API documentation

Foldbun Stack is a pay-per-use computer-vision model API. Segmentation, text recognition, document parsing, embeddings and background removal are available over HTTPS at https://api.stack.foldbun.com. Requests are authenticated with an API key and billed per call from a prepaid balance. Calls inside a pool’s daily allowance are free of charge, and failed calls cost nothing.

The API is synchronous: one request returns one result. Asynchronous jobs and webhooks are not available yet.

Use cases

One page per task: scenarios with figures, model choice and an example call.

Use case	Endpoint	Models
Interactive segmentation	POST /v1/segment, POST /v1/refine	sam3, sam-lite
Concept segmentation	POST /v1/segment	sam3
Text recognition (OCR)	POST /v1/ocr	ppocr-v6
Document parsing	POST /v1/documents/parse	paddleocr-vl
Embeddings and search	POST /v1/embeddings	dinov3, siglip2
Background removal	POST /v1/remove-background	birefnet

Models

Units and prices: Billing. GET /v1/models returns the live catalog for the calling account; the request is free and not metered.

Model	Model name	Use cases	Pool
sam3	SAM 3 (Meta)	Interactive segmentation, Concept segmentation	Paid
sam-lite	SAM 2.1 Hiera-S (Meta)	Interactive segmentation	Free
ppocr-v6	PP-OCRv6 small (PaddlePaddle)	Text recognition (OCR)	Public
paddleocr-vl	PaddleOCR-VL 1.6 (0.9B) with PP-DocLayoutV3 layout analysis (PaddlePaddle)	Document parsing	Paid
dinov3	DINOv3 ViT-B/16 (Meta)	Embeddings and search	Public
siglip2	SigLIP 2 base, patch 16, 224 px (Google)	Embeddings and search	Public
birefnet	BiRefNet, general model (Peng Zheng et al.)	Background removal	Paid

Pools decide which models an account can call: Rate limits and pools.

How a call works

1. Create an API key in the console. See Quickstart.
2. Send the image as base64, or upload it once and reuse the returned img_ handle. See Images.
3. Read the result. Every successful response carries the call id, the model, the image, the cost of the call, the balance after it, a timing object and a billing breakdown. See Billing; every field is described in the API reference.

All coordinates in requests and responses are pixels of the oriented original: the uploaded image at full size with its EXIF orientation applied. Polygons and boxes use the same pixels; a requested mask carries its own grid size in mask.size. See Images and Post-processing.

Where to go next

• Quickstart: first request with curl and Python.
• Use cases: one page per task, with model choice and an example call.
• Authentication: API keys, model scopes and expiry.
• Errors: the error envelope, codes and retry guidance.
• API reference: every endpoint and schema, generated from the OpenAPI document.
• Playground: try every model in the browser without an account.