ERNIE-Image is an open-weight text-to-image generation model developed by the ERNIE-Image team at Baidu. Built on a single-stream Diffusion Transformer (DiT) architecture, the model is designed to provide high-quality visual outputs with a focus on semantic controllability and precise text rendering. It operates within a latent diffusion (LDM) framework and is optimized for bilingual generation in both Chinese and English.
A significant feature of ERNIE-Image is its proficiency in layout-critical tasks, such as generating posters, comic storyboards, and multi-panel compositions. Unlike many general-purpose generative models, it demonstrates strong performance in rendering readable text and maintaining structured image layouts. To assist users, the model is often paired with a lightweight Prompt Enhancer, which expands brief user inputs into richer, more structured descriptions to better utilize the model's capabilities.
The model's architecture consists of 8 billion parameters, allowing it to run on consumer-grade hardware while remaining competitive with significantly larger models. Baidu also released ERNIE-Image-Turbo, a distilled version of the base model that is capable of producing high-quality results in as few as 8 sampling steps. This variant is specifically optimized for high-throughput workflows and rapid content creation without a substantial loss in visual fidelity.