Ernie-Image Text to Image

Strong text rendering: Excels at dense, long, and layout‑sensitive text for posters, UI, and infographics. Instruction fidelity: Reliably handles complex multi‑object, relational, and knowledge‑heavy prompts. Structured generation: Ideal for posters, comics, storyboards, and multi‑panel layouts. Wide style range: From realistic photography to clean design and cinematic aesthetics. Practical deployment: Runs on consumer GPUs with 24GB VRAM. Released versions ERNIE‑Image (SFT): Strong general capability; ~50 inference steps. ERNIE‑Image‑Turbo: DMD + RL optimized; faster and more aesthetic with only 8 steps.