Qwen Image 2.1 - Text-to-Image and Image Edit in ComfyUI (Patreon)

Video : https://youtu.be/F7VeVDunaok

Hey everyone.

New video is up, and this one is about Qwen Image 2.1. It is a 7 billion parameter open weights image model that does text to image and image edit in one model, and it runs natively in ComfyUI on day zero.

Two things surprised me. First, transparency. Back in December Qwen shipped a separate model just for transparent images. Now it is built straight in. You write a prompt and get an RGBA image with a real alpha channel, or you feed it a normal photo and it cuts the subject out into a transparent layer. Second, the editing side takes up to 10 reference images and holds faces and product details across edits.

The two official ComfyUI workflows are in the description, so you can grab those and run it right away.

Patreon Supporters, please check out the member post for your contents. Thanks.

See you in the next one.

— Benji

Resources:

- My Prompt Enhancer custom node: https://github.com/benjiyaya/ComfyUI-Qwen-Image-2.1-Prompt-Enhancer

- Qwen Image 2.1 ComfyUI weights (Comfy Org): https://huggingface.co/Comfy-Org/Qwen-Image-2.1

- Qwen Image 2.1 official weights: https://huggingface.co/Qwen/Qwen-Image-2.1

- Qwen Image 2.1 GitHub: https://github.com/QwenLM/Qwen-Image-2.1

- Qwen Image 2.1 blog: https://qwen.ai/blog?id=qwen-image-2.1

- ModelScope mirror: https://modelscope.cn/models/Qwen/Qwen-Image-2.1

- ComfyUI text to image template: https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_qwen_image_2_1_t2i.json

- ComfyUI image edit template: https://github.com/Comfy-Org/workflow_templates/blob/main/templates/image_qwen_image_2_1_image_edit.json

Diffusion model — folder: ComfyUI/models/diffusion_models/

- qwen_image_2.1_bf16.safetensors (14.2 GB) — full precision, used in the VIP workflow and in my custom workflow

- qwen_image_2.1_int8_convrot.safetensors (7.26 GB) — lower VRAM build, used in the official ComfyUI templates

Text encoder — folder: ComfyUI/models/text_encoders/

- qwen3vl_8b_bf16.safetensors (17.5 GB) — full precision

- qwen3vl_8b_int8_convrot.safetensors (9.35 GB) — lower VRAM build

- qwen3vl_8b_w4a8.safetensors (6.31 GB) — lowest memory build, mentioned for lower VRAM GPUs

VAE — folder: ComfyUI/models/vae/

- qwen_image_2.1_vae_bf16.safetensors (676 MB) — RGBA autoencoder, 16x spatial compression