ERNIE-Image-Turbo

An open text-to-image generation model developed by the ERNIE-Image team at Baidu. It is based on the Diffusion Transformer (DiT) architecture and incorporates additional components to enhance text processing and the handling of structured tasks.

Key Features

• Despite its compact scale, the model delivers performance comparable to significantly larger open-weight text-to-image models.
• The model is designed on the principle that precise content realization is just as critical as visual aesthetics.
• Capable of following intricate prompts involving multiple objects and detailed interactions.
• Highly effective at generating text and structured layouts, such as posters, infographics, and UI-style imagery.
• The model can be deployed on consumer GPUs with 24GB of VRAM.
• Turbo model, optimized by DMD and RL, achieves faster speed and higher aesthetics in only 8 inference steps.

---

The model is a component of the image generation pipeline, consisting of:

• PE: ~3.8B parameters,
• Text encoder: ~3.8B parameters,
• Transformer: ~8B parameters,
• VAE: ~84M parameters, 

Total: ~15.8B parameters

Announce Date: 10.04.2026
Parameters: 7B
Developer: Baidu, Inc.
Diffusers Version: 0.36.0
vLLM-Omni Version: 0.20.0
License: Apache 2.0

Use our pre-built public endpoints for free to test inference and explore ERNIE-Image-Turbo capabilities. You can obtain an API access token on the token management page after registration and verification.

Rent your own physically dedicated instance with hourly or long-term monthly billing.

We recommend deploying private instances in the following scenarios:

• maximize endpoint performance,
• enable full context for long sequences,
• ensure top-tier security for data processing in an isolated, dedicated environment,
• use custom weights, such as fine-tuned models or LoRA adapters.

Contact our dedicated neural networks support team at nn@immers.cloud or send your request to the sales department at sale@immers.cloud.