This is advanced ControlNet workflow based on Flux depth and canny models.
https://huggingface.co/black-forest-labs/FLUX.1-Depth-dev
https://huggingface.co/black-forest-labs/FLUX.1-Canny-dev
The workflow designed to get advantage of using 2 advanced KSamplers, which helps to reduce artifacts caused by CNet and overall produces better results. Gives more control over image generation, allows you to pick different samplers and schedulers.
Other functionalities:
Flux model loaders including GGUF variants.
Lora stacks.
Promting with Florence2 nodes.
Img2Img with autoresizing and cropping.
Additional ControlNet manupilations such as:
CNet image resizing
Blur (optional)
Sharpen (optional)
More info provided inside workflow, including manual on how to use 2 advanced KSamplers.
Description
FAQ
Comments (9)
Very good workflow for Flux Controlnet.
I however struggle to get decent photo-realistic results with the default noise steps. Do you have any recommendation for me?
Thank you! I haven't noticed it before, but the easiest way is to use depth\canny lora with lower strength around 0,5-0,6 instead of full scaled model. You can even isolate lora usage by connecting diffusion model directly to 2nd KSampler, ensuring lora works only for 1st KSampler. Also lowering flux guidance from 30 to even basic 2.0-3.5 works fine + you can try to use different combinations of samplers and schedulers. Generally, samplers like deis, ddim, uni_pc_bh2 with beta and ddim_uniform schedulers give better skin textures. Additionaly, you can use some realistic loras like amateur photography and flux realism . Feel free to ask if you have anymore questions :)
@NexarCG thanks for your detailed explanation! I will try and play with mentioned settings, thank you! :)
Great workflow, thanks for sharing...one note, will be good a workflow to work with canny and depth at same time. thanks
Hello, thanks for the workflow, but i have no result when i add the lora depth (trying 0.5 and 1) Strengh. do you know why ?
hi, maybe you're trying to load both depth/canny with their full scale models simultaneously instead of base flux models? or you just forgot to change preprocessor? otherwise i have no idea why it wouldn't work for you
it's work if i put "flux dev" in Diffusion models, and do not work with flux depth. So i dont understand
@Jubien you either use flux dev with depth/canny lora OR you just use full scaled flux depth dev/flux canny dev models without any lora.
@Nexssar thanks so mutch I didn't know that :)