Image generation
clavue-img is the official image product id (premium pool). Call api.clavue.com — do not ship engine names in product code. Text-to-image: z-image-turbo (1024 / 768x1344 / 1344x768, 9 steps, CFG 0, b64_json). Instruction edit: qwen-image-edit. A face to keep must use /v1/images/edits, never text-to-image. Short prompts spark-rewrite to 60-120 English words.
How to generate
- Web Chat / Imagine: pick clavue-img or ask to draw
- imux Agent Chat: official model clavue-img or imux_generate_image tool
- Images API: POST /v1/images/generations with model=clavue-img and size 1024x1024 / 768x1344 / 1344x768 (or aspect_ratio)
- Chat completions also accept model=clavue-img; reply is a Markdown image link
- Hop locks steps=9 and CFG=0. Face-preserving looks use POST /v1/images/edits. Do not hit ttkk / 192.168.2.* from product clients
curl -sS https://api.clavue.com/v1/images/generations \
-H "Authorization: Bearer $CLAVUE_TOKEN" \
-H "Content-Type: application/json" \
-d '{"model":"clavue-img","prompt":"Photorealistic half-body portrait of a 28-year-old East Asian woman, 85mm, soft window light, visible skin pores, no watermark.","size":"768x1344","response_format":"b64_json"}'Recommended settings
Send a director-style English prompt and an aspect ratio. Do not send steps, guidance, or negative lists. Rewrite Chinese one-liners with auto first if you need a longer prompt.
{
"model": "clavue-img",
"prompt": "Photorealistic half-body portrait of a young woman, 85mm, soft window light, shallow depth, no watermark",
"size": "768x1344",
"response_format": "b64_json"
}Prompt demo
One rewrite, then one image call. Do not mix song captions into the image prompt.
# Rewrite system (model: auto, thinking off)
You are an image prompt engineer. Write a 60–120 word English director description.
Order: subject → wardrobe/action → scene → lens → light → one style → quoted on-image text → positive constraints.
No masterpiece/8k. No negative list. Prompt only.
# User
半身人像，85mm，窗光，没有水印，竖构图。
# Then POST /v1/images/generations
{"model":"clavue-img","prompt":"<rewritten English>","aspect_ratio":"9:16"}Limits & safety
Image generations consume premium quota. Safety filters always apply — prohibited categories cannot be disabled. Do not attempt to strip provenance markers if present.