Research map / Image generation / Fast diffusion sampling
Generative Visual Prompt: Unifying Distributional Control of Pre-Trained Generative Models
Chen Henry Wu, Saman Motamed, Shaunak Srivastava, Fernando De la Torre
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply PromptGen moment constraint to StyleGAN2 | StyleGAN2 FFHQ · race de-biasing | KL divergence (race): 0.860 -> 0.099 | Improved |
| increase moment-constraint scale lambda from 1 to 2 (ablation) | PromptGen · FFHQ race and age | KL divergence (race): 0.286 -> 0.099 | Improved |
| increase moment-constraint scale lambda from 1 to 2 (ablation) | PromptGen · MetFaces gender | KL divergence (gender): 0.000 -> 0.005 | Hurt |
| use PromptGen instead of FairStyle for combined attributes | StyleGAN2 FFHQ · age+gender | KL divergence: 0.0257 -> 0.000558 | Improved |
| use PromptGen instead of FairStyle for single-attribute de-biasing | StyleGAN2 FFHQ · gender | KL divergence: 3.20e-7 -> 1.71e-5 | Hurt |
| amortize CLIP-energy sampling with INN instead of inference-time Langevin optimization (PPGM) | StyleGAN2 · inference latency | seconds per sample: 21.5 -> 0.2 | Improved |
| use PromptGen instead of PPGM n=50 for CLIP energy | StyleGAN2 · baby text | CLIP energy: 0.7134 -> 0.7038 | Improved |
| use PromptGen for pose-guided generation instead of GAN-Control | StyleGAN2 · same-ID embedding distance | distance: 0.68 -> 0.45 | Improved |
| decompose a complex text control into simpler energy terms (ablation) | PromptGen · bald black man with beard | qualitative control | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- FairStyle — FairStyle: Debiasing StyleGAN2 with Style Channel Manipulations (ECCV 2022) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- PromptCoT: Align Prompt Distribution via Adapted Chain-of-Thought (CVPR 2024 · similarity 0.88)
- Direct Discriminative Optimization: Your Likelihood-Based Visual Generative Model is Secretly a GAN Discriminator (ICML 2025 · similarity 0.86)
- Prompting Hard or Hardly Prompting: Prompt Inversion for Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.86)
- AP-Adapter: Improving Generalization of Automatic Prompts on Unseen Text-to-Image Diffusion Models (NeurIPS 2024 · similarity 0.86)
- Learning to Sample Effective and Diverse Prompts for Text-to-Image Generation (CVPR 2025 · similarity 0.86)
- A User-Friendly Framework for Generating Model-Preferred Prompts in Text-to-Image Synthesis (AAAI 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
