Model Bending Experiments: Appendix for "Unboxing Diffusion Models for the Arts"
Fuente:
Zenodo
Enregistré dans:
| Auteurs principaux: | , |
|---|---|
| Format: | Recurso digital |
| Publié: |
Zenodo
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866901778249809920 |
|---|---|
| author | Abuzuraiq, Ahmed Pasquier, Philippe |
| author_facet | Abuzuraiq, Ahmed Pasquier, Philippe |
| contents | <p>This includes the appendix and supplementay material for the paper: <br>Ahmed Abuzuraiq, A., & Pasquier, P. (2026). Unboxing diffusion models for the arts: Interactive model bending and practice-based explainability. In Explainable AI for the Arts (XAIxArts). Springer.</p> <p>It includes sample results of systematic experiments with <em>model bending </em>(manipulation of generative model's internal activations for creative purposes) applied to a Stable Diffusion model using this ComfyUI plugin: <br>https://github.com/abuzreq/ComfyUI-Model-Bending <br><br>This repository includes two formats: a PDF version appendix containing the curated results mentioned in the paper, and a web viewer with a larger collection of results that can be interactivly sorted, grouped, and filtered. </p> <p>All the results are generated using the prompt “Analog style portrait of a person” with model realisticVisionV51_v51VAE.safetensors (seed 42; 20 steps; CFG 7; DPM++ 2M sampler; Karras scheduler; denoise 1; 512×512), while scaling each layer’s activations by 0, 0.5, 1.5, and 2. The experiment setup includes: (1) Same Prompt/Seed, Different Layers, (2) Same Prompt, Different Seeds , (3) Same Seed, Different Prompts, and (4) Same Seed/Prompt/Layer, Different Timesteps. </p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_18986323 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Model Bending Experiments: Appendix for "Unboxing Diffusion Models for the Arts" Abuzuraiq, Ahmed Pasquier, Philippe generative ai AI art Stable Diffusion <p>This includes the appendix and supplementay material for the paper: <br>Ahmed Abuzuraiq, A., & Pasquier, P. (2026). Unboxing diffusion models for the arts: Interactive model bending and practice-based explainability. In Explainable AI for the Arts (XAIxArts). Springer.</p> <p>It includes sample results of systematic experiments with <em>model bending </em>(manipulation of generative model's internal activations for creative purposes) applied to a Stable Diffusion model using this ComfyUI plugin: <br>https://github.com/abuzreq/ComfyUI-Model-Bending <br><br>This repository includes two formats: a PDF version appendix containing the curated results mentioned in the paper, and a web viewer with a larger collection of results that can be interactivly sorted, grouped, and filtered. </p> <p>All the results are generated using the prompt “Analog style portrait of a person” with model realisticVisionV51_v51VAE.safetensors (seed 42; 20 steps; CFG 7; DPM++ 2M sampler; Karras scheduler; denoise 1; 512×512), while scaling each layer’s activations by 0, 0.5, 1.5, and 2. The experiment setup includes: (1) Same Prompt/Seed, Different Layers, (2) Same Prompt, Different Seeds , (3) Same Seed, Different Prompts, and (4) Same Seed/Prompt/Layer, Different Timesteps. </p> |
| title | Model Bending Experiments: Appendix for "Unboxing Diffusion Models for the Arts" |
| topic | generative ai AI art Stable Diffusion |
| url | https://doi.org/10.5281/zenodo.18986323 |