Ernie Image is Baidu's text-to-image model — multilingual prompts (English, Chinese, Japanese), strong CJK text rendering, and two model variants (HD at 1 credit, UHD at 3 credits, or Turbo at 1 credit). Particularly strong on East Asian subjects and CJK signage.
Ernie Image is Baidu's text-to-image model. Its defining traits: multilingual prompt support (you can write prompts in English, Chinese, or Japanese), native CJK text rendering (Chinese, Japanese, Korean characters render with proper letterforms), and strong photoreal output on East Asian subjects, architecture, and signage.
Two variants via the model field. ernie-image is the quality tier — 1 credit at HD (square / landscape / portrait HD) or 3 credits at Square UHD. ernie-image/turbo is the fast tier — 1 credit flat. Both support negative prompts and seed pinning for reproducibility.
Multi
EN / ZH / JA prompts
CJK
Native text rendering
1–3 cr
Per image
Write prompts in English, Chinese, or Japanese — Ernie reads all three natively. Useful for non-English content workflows, and for prompts where the source-language nuance matters (cultural specificity, literary references, place names).
Ernie renders CJK characters with proper letterforms, kerning, and weight — one of the strongest models on this dimension. Wrap exact strings in double quotes. Restaurant signs, calligraphy, packaging, posters all read clean and culturally accurate.
Particularly strong on East Asian subjects — traditional Chinese architecture, Japanese gardens, Korean cuisine, Asian fashion. Cultural detail (specific costume styles, regional architectural features, ceremonial elements) renders with respect and accuracy.
ernie-image HD = 1 credit (square / landscape / portrait HD). ernie-image Square UHD = 3 credits (higher resolution). ernie-image/turbo = 1 credit flat (fast tier, smaller variants). Pick by use case.
Each example shows the exact prompt that produced the result. Copy any prompt with one click.
HD portrait · 1 credit
Editorial portrait of a Chinese tea master in traditional silk hanfu, sitting at a wooden tea table with fine porcelain, soft window light from the left, hyper-detailed skin and fabric texture, photoreal
"Traditional silk hanfu" + cultural detail anchors. Ernie renders hanfu fabric weave, porcelain tea sets, and traditional wooden surfaces convincingly. Soft directional lighting + 3:4 portrait gives editorial framing without needing technical photo anchors.
HD landscape · 1 credit
Cinematic wide shot of Shanghai's Pudong skyline at sunset, neon lights beginning to illuminate against a deep orange sky, the Bund visible in the foreground, painterly photoreal
Place-specific prompts work — name the city + landmark. Ernie composes Pudong's distinctive skyline cleanly. Twilight transition ("neon beginning to illuminate against deep orange sky") gives the right time-of-day lighting layer.
HD square · 1 credit
A handpainted Chinese restaurant sign reading "上海菜館" in elegant calligraphy on a deep red wooden panel, brass mounting, golden afternoon light, photoreal
Quote the exact CJK characters in double quotes — Ernie holds the letterforms cleanly. "Elegant calligraphy" + "deep red wooden panel" + "brass mounting" gives surface and typography direction. One of Ernie's standout strengths.
HD landscape · 1 credit
Traditional Chinese garden pavilion with curved tile roof reflecting in a still pond at dawn, lotus flowers floating on the water, soft mist rising, painterly atmospheric
"Traditional Chinese garden pavilion" + "curved tile roof" — specific architectural detail. Ernie composes the reflection in the pond cleanly with the lotus floating elements layered. "Painterly atmospheric" closes the prompt with aesthetic anchor.
Prompts in Chinese or Japanese sometimes capture cultural nuance that English translation flattens. For specifically East Asian subjects (architecture, calligraphy, ceremony), writing the prompt in the target language can produce more authentic output.
Same recipe as Latin text — wrap exact CJK strings in double quotes. Ernie renders the characters with proper letterforms and weight. "上海菜館", "東京", "한국" all hold up at HD tier when prompted this way.
Square HD, Landscape HD, Portrait HD all cost 1 credit. Use these for everyday production work. Reach for Square UHD (3 credits) only for hero shots where the extra resolution is necessary. Turbo is the speed-optimized 1-credit fast tier.
Cultural specificity unlocks Ernie's strengths. Hanfu, kimono, hanbok, traditional architecture, regional cuisine, calligraphy — all read with cultural accuracy. Vague "Asian-style" produces generic output; specific cultural detail produces convincing output.
"Painterly atmospheric", "soft mist", "golden afternoon light" work better than camera-stack anchors. Ernie reads aesthetic vocabulary literally and renders the atmospheric mood convincingly.
square_hd for 1:1, landscape_hd for 16:9, portrait_hd for 9:16. Square_uhd for hero shots at higher resolution. The image_size field is enum, not a free aspect string — match exactly.
| Setting | Values | Notes |
|---|---|---|
| Model | fal-ai/ernie-image · fal-ai/ernie-image/turbo | Base for quality, Turbo for speed. Both 1 credit at HD. |
| Image size | square_hd · landscape_hd · portrait_hd · square_uhd | Enum, not free aspect. Choose explicitly. |
| Negative prompt | String (optional) | Optional. Short focused negatives work best. |
| Seed | Integer (optional) | Pin for reproducibility across re-runs. |
| Pricing — HD tiers | 1 credit | Square HD, Landscape HD, Portrait HD. |
| Pricing — UHD | 3 credits | Square UHD only. Base model. |
| Pricing — Turbo | 1 credit flat | fal-ai/ernie-image/turbo. Speed-optimized fast tier. |
Both render CJK characters cleanly with proper letterforms. Ernie has slightly stronger Chinese-cultural composition behaviors (traditional architecture, hanfu, regional cuisine read with more authenticity). Qwen Image has broader aspect ratio coverage and the Plus/Max snapshot system. For specifically Baidu/Chinese-cultural subjects, Ernie. For broader work with occasional CJK, Qwen.