I merged the official workflows.(Img-qwen_2.1-Mix)
you can process t2i, image edit in one workflow now.
Reference output:
RTX5080 laptop 16G vRAM + 64G RAM (1536*1536):
- T2I: Prompt executed in 32.14 seconds
- Edit: Prompt executed in 64.43 seconds
Requirements:
Models:
- qwen_image_2.1_int8_convrot.safetensors, Put in: ComfyUI\models\diffusion_models
- qwen3vl_8b_int8_convrot.safetensors; (CLIP) Put in: ComfyUI\models\text_encoders
- VAE Put in: ComfyUI\models\vae
- qwen_image_2.1_vae_bf16.safetensors
https://huggingface.co/Comfy-Org/Qwen-Image-2.1/resolve/main/vae/qwen_image_2.1_vae_bf16.safetensors
ComfyUI Nodes:
- rgthree-comfy
- ComfyUI-KJNodes
Description
FAQ
Comments (3)
thanks for this neat WF, do you know if loras made for 2511 will work for this 2.1 ?
no, It’s need new lora.
I was thinking... since on 16gb of vram, the model+vae+textencoder overflows, we need to put VRAM clearing inbetween the nodes so it always stays in the VRAM and never goes into DRAM.
Anyone tried that?