hidream-ai/hidream-o1-image

HiDream-O1-Image is an open-source AI model for generating high-quality images from text descriptions. What makes it unusual is its architecture: instead of stitching together separate components for understanding text and generating pixels (a common approach in image AI), it uses a single unified model called a Pixel-level Unified Transformer that handles everything in one pass, raw pixels, text, and any additional conditions all share the same processing space. The result is a more coherent system that supports multiple creative tasks without mode-switching: you can generate images from a text prompt, edit existing images with written instructions, or create personalized images that preserve a specific subject's appearance across different scenes. The model runs at 8 billion parameters, which is compact enough to run on consumer hardware, yet its benchmark scores place it competitively against much larger systems and closed commercial models. It can produce images at resolutions up to 2,048 by 2,048 pixels and includes a built-in "reasoning" agent that interprets complex prompts before generating, useful for scenes requiring accurate text rendering, multi-region layouts, or multilingual captions within the image. Two variants are available on Hugging Face: the full model (50 inference steps) and a faster Dev version (28 steps). A web demo is also accessible on Hugging Face Spaces for trying it without any installation.