My TG Channel - https://t.me/StefanFalkokAI
My TG Chat - https://t.me/+y4R5JybDZcFjMjFi
Hi! I introduce my working workflows with Flux.2 Klein 9B for ComfyUI.
I have included workflows with Flux.2 Klein 9B models
You need to have Flux.2 Klein 9B KV Model (https://huggingface.co/black-forest-labs/FLUX.2-klein-9b-kv/blob/main/flux-2-klein-9b-kv.safetensors), clip (https://huggingface.co/Comfy-Org/vae-text-encorder-for-flux-klein-9b/tree/main/split_files/text_encoders) and vae (https://huggingface.co/Comfy-Org/flux2-dev/tree/main/split_files/vae)
if you need Base version - https://huggingface.co/black-forest-labs/FLUX.2-klein-base-9B/blob/main/flux-2-klein-base-9b.safetensors
also recommend download Flux 2 Klein 9B by wikeeyang, it gives a better result - https://civarchive.com/models/2339723/flux2-klein-9b-true
https://huggingface.co/datasets/StefanFalkok/ComfyUI_portable_torch_2.13.0_cu130_cp313_sageattention_triton - my new ComfyUI build with torch 2.13.0cu130 + sageattention + my workflows, also u need download and install CUDA 13.0 (https://developer.nvidia.com/cuda-13-0-0-download-archive) and VS Code (https://visualstudio.microsoft.com/downloads/)
Leave comments if you have trouble or you found the problem with workflows
Description
fixed edit workflow
added source for Flux 2 Klein KV Model
added latent upscale method to get more cool image
FAQ
Comments (6)
Working, but the results look like insane amounts of plastic compared to the non kv. Maybe you set a wrong sampler/scheduler?
1. I use recommend 1mp for getting more stable results
2. I recomend use Flux 2 Klein True instead kv - yoo'll get even better results
it's not sampler/scheduler problem. Flux 2 Klein gives more plastic results when you set resolution from 2mp to 4mp. So set 1mp and adjust upscale latent by parameters
@Stefan_Falkok Nah, what Im saying is I dont have the issue with plastic-y looking faces using a workflow that does not use the new kv. I'm almost certain there are better options and settings for the samples and start/end steps. I'll keep messing around and see what I can come up with.
Removed uni-pc - converted to a two step ksampler with returned noise from the first. Looking much better. Thanks for getting me started with this.
@singularity42 Share workflow? Or at least a screenshot of it?
@freakingpogi3431 I go through work flows like water. Ive already moved way beyond that. Sorry.