DreamOmni2 Interactive Demo

Try DreamOmni2 directly in your browser with this interactive demo. No installation required. Upload your images, provide text instructions, and see the model generate or edit images in real-time.

Generation Demo

The generation demo allows you to create new images from scratch. Upload reference images showing the elements you want to incorporate, provide a text description of the scene you want to create, and watch DreamOmni2 generate a completely new image combining your references.

How to Use the Generation Demo

  1. Upload one or more reference images (up to 4 images)
  2. Enter your text instruction describing the desired output
  3. Adjust the output resolution if needed
  4. Click Generate to create your image
  5. Download or share your result

Demo Examples and Use Cases

Here are some example scenarios you can try with the demo to explore DreamOmni2's capabilities:

Object Replacement

Task: Replace an object in one image with an object from another image.

Example Instruction: "Replace the chair in the first image with the modern sofa from the second image, maintaining the same lighting and perspective."

This demonstrates how DreamOmni2 can swap objects while preserving scene consistency.

Style Transfer

Task: Apply the artistic style from one image to another image.

Example Instruction: "Apply the watercolor painting style from the second image to the portrait in the first image."

Shows abstract concept handling for artistic styles and visual effects.

Character Composition

Task: Combine multiple characters into a new scene.

Example Instruction: "Create a scene where the person from the first image and the person from the second image are standing together in a park, having a conversation."

Demonstrates multi-subject generation with spatial positioning.

Material Application

Task: Change the material or texture of objects.

Example Instruction: "Change the wooden table in the first image to have the marble texture shown in the second image."

Shows how DreamOmni2 handles abstract material concepts.

Pose Imitation

Task: Make a character adopt a pose from a reference image.

Example Instruction: "Make the person from the first image adopt the yoga pose shown in the second image, keeping their identity and clothing."

Demonstrates pose transfer while maintaining identity consistency.

Background Replacement

Task: Change the background while keeping the subject.

Example Instruction: "Place the person from the first image into the beach setting shown in the second image."

Shows precise subject isolation and background composition.

Pattern Application

Task: Apply a pattern or design from one image to another.

Example Instruction: "Apply the floral pattern from the second image to the dress in the first image."

Demonstrates abstract pattern extraction and application.

Lighting Adjustment

Task: Match the lighting conditions between images.

Example Instruction: "Adjust the lighting on the subject in the first image to match the golden hour lighting shown in the second image."

Shows understanding of abstract lighting concepts.

Tips for Best Results

  • Use clear, high-quality reference images with good lighting
  • Be specific in your text instructions about what you want
  • Reference the image order in your instructions (first image, second image, etc.)
  • For editing tasks, place the image to be edited first
  • Experiment with different instruction phrasings if results are not as expected
  • Start with simpler tasks before attempting complex multi-element compositions
  • Allow sufficient processing time for high-resolution outputs

Understanding Generation vs Editing

The demo focuses on generation mode, which creates entirely new images from your references. Generation mode offers more creative freedom since it regenerates the entire image rather than preserving specific pixels from a source.

For tasks requiring precise preservation of non-edited areas (true editing mode), you may want to use the local installation with the editing script, which ensures pixel-perfect consistency in unchanged regions.

Limitations and Considerations

  • Processing time varies based on output resolution and server load
  • Very complex instructions may not always produce expected results
  • The demo has file size limits for uploaded images
  • Generated images should be reviewed and may require multiple attempts
  • Results depend on the quality and relevance of reference images
  • Some abstract concepts may be more challenging than concrete objects

Additional Resources

For more control and advanced features, consider installing DreamOmni2 locally:

  • Visit the Installation page for setup instructions
  • Check the GitHub repository for the latest updates and examples
  • Read the research paper for technical details about the model
  • Join the community discussions to share results and get tips

Note: This demo is provided for educational and research purposes. Generated images should be used responsibly and in compliance with applicable laws and ethical guidelines.