Generation

Four endpoints, one request shape. Every call takes a prompt, an optional provider and model, and a parameters object whose contents depend on the modality.

Omit provider and the account default for that modality is used. Omit model and the provider's default model is used.

Text

POST /v1/generate/text — synchronous.

Field	Type	Required	Notes
prompt	string	yes	Max 10,000 characters
system_prompt	string	no	Max 5,000 characters
provider	string	no	`anthropic`, `openai`
model	string	no	See models
template_id	integer	no	Apply a saved prompt template
template_variables	object	no	Values for the template's placeholders
parameters.max_tokens	integer	no	1–4096
parameters.temperature	number	no	0.0–2.0

request

{
  "prompt": "Write a professional product description for a time-tracking app.",
  "system_prompt": "You are a senior SaaS copywriter.",
  "provider": "anthropic",
  "model": "claude-sonnet-5",
  "parameters": { "max_tokens": 800, "temperature": 0.7 }
}

Image

POST /v1/generate/image — synchronous.

Field	Type	Notes
prompt	string	Required. Max 5,000 characters
negative_prompt	string	What to avoid. Max 2,000 characters
parameters.size	string	`256x256`, `512x512`, `1024x1024`, `1024x1792`, `1792x1024`
parameters.quality	string	`standard` or `hd`. `hd` costs more
parameters.style	string	`vivid` or `natural`
parameters.n	integer	1–4. Billed per image

request

{
  "prompt": "Isometric illustration of a server rack, muted palette",
  "negative_prompt": "blurry, low quality, text",
  "provider": "openai",
  "model": "gpt-image-1",
  "parameters": { "size": "1024x1024", "quality": "hd", "n": 1 }
}

The response result is a URL to the stored image.

Video

POST /v1/generate/video — asynchronous. Returns immediately with status: "processing".

Field	Type	Notes
prompt	string	Required
parameters.duration	integer	Seconds. Billed per second
parameters.ratio	string	e.g. `16:9`, `9:16`

Poll for completion:

polling

curl https://api.genzoai.com/v1/generate/$UUID \
  -H "Authorization: Bearer $GENZO_API_KEY"

# status: pending -> processing -> completed | failed

Poll every 5–10 seconds. Video jobs typically finish in 1–4 minutes. Credits are released automatically if the job fails.

Speech

POST /v1/generate/audio — asynchronous. Text to speech.

Field	Type	Notes
prompt	string	Required. The text to speak
parameters.voice	string	Voice identifier
parameters.speed	number	Playback rate multiplier
parameters.format	string	`mp3`, `wav`

Billed per 1,000 characters of input text.

Retrieving results

Endpoint	Returns
GET /v1/generate/{'{'}uuid{'}'}	One generation with full result and metadata
GET /v1/generate/history	Paginated list. Filter with `?type=`, `?status=`, `?per_page=`

Status values

pending — accepted, queued.
processing — running upstream.
completed — result populated.
failed — error populated; credits released.

Failures do not cost credits. The estimated cost is held when a request starts and released in full if the upstream provider errors. Only completed work is billed, at the real token count.