baidu/ERNIE-Image

ERNIE-Image is an open text-to-image generation model developed by the ERNIE-Image team at Baidu. It is built on a single-stream Diffusion Transformer (DiT), with only 8B DiT parameters, it reaches state-of-the-art performance among open-weight text-to-image models.

What it solves

ERNIE-Image addresses the challenge of generating high-quality images from text prompts, specifically focusing on improving text rendering within images, instruction following for complex scenes, and the ability to create structured visual content like posters and infographics.

How it works

The model is built on a single-stream Diffusion Transformer (DiT) with 8 billion parameters. It is paired with a lightweight Prompt Enhancer (PE) that expands simple user inputs into detailed, structured descriptions to improve generation quality. Two versions are available: a general-purpose SFT model and a "Turbo" version optimized via DMD and RL for faster inference (8 steps vs 50 steps).

Who it’s for

It is designed for creators and developers who need a text-to-image model capable of handling dense text, complex layouts, and diverse styles (from realistic photography to stylized art) that can run on consumer GPUs with 24GB VRAM.

Highlights

  • Superior Text Rendering: High proficiency in generating long-form, layout-sensitive text for posters and UI designs.
  • Complex Instruction Following: Reliable handling of prompts with multiple objects and detailed relationships.
  • Structured Generation: Effective at creating multi-panel compositions, comics, and storyboards.
  • Efficient Scale: Competitive performance against larger models despite its compact 8B parameter size.
  • Flexible Deployment: Supports integration via Diffusers, SGLang, and ComfyUI.