🎨

Image Edit - QWEN-Image-Edit-2511

🛠

Advanced Image-to-Image Editing

An enhanced image editing model featuring significantly improved character and multi-person consistency, integrated LoRA support, enhanced industrial design generation, and stronger geometric reasoning, all driven by natural language instructions.

What it does

Qwen-Image-Edit-2511 is an image editing model developed by Qwen (Alibaba), and is an enhanced version of Qwen-Image-Edit-2509. It takes one or more reference images alongside a natural language instruction and produces a high-fidelity edited output. The model is particularly strong at preserving subject identity across edits, compositing multiple subjects into coherent scenes, and handling practical design and engineering scenarios. It supports the QwenImageEditPlusPipeline via Hugging Face Diffusers and is deployable on CUDA-compatible hardware.

Problem it solves

  • Image drift & consistency – Mitigates drift between input and output, keeping edits faithful to the source subject's identity and visual characteristics
  • Multi-person consistency – Enables high-fidelity fusion of two separate person images into a coherent group scene
  • LoRA integration – Selected community-created LoRAs (e.g., lighting enhancement, novel viewpoint generation) are baked directly into the base model, removing the need for extra tuning steps
  • Industrial design – Supports batch product design generation and material replacement for engineering and manufacturing workflows
  • Geometric reasoning – Can generate auxiliary construction lines and annotations directly within edited images, useful for design and technical applications
  • Creative workflows – Suitable for portrait editing, character storytelling, style transfer, and multi-subject scene composition

Input/Output

  • Input: Images and text instructions
    • Required:
      • Image: RGB image to edit (the main image being modified)
      • Prompt: Natural language text describing desired edits
    • Optional:
      • Reference Image: Additional image used as style/content reference for edits
    • Diffusion Steps: 5-40 (higher = better quality, slower)
    • Random Seed: Enter a number for reproducibility, or use -1 to randomly generate.
    • Output Image Size - Width x Height: 512 ~ 1600
      • image
  • Output: High-quality RGB image with edits applied
    • If reference image provided: Uses reference as guidance (e.g., copy specific details from reference)
    • If no reference image: Generates based on text prompt alone (e.g., creates random variation)

Accuracy & Speed

  • Key Enhancements over Qwen-Image-Edit-2509
    • Feature
    Improvement
    Image Drift
    Mitigated — edits stay closer to source identity
    Character Consistency
    Significantly improved for single-subject edits
    Multi-Person Consistency
    New — fuses two separate portraits into coherent group photos
    LoRA Support
    Community LoRAs integrated natively (lighting, viewpoints, etc.)
    Industrial Design
    Enhanced batch product design & material replacement
    Geometric Reasoning
    New — generates construction lines and design annotations

Technical Details

Architecture
Diffusion-based image editing pipeline (QwenImageEditPlusPipeline)
Framework
Hugging Face diffusers (latest version required)
Precision
bfloat16
Default Inference Steps
40
Guidance Scale
1.0 (default); true CFG scale: 4.0
Multi-Image Input
Supported (list of images)
LoRA
Community LoRAs integrated into base model weights
Languages
English, Chinese

Compliance & Provenance

Provider
Open-source (Alibaba)
Provider type
Specialized
License
Apache 2.0
EU AI Act risk class
Limited Risk
Art. 50 transparency
Required — outputs are marked. See AI Policy §2.
Region availability
Available globally
Training data summary
Pending — provider has not yet published per Art. 53(d)

For more on how we classify models and mark outputs, see our AI Policy.

Model Source