ImIR: Image-Instruction Tuning for All-in-One Image Restoration
ImIR replaces text prompts with image-derived instructions so one Qwen adapter restores many degradations.
ImIR adapts a pretrained Qwen-Image-Edit model to all-in-one image restoration by replacing the text prompt with a continuous instruction derived from the degraded image. Structure comes from the model's VAE, while a lightweight token mapper shifts the vision-language embedding toward a clean-image embedding. One low-rank adapter covers six tasks after about three hours of training on a single GPU. Image instructions beat matched text conditioning and support label-free restoration, including scalable outputs for non-unique targets such as low-light enhancement.
- One Qwen-Image-Edit adapter covers six restoration tasks.
- Training takes about three hours on a single GPU.
- Continuous image instructions outperform matched text prompts.
- No degradation label is required for task-agnostic restoration.
Full article166 words · extracted from huggingface.co · click to collapse
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.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.25267