Instructions to use Alissonerdx/flux.1-dev-SRPO-LoRas with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use Alissonerdx/flux.1-dev-SRPO-LoRas with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("Alissonerdx/flux.1-dev-SRPO-LoRas", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
- DiffusionBee
Question about clamping the LoRA weights
#1
by woctordho - opened
Hi, just a quick check: Did you clamp the LoRA weights when extracting?
It seems the clamping is a common practice, but it causes quality loss, see https://github.com/kijai/ComfyUI-FluxTrainer/issues/183
Yes, I extracted the Loras with the clamp activated but I thought it was only used in the quantile type, anyway I am extracting the Loras from the original model and I disabled this option, thanks for letting me know.
It seems the clamping is a common practice, but it causes quality loss, see https://github.com/kijai/ComfyUI-FluxTrainer/issues/183
This explains why I had so many difficulties extracting good LoRAs recently. Thank you!