Ming Image Design Layer

Ming-Image-0.1-Design-Layer takes a finished, flattened design and pulls it back apart into separate transparent layers. Give it an image and a layer plan and it returns the requested number of RGBA layers - background, shapes, text - each one editable on its own. It is the companion to Ming-Image-0.1-Design, and a different checkpoint with a different job.

Originally released by inclusionAI on Hugging Face under the MIT license. All credit for the model goes to the inclusionAI team. Civitai is hosting a mirror so the weights are easy to get hold of - please head to the original repository for updates and to follow the project directly.

Download only - this one does not run on-site

Layer decomposition is not a workflow the Civitai generator offers, so there is no Generate button here. The files below are for running it yourself, in ComfyUI or with inclusionAI's own inference code. If you want to generate Ming designs on-site, that is the Ming Image Design model.

Built by

• inclusionAI - the Ming-Image series and the Ling-mini-2.0 text encoder it runs on
• The Ming-Image repository - reference inference code and the layer-decomposition demo
• Comfy-Org - the repackaged single-file weights this mirror is built from

What it produces

One input image plus a prompt describing the layer plan, and you get back that many RGBA layers with real alpha. The released card-making example splits a greetings card into six. Layer count follows the plan you give it rather than being fixed, and the output keeps the reference image's aspect ratio.

Recommended settings

• Resolution: 1024 for quality, 512 for speed. Unlike text-to-image, output is not square - the working size is picked from the selected bucket and the reference ratio is preserved.
• Steps: 12.
• CFG scale: 2.0. Note this differs from Design, which wants 1.0.
• Precision: BF16.

Files here

The Comfy-Org repackaging of both halves of the Layer checkpoint: the diffusion model in BF16 and an int8 quantization, and its own Ling-mini-2.0 text encoder in BF16 and int8. The text encoders are specific to Layer and are not interchangeable with the ones on the Design model.

Links

• Source: inclusionAI/Ming-Image-0.1-Design-Layer on Hugging Face
• Code: github.com/inclusionAI/Ming-Image
• ModelScope: inclusionAI/Ming-Image-0.1-Design-Layer
• Demo: Hugging Face Space
• Agent skills: Ling UI Design and Image to Editable PPT