Ming-Image-0.1-Design
Ant Group's MIT-licensed design image model — legible text, native transparent backgrounds, editable layers.
About
Ming-Image-0.1-Design, released September 17, 2026 by inclusionAI (the AI lab associated with Ant Group) under the MIT license, is a text-to-image model aimed at text-heavy graphic design rather than photography: UI mockups, dashboards, infographics, and posters where rendered text must stay legible.
The diffusion transformer is 6.15B parameters, paired with a 17.01B Ling-mini-2.0 text encoder, totaling roughly 52.9 GB of weights; the validated config is a single CUDA GPU with 80 GiB of VRAM. Inference runs at 12 steps with CFG 1.0 in BF16 at 2048x2048. It emits native RGBA, so genuine transparent backgrounds come out of the sampler rather than a matting pass.
A companion model, Ming-Image-0.1-Design-Layer, takes a flattened design and splits it back into editable RGBA layers (text, cards, subject, background). It sits #1 among open-weight models on Artificial Analysis's UI/UX Design leaderboard with an Elo of 1082.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Independent (Artificial Analysis): #1 open-weight on the UI/UX Design board (Elo 1082, 17th of 81 overall); 995 Elo on the general text-to-image board. Provider-reported: Layer variant leads all 12 evaluated Crello settings, 4.3x faster than a 20B open-weight baseline.
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Release facts (September 17, 2026; inclusionAI/Ant Group; MIT license; 6.15B DiT + 17B text encoder; native RGBA; AA UI/UX Design Elo 1082 #1 open-weight) corroborated across Orcarouter, AlphaSignal, Digital Applied, and the public model cards. Docked for: (1) ~53GB download and 80GB VRAM requirement put it out of reach of consumer hardware despite the open license; (2) the Layer variant's 4.3x speedup figure is the vendor's own evaluation.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 5.0/10
- Maintenance · recency, releases, issue load
- 8.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 8.0/10
- · Adoption 5: specialist design model; ~53GB download and 80GB VRAM requirement limit audience
- · Maintenance 8: released 2026-09-17; card refreshed Sep 22-23; community runtime support emerging (sd.cpp docs, vLLM-Omni)
- · Openness 10: MIT (0.7*10) + public weights on Hugging Face verified (0.3*10)
- · Manual 8: release facts (Sep 17 2026; inclusionAI/Ant Group; MIT; 6.15B DiT + 17B encoder; native RGBA; AA UI/UX Elo 1082 #1 open-weight) corroborated across Orcarouter, AlphaSignal, Digital Applied, public cards; docked for consumer-unreachable hardware requirement, vendor's own Layer speedup figure