Qwen-Image-2.1 Demo & ComfyUI Hub

Alibaba Qwen & Comfy-Org

Qwen-Image-2.1 Multimodal Image Studio

Supports native 2K resolution (2048x2048), bilingual typography (Chinese/English), multi-image sequential editing, and transparent RGBA alpha output.

Model Resolution

Native 2K & 1K (2688x1536)

Transparency

Native RGBA Alpha

Generation Mode

Input Reference Images (1 to 10 images)

Click or drag & drop reference images

Supports PNG, JPG, WEBP (up to 10 images)

Prompt

Curated Capability Presets:

Advanced Generation Settings

Resolution & Aspect Ratio

Inference Steps

Seed (0 = Random)

Negative Prompt

Generation Output

Ready

Synthesizing with Qwen-Image-2.1...

Download PNG

Resolution: 2048x2048 | Steps: 30 | Model: Qwen-Image-2.1

ComfyUI Official Workflow Hub

Repackaged weights and ready-to-run workflow JSON templates for Comfy-Org/Qwen-Image-2.1.

ComfyUI v0.3+ Native

Model Weight Directory Placement Place in your local ComfyUI folder

Diffusion Model: qwen_image_2.1_bf16.safetensors

ComfyUI/models/diffusion_models/

Text Encoder: qwen3vl_8b_bf16.safetensors

ComfyUI/models/text_encoders/

VAE Model: qwen_image_2.1_vae_bf16.safetensors

ComfyUI/models/vae/

Standard Text-to-Image (T2I)

High fidelity 2K generation with Qwen3-VL conditioning and Euler sampler.

Image Edit & Multi-Ref

Supports multi-image visual conditioning for continuous character turnarounds & storyboards.

RGBA Transparent Generation

Direct alpha-channel generation with native transparent background for stickers and assets.

Capability 01

Flawless Multilingual Typography

Qwen-Image-2.1 leads open-weight models in rendering crisp, accurately spelled English phrases and complex Chinese characters directly in the generated graphics.

Capability 02

Multi-Panel Sequential Storyboards

Generate 6-panel or 3-view character Turnaround sheets with strict character, costume, and lighting consistency throughout the frames.

Capability 03

Native Alpha (RGBA) Transparency

Export true RGBA PNG files with transparent backgrounds without requiring a separate background matting model.

Capability 04

Native 2K Resolution

Directly generates up to 2688x1536 or 2048x2048 images in a single pass with fine-grained micro-details.

Python & Diffusers ZeroGPU Implementation

The code below is included in app.py and can run seamlessly on Hugging Face ZeroGPU:

import spaces  # Must come before torch / diffusers
import torch
import gradio as gr
from diffusers import DiffusionPipeline

MODEL_ID = "Qwen/Qwen-Image-2.1"

# Load model eagerly at module scope
pipe = DiffusionPipeline.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.bfloat16
).to("cuda" if torch.cuda.is_available() else "cpu")

@spaces.GPU(duration=120)
def generate_image(prompt, negative_prompt="", width=1024, height=1024, steps=30, seed=42):
    generator = torch.Generator("cuda").manual_seed(int(seed))
    image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt or None,
        width=int(width),
        height=int(height),
        num_inference_steps=int(steps),
        generator=generator,
    ).images[0]
    return image

gr.Interface(
    fn=generate_image,
    inputs=[
        gr.Textbox(label="Prompt", lines=3),
        gr.Textbox(label="Negative Prompt"),
        gr.Slider(512, 2048, value=1024, step=64, label="Width"),
        gr.Slider(512, 2048, value=1024, step=64, label="Height"),
        gr.Slider(10, 50, value=30, step=1, label="Steps"),
        gr.Number(value=42, label="Seed"),
    ],
    outputs=gr.Image(label="Result"),
    title="Qwen-Image-2.1 ZeroGPU Demo"
).launch(mcp_server=True)