Instructions to use XLabs-AI/flux-controlnet-canny with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use XLabs-AI/flux-controlnet-canny with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("XLabs-AI/flux-controlnet-canny", 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
- Draw Things
- DiffusionBee
why flux1-dev produce noisy image with canny controlnet
sadly the results I get are vastly inconsistent but often they look like this. 90% of the time the end result is unusable, then randomly i will get a good gen. Been researching for about 7 hours today, haven't found a consistent solve.
I spent 70 hours since it came out, and come to same conclusion
Hi,
Currently, ControlNet is not working as expected in ComfyUI, because we use different implementation of sampling and other features in our code. We are working on integrating our implementation into ComfyUI directly, but it is not functional at the moment. Thank you for your understanding.
I have used Canny Controlnet in ComfyUI with the "Apply Controlnet (Advanced)" node. I'm getting good results by setting end_percent very low, like 0.01-0.05 to do just 1 or 2 steps. That points the image in the direction I want and then leaves Flux free to do what it wants. With end_percent too high the image quality degrades badly.
