Qwen Image 2.1 Text-to-Image
Qwen Image 2.1 Makes Detailed 2K Images Fast
AI model workflow
Qwen Image 2.1 ComfyUI Workflows
Create images from text or edit photos with Qwen Image 2.1 in ComfyUI. Explore reference-guided edits, outfit changes, face swaps, multi-image compositions, and masked inpainting. Run on RunComfy Cloud or download the workflow JSON.
6 workflows
Featured workflow
Qwen Image 2.1 ComfyUI Workflow: Text-to-Image
The workflows
6 Qwen Image 2.1 ComfyUI Workflows
Choose from 6 Qwen Image 2.1 ComfyUI workflows. Start from a prompt, a source photo, multiple references, or an image and mask. Each example includes instructions, a preview, and JSON download. Run it on RunComfy Cloud, download the workflow JSON, or deploy supported workflows as APIs.
Qwen Image 2.1 Inpainting
Qwen 2.1 local edits add or replace details with precision.
Qwen Image 2.1 Multi-Image Editing
Powerful multi-reference editing creates precise scenes from separate images.
Qwen Image 2.1 Face & Head Swap
Qwen 2.1 face swaps deliver natural faces while keeping every scene.
Qwen Image 2.1 Outfit Change
Qwen Image 2.1 swaps outfits fast with precise, realistic detail.
Qwen Image 2.1 Image Editing
Qwen 2.1 Edit delivers precise image changes fast.
6 of 6 workflows shown
About Qwen Image 2.1
What can Qwen Image 2.1 do in ComfyUI?
Qwen Image 2.1 is Alibaba's Qwen model for creating and editing images with the same underlying weights. Its visual generation component has 7 billion parameters. The release brings native transparency, richer image detail, and text rendering that considers lettering and layout alongside the rest of the composition.
Choose a graph for your inputs: text-to-image starts from a prompt; general editing uses a base image and optional references; outfit and face swaps use a source photo plus a matching reference. The multi-image setup combines separate visual inputs, while inpainting adds a mask for local changes.
Start with the featured INT8 text-to-image workflow for a 2048 × 2048 image. Qwen Image 2.1 is a distinct release from Qwen Image 2512 and Qwen Image Edit 2509 or 2511; use the matching model files and encoder for your chosen workflow. Open its detail page to inspect the graph, examples, and setup instructions.
References and model resources: Qwen Image 2.1 announcement, Qwen Image 2.1 model card, Qwen Image 2.1 GitHub, Explore Qwen Image 2.1 on Hugging Face, Explore Qwen Image 2.1 on Civitai.
Capabilities
Explore Qwen Image 2.1 capabilities in ComfyUI
Explore the tasks available in the workflows below. Open an example to see its image inputs, controls, and output format.
Text-to-image generation
Turn a written brief into an image, with control over the prompt, seed, and canvas size in the featured 2K setup.
Instruction-based image editing
Edit a base image with text and optional references to place products, replace image text, or change style, lighting, and weather.
Multi-image editing
Combine subject, object, and scene references into one image. The workflow uses reference-aware prompt enhancement and supports up to 10 image inputs.
Inpainting and local edits
Paint a mask and describe the detail to add or replace. Aligned image and mask sizing keeps the edit focused on your selected area.
Qwen Image 2.1 on RunComfy
Create with Qwen Image 2.1 in a prepared cloud workspace
Open a setup with its matching Qwen Image 2.1 weights, encoder, VAE, and ComfyUI nodes ready to use. Adjust prompts, reference images, masks, and canvas size in the graph. Editing examples that use prompt enhancement include that stage in the workflow.
Run Qwen Image 2.1 nowEditable ComfyUI graph
Control prompts, seeds, and dimensions in the actual workflow, with the option to customize its nodes.
Workflow JSON to keep
Download the graph for your own ComfyUI installation and follow its matching model requirements.
Matching model components
Start with the configured 2.1 generation weights, text encoder, and VAE for the selected example.
Cloud GPU options
Choose GPU memory for your output resolution, weight format, and any reference-image processing.
Applications
What you can make with Qwen Image 2.1
Choose a published example for your visual goal, then adapt its prompts and available inputs to your own project.
Portraits and Cinematic Stills
Explore lighting, wardrobe, atmosphere, and camera framing from a written scene description.
Product Placement and Scene Restyling
Place a reference product into a setting or explore a new lighting, weather, or style treatment for an existing scene.
Outfit Changes and Virtual Try-On
Pair a person photo with an outfit reference to explore garment shape, color, and fabric for styling previews or fashion editorials.
Face and Head Swaps
Recast a character using a scene image and an identity portrait. Guide the face, hairstyle, and facial hair, then review likeness and scene fit.
Group Portraits and Styled Interiors
Build a group portrait from separate photos or style a room with furniture and decor references. Specify each image's role in the composition.
Targeted Image Refinements
Add or replace a person, object, or clothing detail inside a painted mask. Compare the result with the original to check edges and surrounding detail.
Frequently asked questions
Qwen Image 2.1 in ComfyUI: FAQ
Choose the right editing setup, understand references and masks, and check transparency, LoRA training, memory needs, and JSON downloads.
Which Qwen Image 2.1 ComfyUI workflow should I start with?
Start with text-to-image for a new image from a prompt, or reference-guided editing to change an existing photo. Choose the outfit or face-swap setup for those specific tasks, multi-image editing to combine references, and inpainting when you need a mask to define the edit area.
Is Qwen Image 2.1 the same as Qwen Image Edit 2511?
No. Qwen Image 2.1 unifies generation and editing in a new release. Qwen Image Edit 2511 belongs to the earlier editing line. Follow a workflow built for the version you want to run; replacing only the checkpoint in an older graph does not establish compatibility. The Qwen Image Edit workflows page covers those earlier editing setups.
Can Qwen Image 2.1 generate transparent PNG images?
Yes, the model supports native RGBA generation. Ask explicitly for a transparent background and use a graph that preserves the alpha channel through decoding and saving. Saving a PNG alone does not prove that its background is transparent; inspect the output's alpha channel before using it as a cutout or overlay.
How many reference images can Qwen Image 2.1 use?
The model and the multi-image editing workflow support up to 10 reference images. The supplied example uses four active images, with the remaining reference loaders muted. Enable those loaders and supply images when needed. Describe each image's role, such as the person, furniture, or setting. These references guide one composition, not a batch of independent edits.
How do I change an outfit or swap a face with Qwen Image 2.1?
Use the outfit-change workflow with a person photo and a clothing reference. Use the face and head-swap workflow with the original scene and an identity portrait. Write a short instruction saying what to transfer and what to keep. Review likeness, garment details, lighting, and unintended changes; preservation is not guaranteed.
Does Qwen Image 2.1 inpainting leave everything outside the mask unchanged?
Not necessarily. The inpainting workflow aligns the image and mask and applies noise inside the masked region, but decoding can still alter surrounding pixels. Keep the mask focused on the change and compare the output with the original, especially at the edges.
How much VRAM do I need for Qwen Image 2.1 in ComfyUI?
Memory use depends on weight precision, the text encoder, output size, reference images, and any prompt-enhancement stage. The featured workflow uses INT8 weights, but that does not establish a universal minimum VRAM requirement. Check the selected workflow's requirements, or use RunComfy Cloud with a GPU suited to that setup.
Can I use older Qwen Image LoRAs with Qwen Image 2.1?
Do not assume compatibility. A LoRA must match the model architecture and version used for inference. Choose Qwen Image 2.1 in RunComfy Trainer as your training base model. Follow the 2.1 LoRA training guide for generation, reference-guided editing, or transparency datasets; an older Qwen LoRA is not automatically compatible.
Where can I download the Qwen Image 2.1 workflow JSON?
Use Download JSON beside a workflow or on its detail page. Import the file into a compatible ComfyUI installation and add the model files and nodes specified by that example. JSON contains the graph, not the model weights. The featured text-to-image example is a starting point for a local setup.
Related resources
More ways to use Qwen Image 2.1
Train a Qwen Image 2.1 LoRA, use Qwen models without a full ComfyUI graph, or compare earlier editing releases.
Train a Qwen Image 2.1 LoRA
Customize Qwen Image 2.1 in RunComfy Trainer. Select the 2.1 base model and prepare a dataset for generation, reference-guided editing, or transparency.
Train Qwen Image 2.1Explore Qwen Image Models
Use Qwen Image models directly, including 2.1 text-to-image and editing, when you do not need a full ComfyUI graph. Check each model's inputs and version.
Browse Qwen Image ModelsCompare Earlier Qwen Image Edit Workflows
Explore the 2509, 2511, and other Qwen editing setups when your existing graph or LoRA uses an earlier model.
Browse Qwen Image Edit Workflows