{"slug": "imir-image-instruction-tuning-for-all-in-one-image-restoration", "title": "ImIR: Image-Instruction Tuning for All-in-One Image Restoration", "summary": "Researchers introduced ImIR, an image-instruction tuning method that adapts a single Qwen-Image-Edit model to six image restoration tasks using one low-rank adapter trained in about three hours on one GPU. ImIR replaces the text prompt used in prior restoration recipes with a continuous instruction vector derived from the degraded image, combining structure from the model's VAE with a semantic instruction from a lightweight token mapper. The image instruction outperformed text conditioning under a matched comparison and enabled task-agnostic restoration without a degradation label, which the text variant could not do.", "body_md": "arXiv:2609.25267v1 Announce Type: new \nAbstract: Degradations vary widely across images, so a practical restoration system has to handle many degradation types with one model. A recent and effective recipe adapts a large pretrained image-editing model to restoration using a small low-rank adapter with a text prompt. We replace that prompt with an instruction derived from the degraded image itself. The image reaches the editor through two paths: its structure comes from the model's VAE, and its semantic instruction comes from a lightweight token mapper that shifts the degraded image's vision-language embedding toward the embedding a clean image would produce. Because the instruction is a continuous vector, scaling it yields a family of valid restorations for tasks whose target is not unique, such as low-light enhancement. We adapt one Qwen-Image-Edit model to six tasks with a single adapter trained in about three hours on one GPU. The image instruction outperforms text conditioning under a matched comparison, and it supports task agnostic restoration without a degradation label, which the text variant does not.", "url": "https://wpnews.pro/news/imir-image-instruction-tuning-for-all-in-one-image-restoration", "canonical_source": "https://arxiv.org/abs/2609.25267", "published_at": "2026-09-23 04:00:00+00:00", "updated_at": "2026-09-23 04:27:03.312196+00:00", "lang": "en", "topics": ["computer-vision", "generative-ai", "ai-research", "machine-learning"], "entities": ["ImIR", "Qwen-Image-Edit"], "alternates": {"html": "https://wpnews.pro/news/imir-image-instruction-tuning-for-all-in-one-image-restoration", "markdown": "https://wpnews.pro/news/imir-image-instruction-tuning-for-all-in-one-image-restoration.md", "text": "https://wpnews.pro/news/imir-image-instruction-tuning-for-all-in-one-image-restoration.txt", "jsonld": "https://wpnews.pro/news/imir-image-instruction-tuning-for-all-in-one-image-restoration.jsonld"}}