Use one prompt and one image to train qwen-image-edit the help it to learn the visual features and understanding ability of special scene, e.g. the face feature. I think this way can improve the model’s edit capacity by enhancing its understanding ability if we lack high quality image pairs which are usually needed in training image-edit model.
Is there any way to make it ?
1 Like
Fine-tuning with just one image makes it quite difficult to achieve good results.
If you absolutely must use training as a means, I think it’s better to augment the dataset beforehand by synthesizing data, etc.