Make it a mandatory part of the model upload to state on what resolution the model was trained on
It would be nice if when you upload a model you are required to state on which resolution you trained the model on.
Like when you upload a model, you have to state the name, then the "Early Access", then what base model you trained it on.
I think that after the base model input, you should have to input on what resolution you trained the model on.
E.g. my model was trained on a 1024 resolution on 1.5 SD.
I know that the tag system exists, but that is voluntary. I would like this to be a mandatory thing of any model upload, so that one can see at a quick glance on what resolution one has to run inference at.
More importantly, it would help people like me to quickly find similar 1024 res trained models, which currently is basically impossible.
Currently most models are trained using a 512 or 768 resolution, but I think higher resolution and more diverse resolution training will spread (especially when SDXL releases), so I think this feature just makes sense.
Uhm, A LOT Of lora are trained on MIX resolutions, none of mine are DIRECTLY that size, when I train lora, i never resized the images - You never have to evne with Lycoris. So no, i don't think this feature makes any sense, it buckets it when you have multiple sizes.