88Z AI, 15 April — Baidu Wenxin has officially open-sourced the ERNIE-Image text-to-image model. With only 8 billion parameters, it achieves state-of-the-art open-source performance, matching commercial closed-source models in text rendering and complex instruction following. The model supports multilingual glyph generation, runs on 24GB VRAM, and its resources are available on Hugging Face.
At a glance
- 8B parameters achieve open-source SOTA, with text rendering rivaling commercial models
- Runs on 24GB VRAM, supports multiple languages, and is available on ComfyUI

The update
Baidu Wenxin has officially open-sourced its text-to-image model, ERNIE-Image. With just 8 billion parameters, the model achieves state-of-the-art performance in the open-source community. In terms of key capabilities, it rivals commercial closed-source models like Nano Banana in text rendering accuracy and adherence to complex instructions. Regarding deployment and compatibility, ERNIE-Image is hardware-efficient, requiring only 24GB of VRAM to run. It supports precise glyph generation for Chinese, English, Japanese, and Korean. The model is now available on ComfyUI and includes a GGUF quantization scheme to optimize performance. Relevant model weights and inference code have been made publicly available via Hugging Face.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 百度文心Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.