MAI-Image-2
#13microsoft · v2 · seit 19. März 2026 · 17× · zuletzt 30. Juni 2026
25
Momentum
MAI-Image-2 is Microsoft's second-generation proprietary text-to-image model, built on a flow-matching diffusion architecture with 10–50 billion non-embedding parameters. It is optimized for photorealism, reliable in-image text rendering, and complex creative workflows. Released on March 19, 2026 in MAI Playground and Microsoft Foundry, it debuted as a top-3 model family on the Arena.ai leaderboard. API access is available through Microsoft Foundry, and the model is also integrated into Copilot, Bing Image Creator, and PowerPoint.
Historique du momentum
04.04.03.07.
Fonctionnalités
| Fine-Tuning | Not documented/available for MAI-Image-2; weight tuning first announced for MAI-Image-2.5 (Build 2026) |
| Generation Time | 2–4 sec. (typical, Foundry API, P50); MAI-Image-2-Efficient: ~13.7 sec. P50 median per official benchmark |
| License | Product-specific terms of use per platform (MAI Playground Terms, Foundry Public Preview Terms) – no open-source license |
| Max Resolution | 1024 × 1024 px (max.); also 1365×768 (landscape) and 768×1365 (portrait) available |
| Platform | Microsoft Foundry (API), MAI Playground, Copilot, Bing Image Creator, PowerPoint |
| Price | $5 / 1M text input tokens; $33 / 1M image output tokens (Foundry API) |
| Release Date | March 19, 2026 (MAI Playground & Foundry Public Preview) |
