Research map / Image generation / Text-to-image diffusion
UniVG: A Generalist Diffusion Model for Unified Image Generation and Editing
Tsu-Jui Fu, Yusu Qian, Chen Chen, Wenze Hu, Zhe Gan, Yinfei Yang
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 |
|---|---|---|---|
| add instruction-based editing to T2I training (ablation) | UniVG · GenEval | GenEval score: 0.71 -> 0.70 | No clear effect |
| add instruction-based editing to T2I training (ablation) | UniVG · HPSv2 | HPSv2: 28.4 -> 28.3 | No clear effect |
| add ID-preserving generation without auxiliary tasks (ablation) | UniVG · MagicBrush | CLIP-T: 28.3 -> 26.9 | Hurt |
| add ID-preserving generation without auxiliary tasks (ablation) | UniVG · EmuEdit | CLIP-T: 25.2 -> 24.9 | Hurt |
| add auxiliary tasks and layout generation to T2I+editing (ablation) | UniVG · MagicBrush | CLIP-T: 28.3 -> 29.8 | Improved |
| add auxiliary tasks and layout generation to T2I+editing (ablation) | UniVG · EmuEdit | CLIP-T: 25.2 -> 26.2 | Improved |
| add auxiliary tasks when ID generation is included (ablation) | UniVG · MagicBrush | CLIP-T: 26.9 -> 29.5 | Improved |
| add auxiliary tasks when ID generation is included (ablation) | UniVG · Unsplash-50 | CLIP-T: 28.9 -> 28.1 | Hurt |
| add auxiliary tasks when ID generation is included (ablation) | UniVG · Unsplash-50 | ID similarity: 0.327 -> 0.329 | No clear effect |
| use channel-wise concatenation of image latents instead of sequence concatenation | UniVG vs OmniGen · Editing inference time | time (sec): 36.8 -> 10.4 | Improved |
| scale model from B to XL (ablation) | UniVG · T2I and editing | benchmarks | Improved |
| add unified multi-task training vs task-specific SD3 | UniVG vs SD3 · GenEval | GenEval score: 0.71 -> 0.70 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- OmniGen — OmniGen: Unified Image Generation (CVPR 2025) · better on 7 of 9 shared results
- EmuEdit — Emu Edit: Precise Image Editing via Recognition and Generation Tasks (CVPR 2024) · better on 2 of 4 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
- R&B: Region and Boundary Aware Zero-shot Grounded Text-to-image Generation (ICLR 2024 · similarity 0.91)
- EditAR: Unified Conditional Generation with Autoregressive Models (CVPR 2025 · similarity 0.90)
- DreamDistribution: Learning Prompt Distribution for Diverse In-distribution Generation (ICLR 2025 · similarity 0.90)
- Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs (ICML 2024 · similarity 0.90)
- ACE: All-round Creator and Editor Following Instructions via Diffusion Transformer (ICLR 2025 · similarity 0.89)
- Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models (CVPR 2024 · similarity 0.89)
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.
