baidu / ERNIE-Image-Turbo
ERNIE-Image-Turbo
A high-quality text-to-image model developed by Baidu that supports English, Chinese, and Japanese prompts, with automatic prompt expansion built in.
# ERNIE-Image-Turbo
A distilled, high-speed version of ERNIE-Image.
## Highlights
- Very fast: Produces high-quality images in just 8 steps, dramatically faster than the standard model.
- Quality retained: Keeps ERNIE-Image's strengths in text rendering, instruction following, and structural consistency.
- Strong text rendering: Clear, legible typography for posters, banners, and UI-style visuals.
- Wide style coverage: Photorealistic, design-oriented, cinematic, and stylized looks.
- Runs on consumer GPUs: Accessible on 24 GB VRAM cards.
## When to Use
- When speed is critical — demos, previews, interactive tools, or bulk generation.
- When you want to generate many variations from a single prompt and pick the best.
- When you want high quality without long waits.
## How It Differs from ERNIE-Image
- ERNIE-Image: More steps, maximum quality and detail.
- ERNIE-Image-Turbo: Far fewer steps, near-equal quality, much faster results.
API quick start
Run ERNIE-Image-Turbo with a single API call.
{
"prompt": "This is a photograph depicting an urban s…",
"negativePrompt": "...",
"steps": 8,
"scale": 1.0
}