ostris
Control images for QwenImageEditPlusModel are not mandatory, but AI Toolkit fires errors when not providing any of them
In ComfyUI it is possible to generate images with Qwen Image Edit 2509 even when no control images are provide. This works very well, as if Qwen Image Edit 2509 was a generative model. Still AI Toolkit won't run the training lamenting the famous " Missing control images for QwenImageEditPlusModel" error when no control images are provided. Please, can this be corrected? Sometimes you need a generic LORA which does not need any control image to operate, for example for character LORAs. These kind of LORAs should be possible to train even for Qwen Image Edit 2509.
AI toolkit can train Flux.2 both with control images and without control images. Clearly this is something that can be fixed. As a workaround I just select Qwen Image as the model architecture and then manually enter Qwen/Qwen-Image-Edit-2511 as the model name. It trains without an issue and the results are good.
