Vision

Perceptron provides a video understanding API. Vision models accept video (and optionally image) input and return text output — descriptions, analysis, or Q&A. You pay using credits purchased with USDT.

Base URL

https://perceptron.cloud/api/v1

Authentication

Include your API key in the Authorization header:

Authorization: Bearer YOUR_API_KEY

Generate an API key from the Keys page in the Perceptron app.

Available Models

curl https://perceptron.cloud/api/v1/vision/models

Vision models (the video input modality, optionally including image) are listed separately from chat models. Query GET /api/v1/vision/models for the current list and pricing. The reference model is MiniMax M3.

Analyze

POST /api/v1/vision

Analyze a video or image and return text output (description, analysis, or Q&A). OpenAI-compatible multipart/form-data upload. Supported video formats: mp4, webm, mov, mkv.

Request

curl https://perceptron.cloud/api/v1/vision \
  -H "Authorization: Bearer $PERCEPTRON_API_KEY" \
  -F "model=minimax/minimax-m3" \
  -F "prompt=Describe what is happening in this video." \
  -F "file=@clip.mp4"

Send only images (no video):

curl https://perceptron.cloud/api/v1/vision \
  -H "Authorization: Bearer $PERCEPTRON_API_KEY" \
  -F "prompt=Compare these two screenshots." \
  -F "image_file=@screenshot1.png" \
  -F "image_file=@screenshot2.png"

Part	Type	Required	Description
`file`	file	yes*	The video or image file (mp4, webm, mov, mkv, png, jpg, gif, webp)
`image_file`	file	no	Additional image (repeatable for multiple images)
`prompt`	string	yes	Text prompt (description, question, or instruction)
`model`	string	no	Vision model ID from `GET /api/v1/vision/models`
`artifact_id`	string	no	Artifact reference (existing artifact id, alternative to `file`)
`image_artifact_id`	string	no	Image artifact reference (repeatable for multiple; alternative to `image_file`)
`max_tokens`	uint64	no	Maximum output tokens

* At least one media input is required: file, image_file, artifact_id, or image_artifact_id. You can send only images (no video) by omitting file and using image_file parts.

Response

{
  "text": "A dog runs across a sunny beach, chasing a thrown ball.",
  "usage": {
    "prompt_tokens": 24,
    "completion_tokens": 16,
    "total_tokens": 40
  }
}

See the full reference at API Reference.

Pricing

Vision analysis is billed from your credits balance. See Billing for more details.