Research map / Image generation / Editing and style transfer
UniEdit-I: Training-free Image Editing for Unified VLM via Iterative Understanding, Editing and Verifying
Chengyu Bai, Jintao Chen, Xiang Bai, Yilong Chen, Qi She, Ming Lu, Shanghang Zhang
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 |
|---|---|---|---|
| edit in CLIP semantic latent space instead of VAE space (ablation) | FlowEdit trajectory · Intermediate artifact score (100 samples) | Artifact score (1-10): 5.35 -> 8.10 | Improved |
| edit in CLIP space instead of VAE space for feedback stability (ablation) | FlowEdit trajectory · Intermediate states (100 samples) | Feedback stability (std of CLIP-Sim, lower is better): 0.063 -> 0.025 (Table 2); text states reverse values | Mixed |
| replace fixed gain with full dynamic gain (ablation) | UniEdit-I · GEdit-Bench-EN (full set) | Overall: 5.66 -> 7.06 | Improved |
| replace linear decay gain with full dynamic gain (ablation) | UniEdit-I · GEdit-Bench-EN (full set) | Overall: 5.97 -> 7.06 | Improved |
| add task-completion score pt to dynamic gain (ablation) | UniEdit-I · GEdit-Bench-EN (full set) | Overall: 6.77 -> 7.06 | Improved |
| dynamic gain on perceptual quality (ablation) | UniEdit-I · GEdit-Bench-EN (full set) | Perceptual Quality: 7.39 -> 7.40 | No clear effect |
| apply training-free UniEdit-I loop on BLIP3-o-8B versus trained editors | BLIP3-o-8B · GEdit-Bench-EN (full set) | Overall: 6.70 (Step1X-Edit) -> 7.06 | Improved |
| training-free closed-loop editing versus GPT-4o | UniEdit-I · GEdit-Bench-EN (full set) | Overall: 7.53 (GPT-4o) vs 7.06 | Hurt |
| rely on BLIP3-o semantic space for text change tasks | UniEdit-I · GEdit-Bench-EN text change | Overall | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- OmniGen — OmniGen: Unified Image Generation (CVPR 2025) · better on 1 of 1 shared result
- Instruct-Pix2Pix — InstructPix2Pix: Learning To Follow Image Editing Instructions (CVPR 2023) · better on 1 of 1 shared result
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
- FireEdit: Fine-grained Instruction-based Image Editing via Region-aware Vision Language Model (CVPR 2025 · similarity 0.91)
- ReasonEdit: Towards Reasoning-Enhanced Image Editing Models (CVPR 2026 · similarity 0.90)
- UnicEdit-10M: A Dataset and Benchmark Breaking the Scale-Quality Barrier via Unified Verification for Reasoning-Enriched Edits (CVPR 2026 · similarity 0.90)
- Guiding Instruction-based Image Editing via Multimodal Large Language Models (ICLR 2024 · similarity 0.89)
- EdiVal-Agent: An Object-Centric Framework for Automated, Fine-Grained Evaluation of Multi-Turn Editing (ICLR 2026 · similarity 0.88)
- VLKEB: A Large Vision-Language Model Knowledge Editing Benchmark (NeurIPS 2024 · similarity 0.88)
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.
