Refinement Is Inherently Editable: Training-Free Prompt-to-Prompt Image Editing with Generative Refinement Network
RefineEdit is a training-free image editing framework achieving best background preservation on PIE-Bench through generative refinement.
RefineEdit is a training-free prompt-to-prompt image editing framework that couples edit localization with content generation through global refinement of binary image codes. The method uses adaptive spatial freezing and finite bit locking to stabilize editing across refinement steps. Across nine editing categories of PIE-Bench, RefineEdit achieves the best background-preservation scores in PSNR, LPIPS, MSE and SSIM.
- RefineEdit is a training-free prompt-to-prompt image editing framework using generative refinement.
- Method couples edit localization with content generation through global refinement of binary codes.
- Achieves best background preservation scores across PIE-Bench metrics.
- Requires no additional training, external masks, or attention control.
- Products
- RefineEdit
- AI models
- RefineEdit
From the original article
1 min read at huggingface.co
Text-guided image editing must introduce the requested changes while preserving unrelated source content. Diffusion-based editors rely on spatial controls whose inaccuracies can leave edits incomplete or alter unrelated regions. Causal autoregressive editors face a further constraint: their fixed decoding order limits revision of earlier decisions.
By Yulong Chen, Ziqian Zhang, Haoyu Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.