MOSS-Image
MOSS-Image is an open-weight text-to-image model family for generating images from natural-language prompts and structured layout instructions. Its Base and fine-tuned SFT checkpoints share an inference interface and support prompts of up to 4,096 tokens. Users can describe a scene in text or provide JSON layout data to guide where objects and rendered text appear in the composition. The project includes Python and Gradio workflows, a Diffusers pipeline, model weights, and guidance for local deployment and fine-tuning. It can be used to create illustrations, photorealistic images, posters, and other designed compositions, including images containing specified text. Developers can also use supported image-generation runtimes and a hosted API option. The software is aimed at developers, researchers, and creative practitioners who want to generate or adapt images with layout control.
Learn more
FLUX.2 [klein]
FLUX.2 [klein] is the fastest member of the FLUX.2 family of AI image models, designed to unify text-to-image generation, image editing, and multi-reference composition into a single compact architecture that delivers state-of-the-art visual quality at sub-second inference times on modern GPUs, making it suitable for real-time and latency-critical applications. It supports both generation from prompts and editing existing images with references, combining high diversity and photorealistic outputs with extremely low latency so users can iterate quickly in interactive workflows; distilled versions can produce or edit images in under 0.5 seconds on capable hardware, and even compact 4 B variants run on consumer GPUs with about 8–13 GB of VRAM. The FLUX.2 [klein] family comes in different variants, including distilled and base versions at 9 B and 4 B parameter scales, giving developers options for local deployment, fine-tuning, research, and production integration.
Learn more
Bonsai 27B
Bonsai 27B is the new multimodal flagship of the Bonsai family and the first 27B-class model to run on a phone. Based on Qwen3.6 27B, it brings a new capability tier to local devices; multi-step reasoning, structured tool calls, vision tasks, and computer-use agentic loops that stay coherent across many steps. Bonsai 27B comes in two variants. Ternary Bonsai 27B uses ternary weights with FP16 group-wise scaling, giving 1.71 effective bits per weight and a 5.9 GB footprint for the quality-oriented laptop-class version. 1-bit Bonsai 27B uses binary weights with the same group-wise scaling, giving 1.125 effective bits per weight and a 3.9 GB footprint that fits within the memory budget of an iPhone 17 Pro. Both variants run end-to-end across the language network, embeddings, attention, MLPs, and LM head with no higher-precision escape hatches. They are multimodal, with a compact 4-bit vision tower, so on-device workflows can understand screenshots, documents, and camera input.
Learn more
FLUX 3
FLUX 3 is a multimodal foundation model that jointly learns from images, video, and audio within one unified architecture, building a representation of how objects hold together, how things move, and how events sound. Built on the Self-Flow approach, it aligns multimodal generation and understanding in the same backbone so each modality constrains the others, sound matches impact, motion follows physical properties, and future events follow from the past. FLUX 3 can mix modalities and jointly generate images, video, and native audio from text prompts or references such as images, video, and audio. Its video capabilities include text-to-video, image-to-video animation, video-to-video transformation, generative video-and-audio continuation, keyframe-controlled transitions, multilingual dialogue, animated typography, diverse styles and aspect ratios, and agentic chaining into longer multi-shot sequences.
Learn more
