UniPic is a series of multimodal image models open-sourced by SkyworkAI on GitHub. The repository hosts model weights and official implementations, positioned as a unified multimodal model for image editing, generation, and understanding.
In terms of use, UniPic covers three core task categories: text-to-image generation (high-fidelity synthesis based on natural language prompts), image editing (seamless inpainting, outpainting, and object manipulation), and image understanding (perception capabilities for various visual tasks).
The main capabilities are divided by three modeling paradigms:
UniPic-3: Described as an open-source SOTA multi-image editing model, it provides a unified framework for single-image editing and multi-image composition, supports 1 to 6 input images, and supports flexible resolutions; it achieves 8-step inference and approximately 12.5x acceleration through CM + DMD distillation.
UniPic-2: Includes SD3.5M-Kontext and MetaQuery variants, based on an efficient architecture and diffusion post-training, providing leading performance in text-to-image generation, fine-grained image editing, and multimodal reasoning.
UniPic-1: 1.5B parameters, employing unified autoregressive modeling, jointly combining visual understanding and generation, enabling a single Transformer to simultaneously handle perception and synthesis tasks.
In terms of usage, the repository provides READMEs and official implementations for each version, and users can obtain model weights and code through the GitHub repository; specific installation commands, dependency configurations, and inference steps are not currently provided in the available materials, and the corresponding README or repository files should be consulted.
Applicable scenarios include: image editing workflows requiring multi-image input and composition, text-to-image and fine-grained editing tasks, and unified model research or applications requiring simultaneous visual understanding and generation.
In terms of limitations, the available materials do not provide information such as fees, pricing, platform support scope, specific performance metrics, or hardware requirements; the project uses the MIT license, see the LICENSE file in the repository for details.
