ERNIE-Image
ERNIE-Image combines an 8B diffusion transformer with a lightweight prompt enhancement component. This is the standard supervised fine-tuned release, intended to preserve image quality across general generation tasks.
Model design and access
The card reports support for posters, comics, detailed prompts and text within images. It describes operation on a consumer GPU with 24 GB of memory. That example depends on the supplied inference configuration; it is not a fixed memory guarantee for every image size.ERNIE-Image official model card
Practical use
Practical guidance. Choose this version when output quality matters more than the number of sampling steps. Inspect small text, repeated symbols and layout alignment in the final image. A prompt enhancer can change the effective instruction, so retain its expanded prompt when diagnosing an unexpected result.ERNIE-Image official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| License | Apache 2.0ERNIE-Image official model card |
| Architecture | Single-stream diffusion transformerERNIE-Image official model card |
| Reference sampling | 50 stepsERNIE-Image official model card |
Related models
- ERNIE-Image-Turbo
Baidu ยท ERNIE-Image
Distilled eight-billion-parameter image model that generates images in eight sampling steps.
Launch date not confirmedDownloadable