ERNIE Image
Browser access to Baidu's open-weight 8B image model
About ERNIE Image
ERNIE Image is a browser front end for Baidu's open-weight image model, an eight billion parameter diffusion transformer. The reason it is worth knowing about separately from the many image generators available is what this particular model is good at: text rendering. Most diffusion models produce beautiful images and then fall apart the moment a word needs to appear inside them, generating letterforms that look like writing without being writing. ERNIE Image is specifically positioned around structured, text-rich output such as posters, infographics, comic panels and bilingual labelling, which is precisely where the general-purpose alternatives are weakest.
This is a third-party web interface for an open-weight model rather than a Baidu-operated property, which is worth understanding: the weights are openly published, and this site provides hosted access without requiring you to run them yourself. That removes the GPU requirement and the setup, at the cost of a per-generation credit. The studio exposes real controls rather than a bare prompt box, with six aspect ratios from 16:9 through 2:3 plus explicit width and height, and output at defined pixel dimensions. Credits are purchased in one-time packs that never expire, with the per-credit cost falling on larger packs, and a seven day money-back guarantee. Generation is free to start.
Describe what you want in the prompt field, then set the shape. Choose from the six preset ratios covering 16:9 at 1280 by 720, 9:16 at 720 by 1280, 4:3, 3:4, 3:2 and 2:3, or set width and height directly for a size the presets do not cover. Generate, and the result appears in the studio ready to download; the site notes this is the same flow used inside the product itself, so the homepage demo is not a simplified preview. Because the model handles embedded text well, prompts describing posters, labelled diagrams or panels with dialogue are worth attempting rather than avoiding. Credits deduct per generation, purchased in advance in one-time packs.
- •Strong Text Rendering - Built around structured, text-rich output including posters, infographics, comic panels and bilingual labels, the weak point of most diffusion models
- •8B Diffusion Transformer - Baidu's open-weight ERNIE architecture, published openly rather than kept proprietary
- •No Installation or GPU - Runs in the browser, removing the hardware and setup burden of self-hosting an open-weight model
- •Six Aspect Ratios Plus Custom - 16:9, 9:16, 4:3, 3:4, 3:2 and 2:3 presets at defined pixel dimensions, or set width and height directly
- •One-Time Credit Packs - Buy credits outright rather than subscribing, with the per-credit price dropping on larger packs
- •Credits Never Expire - Purchased credits remain usable indefinitely, suiting irregular usage
- •Photorealistic and Stylised Range - Covers portraits, film stills, manga and product sets alongside the text-heavy cases
- •Seven Day Refund - Money-back guarantee on credit purchases, with payment handled through Stripe
Best for anyone generating visuals that must contain readable text: poster and flyer designers, people building infographics and labelled diagrams, and comic or storyboard creators who need dialogue rendered in panel. It suits bilingual and Chinese-language work particularly well given the model's origins. Marketers producing product sets and social graphics get general-purpose capability alongside the text strength. Developers curious about the open-weight ERNIE model can evaluate its output here before committing to self-hosting it.
Pricing
One time purchase, not a subscription
Free to start - one-time credit packs: Basic $9.90 (150 credits), Major $29.90 (540), Profession $49.90 (1,100) - credits never expire - 7-day refund
From the vendor pricing page, 2026-09-16
