ostris
Image quality degradation in new versions
I trained my flux model on replicate but as new version ID of the model comes, it degrades my image quality. I've a product developed on flux replicate LoRA trainer. This is seriously concerning as I have to start over and spend more effort to get same quality
Suppose I trained my model on version 2. After some time version 10 comes but version 2 is unavailable and model is trained on minimum available version like 5. How will I get same consistency in image quality.
I trained a flux LoRA model in late September. Now I train a new model on replicate platform on same person images with same parameter and same prompt, I get a cartoony image but previous I was getting realistic high quality image.
I again trained model in replicate on same dataset, same parameters, same prompt but the september version is still giving different quality.
I'm developing a product and with these uncertainties how will I ensure in future my users will get same results I would be great if you guys can share your thoughts @jaretburkett @apolinario @fofr @FoundSol @liaoliaojun
as i noticed there was automatic image captioning in earlier versions, last version doesn't have that by default
