Image and Video Models

Image and video models on the platform bill two ways: per token (multimodal chat models that draw) and per call (most dedicated image/video models, marked "$/call" on the pricing page). Both are called through the OpenAI-compatible endpoint.

Image generation (images/generations)#

Endpoint: POST https://api.zaimod.com/v1/images/generations

resp = client.images.generate(
    model="gpt-image-2",
    prompt="Cyberpunk city at night, neon rain, cinematic composition",
    size="1024x1024",
    n=1,
)
print(resp.data[0].url)        # or b64_json

Main image models on the platform (per-call reference pricing):

In-chat image models (gpt-4o-image family)#

Some models live on the chat endpoint — iterate on an image through conversation:

resp = client.chat.completions.create(
    model="gpt-4o-image",
    messages=[{"role": "user", "content": "Draw a ginger cat wearing glasses, flat illustration"}],
)

Billed per token or per call (see the pricing page). Ideal for generate → tweak → regenerate loops.

Video generation (async tasks)#

Video models are asynchronous: submit a task, then poll.

1. Submit

curl https://api.zaimod.com/v1/videos/generations \
  -H "Authorization: Bearer $ZAIMOD_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2-0-260128",
    "prompt": "A corgi running on grass, tracking shot, slow motion",
    "duration": 5
  }'

Returns a task_id.

2. Poll

curl https://api.zaimod.com/v1/videos/generations/{task_id} \
  -H "Authorization: Bearer $ZAIMOD_API_KEY"

Status flows submitted → running → succeeded/failed; success returns a video URL.

Main video models:

General notes#

• Per-call models don't charge on failure, but duplicate submissions do bill — make task submission idempotent
• When size/duration affects price, it's marked on the pricing page or in the console
• For batch jobs, trial 10–20 items first and check the success rate before scaling up