Workflow for this video: https://youtu.be/lgFxqG8eam4

If you want to see a Flux deep dive, watch it.

This Workflow is a collection of four different pipelines:

- a basic txt2img Flux Dev fp16 workflow;

- a basic txt2img Flux Schnell fp8 quantized workflow;

- a LoRA + ControlNet Canny Flux Dev fp16 workflow;

- a LoRA + ControlNet Canny Flux Dev fp16 + 1.5 IC-Light pipeline workflow

Want to support me? You can buy me a coffee here: https://ko-fi.com/risunobushi

Download and place the dev FP16 version of Flux in the /models/unet folder:

https://huggingface.co/black-forest-labs/FLUX.1-dev/blob/main/flux1-dev.safetensors

Download and place the VAE in the VAE folder:

https://huggingface.co/black-forest-labs/FLUX.1-dev/tree/main/vae

Download and place the text encoder models (clip_l and either t5xxl fp8 or both fp8 and 16) in the /models/clip folder:

https://huggingface.co/comfyanonymous/flux_text_encoders/tree/main

If you run out of memory, under Load Diffusion Model's weight_dtype, select fp8_e4m3fn

If you run out of memory, under DualCLIPLoader select the t5xxl_fp8 model

The LORA can be bypassed if you don't care about the realism LORA.