Description
Explore Ming Image 0.1 in one organized ComfyUI workspace. This workflow combines nine processing modes on a single canvas, with bright color-coded groups, a compact grid, detailed instructions, practical tips and direct model download links.
Two complete versions are included: English and German, with identical connections and functionality.
What’s inside
Text-to-Image / Text-to-Design — Create posters, advertising, editorial layouts, infographics, product visuals, website concepts, UI mockups, illustrations and photographic scenes.
Local Prompt Enhancement — Turn a short idea into a detailed layout prompt using an optional local language model.
Transparent RGBA Generation — Generate PNG images with a real alpha channel for isolated graphics, characters and compositing assets.
Single-Image Editing — Use an input image and a prompt to guide changes through Ming’s native reference conditioning.
Multiple Reference Images — Combine reference inputs for image-guided generation. Two inputs are connected in the workflow; the Core node supports up to eight.
Vision-Only Guidance — Explore image conditioning through the vision encoder without adding VAE reference latents.
Experimental Img2Img — Start from an existing image and explore variations through denoise control.
Layer Decomposition — Use the separate Ming Design-Layer model to split a flattened design into transparent raster layers.
Experimental Masked Img2Img — Explore localized variations, with the source image composited back outside the resized mask.
Built for practical use
Prompt tips sit directly below the prompt areas, while setting notes explain the relevant controls. Download information covers the Design model, Layer model, text encoders and VAE, with separate compatibility notes for available quantized variants.
Only the direct generation section is enabled initially. Activate the section you want to use and mute the others, so optional pipelines do not run together.
For detailed design work, 2048 × 2048 provides approximately 4.19 megapixels. Other aspect ratios are supported, including portrait and landscape compositions. The included examples contain actual local Ming outputs, recorded prompts and generation settings, plus before-and-after comparisons.
Requirements
A recent ComfyUI installation with Ming support. The workflow was checked with ComfyUI 0.38.1.
Ming Design weights, a compatible Ling text encoder and the matching VAE.
Separate Design-Layer weights and its text encoder for decomposition.
An additional local language model for the optional prompt enhancement section.
All processing nodes come from ComfyUI Core; no additional custom node pack is required. Model weights are downloaded separately.
Validation and limitations
Both language versions were imported into ComfyUI, and all nine sections passed node and connection validation. Text-to-image, transparent generation, single-image editing, experimental Img2Img and layer decomposition were also exercised in actual local generation runs. The other optional sections have been validated structurally but have not all been tested through GPU inference.
Long text, anatomy and precise layout can still produce mistakes. Experimental modes are clearly labeled. Layer decomposition produces raster PNG layers; website and UI outputs are visual mockups. GGUF and FP8 references are informational and are not presented as validated drop-in configurations.