ERNIE-Image
Open text-to-image generation model from Baidu
ERNIE-Image is Baidu's open text-to-image generation model release, built on a single-stream Diffusion Transformer with 8B DiT parameters in a latent diffusion framework and paired with a lightweight Prompt Enhancer. The standard ERNIE-Image release is described as the SFT model focused on stronger general-purpose quality and instruction fidelity, typically using 50 inference steps.
Pricing
Listed pricing interpreted as RMB 0.04 per generated image (元/张) for ERNIE-Image; verify live Baidu page for any current free quotas, regional billing rules, or discounts.
Baidu AI Cloud/Qianfan first-party pricing documentation is the relevant vendor source for ERNIE-Image. Exa access returned credit-limit errors during this run, so the finding is recorded with low confidence despite using the official pricing-document URL.