 and H2.

6.2. Limitations and Future Work

This study has the following limitations that warrant further investigation:

First, limitations in expert panel size and user sample. The current AHP weight derivation relied on evaluations from only five experts. Although their backgrounds covered the entire chain of ICH heritage, design, algorithms, and operations, the small number of experts introduces potential subjectivity that is difficult to eliminate entirely. Similarly, the user evaluation sample consisted of only 30 participants, which falls within the scope of exploratory research and limits statistical generalizability. Future work could incorporate user perception data (e.g., eye-tracking, click-through feedback) for dynamic weight optimization and expand the user sample size to enhance the generalizability of the findings.

Second, limitations in the scale and representativeness of the training dataset. The ComfyUI workflow in this study is based on the SDXL 1.0 base model, whose training data are predominantly Western visual content, resulting in insufficient learning samples for the specific forms, materials, and craftsmanship details of Chinese ICH furniture. Although structural control and style transfer were achieved through ControlNet and IPAdapter, the generation of highly culture-specific elements—such as “lacquer decoration”, “close-up of mortise-and-tenon joints”, and “traditional carved patterns”—may still suffer from morphological distortion or semantic confusion. Future work should construct a dedicated image dataset for ICH furniture (with a recommended size of no fewer than 5000 high-quality annotated images) and train or fine-tune specialized LoRA models on this dataset to improve the accuracy of cultural semantic generation.

Third, potential cultural bias in the base model. The base diffusion model (SDXL) may have embedded certain stereotypes during pre-training, for example, associating “Chinese furniture” by default with “dark tones, heavy carvings, and outdated scenes”—while performing less effectively for “modern Chinese” or “new Chinese” styles. Such cultural bias can drive generated images excessively toward a “traditional museum” aesthetic, deviating from the “heritage, yet contemporary” balance pursued in this study. Although the AHP-weighted prompts partially correct this bias, the fundamental elimination of such biases requires enhancing cultural diversity and applying debiasing techniques at the training data level.

Fourth, challenges in cross-cultural application. The AHP-weighted prompt framework and ComfyUI workflow proposed in this study are primarily designed for Chinese ICH furniture and its contextual migration into modern domestic spaces. When applying this method to other cultural backgrounds (e.g., Japanese, Southeast Asian, or European traditional furniture), the indicators at both the criterion level and the alternative level must be recalibrated, as different cultures place different emphases on “form”, “technique”, “material”, and “context”. For instance, Japanese traditional furniture may prioritize “natural wood grain texture” and “minimalist Zen spirit”, whereas European classical furniture may focus more on “ornamental complexity” and “palatial ambience”. Therefore, cross-cultural application of this framework would require renewed expert surveys and AHP weight calculations rather than directly adopting the existing weight system.

Fifth, the gap between generated images and physical artifacts. The images generated by the system still exhibit a certain gap compared to physical photographs, particularly in terms of complex lighting, shadows, and material details, such as metal inlays and the high-gloss finish of lacquerware. Furthermore, the current workflow supports only static image generation and does not extend to video or interactive 3D scenes, limiting its use in immersive communication.

Future work will focus on: ① constructing a dedicated image dataset and LoRA model for ICH furniture to improve the accuracy of cultural semantic generation and reduce cultural bias; ② exploring dynamic scene generation based on video diffusion models; ③ integrating augmented reality (AR) technology to enable real-time furniture embedding visualization in user-defined spaces; and ④ conducting cross-cultural comparative studies to validate and adapt the proposed framework for different types of traditional furniture.