ostris
Black Images and LoRAs do not work.
Hello, I am getting this error when starting the training on Flux.1 dev
image_processor.py:148: RuntimeWarning: invalid value encountered in cast
It is generating black images as samples. Training also shows nan loss
It is training faster than it should and the resulting LoRA files are correct size but they do not work, the images I generate produce random people instead of the person I've trained.
I am using windows 11 - RTX 5090
You didnt even tell what you train...
Updated post. I am training a LoRA on the flux.1 dev model.
Similar problem when training WAN2.2. I'll roll back a version and see if that fixes anything.
Edit: Rolling to previous commits doesn't help.
I get this error during sampling. Perhaps something the version of diffusers.
Loading Vae
/venv/main/lib/python3.12/site-packages/diffusers/image_processor.py:148: RuntimeWarning: invalid value encountered in cast
images = (images * 255).round().astype("uint8")
Similar
Still not working, even tried chatgpt which suggested disabling quantization for transformer and text encoder but doing so casues Cuda out of memory errors