Repolode
Submit
inclusionAI

inclusionAI/Ming-Image

Open-source 6B image models for generating visual designs with text and decomposing flattened design images into editable transparent layers.

Open-source alternative to:Canva

View on GitHub
Stars
215
Forks
14
Open issues
5
Last commit
22 days ago

What is it?

Ming-Image-0.1-Design is a series of two 6B-parameter models for visual design. The Design model generates complete layouts for UI screens, infographics, posters, and other text-rich compositions, and can produce transparent-background images. The Layer model takes a flattened design image and decomposes it into independently editable RGBA layers, with the layer count set by the prompt.

The repository provides a command-line inference script (infer.py) that loads checkpoints from a local directory or a Hugging Face Hub repo ID, with options for pinning revisions, choosing attention implementation, and validating a checkpoint without loading weights. The default configuration targets one GPU with at least 80 GiB of memory running in BF16. Prompt rewriting with a vision-language model is described as a separate pre-processing step, and the repo includes the system prompts used for it.

VLLM-Omni is listed as a recommended serving framework. Two agent skills built on the models, for UI design and image-to-editable-PPT conversion, are linked from the project page.

Key features

  • 01Generates complete visual designs with text for UI, infographics, and posters.
  • 02Produces transparent-background images with alpha channels.
  • 03Decomposes flattened design images into separate transparent RGBA layers.
  • 04CLI inference script with local or Hugging Face Hub checkpoint loading.
  • 05Supports multiple output sizes and aspect ratios for text-to-image generation.
  • 06Includes prompt-rewriting system prompts for structured, layer-based prompts.

Use cases

  • Generating UI mockups or landing page drafts from a text description.
  • Creating poster or infographic drafts with rendered text.
  • Separating a flattened design into editable layers for revision.
  • Converting a generated slide image into an editable PowerPoint slide with an agent workflow.

Quick start

  1. 1

    Install the runtime dependencies

    pip install -r requirements.txt
  2. 2

    Restrict the process to one GPU

    export CUDA_VISIBLE_DEVICES=0
  3. 3

    Check the checkpoint and task without loading weights

    python infer.py --model inclusionAI/Ming-Image-0.1-Design --task text-to-image --prompt "A red circle on a white background" --width 2560 --height 1440 --validate-only

Things to know

  • The validated setup requires one GPU with at least 80 GiB of memory running in BF16.
  • Full model inference needs CUDA-capable PyTorch and Python 3.10 or newer.
  • Good prompts rely on an external vision-language model for prompt rewriting, which the CLI does not include.

Topics

#image-generation#text-to-image#diffusion-model#design#layer-decomposition#transparent-background#python

FAQ

What is Ming-Image?
Open-source 6B image models for generating visual designs with text and decomposing flattened design images into editable transparent layers. Ming-Image-0.1-Design is a series of two 6B-parameter models for visual design. The Design model generates complete layouts for UI screens, infographics, posters, and other text-rich compositions, and can produce transparent-background images. The Layer model takes a flattened design image and decomposes it into independently editable RGBA layers, with the layer count set by the prompt.
Is Ming-Image free to use? What license does it have?
Ming-Image is open source under the MIT license. Check the license terms before using it in commercial products.
Is Ming-Image still maintained?
Actively maintained. The last commit to Ming-Image was on Sep 28, 2026.
What can I use Ming-Image for?
Generating UI mockups or landing page drafts from a text description. Creating poster or infographic drafts with rendered text. Separating a flattened design into editable layers for revision. Converting a generated slide image into an editable PowerPoint slide with an agent workflow.
How do I install Ming-Image?
Install the runtime dependencies: pip install -r requirements.txt Restrict the process to one GPU: export CUDA_VISIBLE_DEVICES=0 Check the checkpoint and task without loading weights: python infer.py --model inclusionAI/Ming-Image-0.1-Design --task text-to-image --prompt "A red circle on a white background" --width 2560 --height 1440 --validate-only
Where can I get Ming-Image?
The source code and installation instructions are on GitHub: https://github.com/inclusionAI/Ming-Image

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects