[Feature Request]: set different models to use for txt2img and img2img and also make an option that prevents prompts from being sent to img2img

[molkemon]

trafficstars

Is there an existing issue for this?

• [X] I have searched the existing issues and checked the recent builds/commits

What would your feature do ?

The SD 1.5 inpainting model is so vastly superior to any other model for inpainting, it's not even close, the support for inpainting conditioning mask strength really makes it the only feasable choice.

Therefore it should always be used for inpainting. As it is now, if I want to generate an image in a specific artstyle, first genning the principal picture but without artstyle and then sending it to img2img and then have the style inpainted usually produces much better results than having the picture genned with the artstyle to begin with.

To facilitate this worklow, it would be great if model used for txt2img and img2img could be set seperatly, so you don't always have to switch.

Also, when I send a genned img to img2img, the used prompt gets sent aswell. Please make an option that disables this, as the img2img prompt should only be the artstyle in this workflow.

Proposed workflow

Add options to have seperate models for txt2img and img2img and also add an option that prevents the prompt being sent to img2img if you click the button from txt2img.

Additional information

No response

[fredconex]

I think this is not possible due to vram limitations, having both means we would require to have both loaded to vram, so you must change it manually.

[a-l-e-x-d-s-9]

@fredconex The models don't need to be loaded in the memory at the same time. The UI can remember which model was selected by the user for each mode: txt2img, img2img. And swap the model if necessary before the generation. Also I think that prefered working mode can be the same model for all modes, but with easy option to change to a different model for each mode.

[maniac-0s]

This would be great!

Of course the model would need to be reloaded but sometimes you need to do more in img2img. Like inpaint, inpaint sketch, have several attempts and so on. Here it would be great if we could have a different model loaded until we switch back to txt2img.

My workaround is now to use refiner switch at 0.01, but that of course means that the original model is always loaded and then kicked after 1st step and to load the refiner.