MAI-Image-1
MAI-Image-1 is the first fully in-house text-to-image generation model from Microsoft that has debuted in the top ten on the LMArena benchmark. It was engineered with a goal of delivering genuine value for creators by emphasizing rigorous data selection and nuanced evaluation tailored to real-world creative use cases, and by incorporating direct feedback from professionals in the creative industries. The model is designed to deliver real flexibility, visual diversity, and practical value. MAI-Image-1 excels at generating photorealistic imagery, for example, realistic lighting (bounce light, reflections), landscapes, and more, and it offers a compelling balance of speed and quality, enabling users to get their ideas on screen faster, iterate quickly, and then transfer work into other tools for refinement. It stands out when compared with many larger, slower models.
Learn more
MAI-Image-2
MAI-Image-2 is an advanced text-to-image model developed to enhance creative workflows with highly realistic and detailed visual outputs. It is ranked among the top three model families on the Arena.ai leaderboard, reflecting strong real-world performance. The model is designed in collaboration with creatives, including photographers and designers, to meet practical artistic needs. It delivers enhanced photorealism with accurate lighting, textures, and lifelike environments. MAI-Image-2 also improves in-image text generation, enabling users to create posters, infographics, and visual content with embedded typography. The model supports complex and imaginative scene creation, from cinematic visuals to abstract compositions. Available through platforms like MAI Playground, Copilot, and Bing Image Creator, it allows users to experiment and generate high-quality visuals.
Learn more
FLUX.1 Kontext
FLUX.1 Kontext is a suite of generative flow matching models developed by Black Forest Labs, enabling users to generate and edit images using both text and image prompts. This multimodal approach allows for in-context image generation, facilitating seamless extraction and modification of visual concepts to produce coherent renderings. Unlike traditional text-to-image models, FLUX.1 Kontext unifies instant text-based image editing with text-to-image generation, offering capabilities such as character consistency, context understanding, and local editing. Users can perform targeted modifications on specific elements within an image without affecting the rest, preserve unique styles from reference images, and iteratively refine creations with minimal latency.
Learn more
Bonsai Image
Bonsai Image Ternary 4B MLX 2-bit is a ternary-weight text-to-image diffusion transformer deployment for Apple Silicon. It is built as a quality-oriented Bonsai Image variant, using ternary {−1, 0, +1} transformer weights with FP16 group-wise scaling in the matrix-heavy transformer layers, including Q/K/V projections, output projections, and MLP weights. The model reduces the FLUX.2 Klein 4B transformer from 7.75 GB FP16 to a 1.21 GB Bonsai Image transformer, a 6.4× smaller footprint, while keeping visual quality and prompt fidelity close to the original model. The Apple Silicon deployment payload is 3.88 GB, including the MLX 2-bit diffusion transformer, a 4-bit Qwen3-4B text encoder, and an FP16 Flux2 VAE. After prompt encoding, the text encoder is offloaded, so the denoising loop only keeps the compact transformer and VAE resident. The model uses a 4-step FlowMatchEuler sampler with guidance 1.0 and shift 3.0, with no CFG and no negative prompts required.
Learn more
