ERNIE Image: concise review

ERNIE Image is Baidu’s open-weight text-to-image model, presented on the project site under an Apache 2.0 license. The site positions it around text-accurate generation, structured layouts, and complex multi-object prompts, and it offers a browser-based image generator for testing.

What it does

ERNIE Image turns short text prompts into images, with a stated emphasis on readable text inside images and on maintaining layout structure. The project materials describe it as an 8B diffusion transformer and note support for bilingual prompts. A web generator is available on the site.

Key features

• Text-accurate image generation with a focus on legible in-image text.
• Open-weight 8B diffusion transformer architecture.
• Structured layouts and complex multi-object prompts.
• Bilingual prompt support.
• Adjustable output resolution from 300 to 2048 px.
• Credit-based generation with Turbo or Standard modes and an optional Prompt Enhancer.
• Browser-based image generator.

Pricing and usage

ERNIE Image uses prepaid credit packs rather than subscriptions. The pricing page says commercial usage is included in all packs and describes the model as “simple pricing” with “no surprises.” Users can choose Turbo or Standard generation and set custom width and height within the documented range.

A refund policy is published. Refund eligibility for unused services depends on the time since purchase, and the policy summary indicates refunds can be requested within 7 days of purchase for eligible unused services.

Who this is for / not for

Who this is for:
- Creators and designers working on text-heavy visuals such as posters, labels, or UI mockups.
- Users who want structured image layouts and bilingual prompt support.
- People who prefer a web-first, pay-as-you-go credit model.

Who this is not for:
- Users who need output sizes beyond the documented 300–2048 px range.
- Teams that need detailed API documentation, self-hosting instructions, or training-data provenance before evaluating the tool.
- Buyers who need confirmed volume or enterprise pricing details, which are not shown on the public pages reviewed.

Risks and limitations

• Output size is bounded by the documented 300–2048 px range.
• The prepaid credit model means you should check refund terms before purchasing.
• The public pages emphasize text rendering and structured layouts, so those appear to be the model’s main strengths.
• Public pages do not fully document API availability, self-hosting, exact per-pack credit pricing, or training-data details.

Bottom line

Based on the project site, ERNIE Image is a focused text-to-image product for users who care about readable text, structured layouts, and bilingual prompts. Its main practical advantages are the open-weight release, browser access, and credit-based pricing with commercial use included. Before using it for a production workflow, verify the license files, refund terms, and any documentation you need for API or self-hosting use.