Why is generated image noisy?
Hi, I tried your example script and this is the result.
Hi, I tried your example script and this is the result.
same, i'm training lora on this and get exact same problem. Smth does not fully connect to vae
I apologize. I must have screwed up the scaled weight conversion in this instance. It was bugging out for me as well (though I wasn't sure if it was just me). But it seems I was negligent to have kept this repo up, especially without warning. For training, at least if you're using AI-toolkit, I would highly suggest this hqq/nf4 conversion of Kontext at https://huggingface.co/HighCWu/FLUX.1-Kontext-dev-bnb-hqq-4bit for faster/lower-VRAM training (which also enables training at higher resolutions if you do have more VRAM for training: highly recommended for Kontext). This 4bit will definitely work for training with HighCWu's fork at https://github.com/HighCWu/ai-toolkit or my fork at https://github.com/AlekseyCalvin/ai-toolkit-soonr .
For inference, I've just kept using https://huggingface.co/buildborderless/FLUX.1_Kontext-Lightning-8step