# Pinwing

> Pinwing (pinwing.ai) turns 3D models — GLB/glTF files or zipped OBJ scans — into shareable turntable videos and multi-view image datasets, rendered on GPU nodes. Browser studio, REST API, and MCP server. Credit-based: Standard 540p video costs 1 credit per second; new accounts get 100 free credits.

Entity note: Pinwing is the product brand; DXGL OÜ (Estonia) is the operating legal entity — new API keys use the pinwing_sk_ prefix, and legacy dxgl_sk_ keys plus dx.gl share links belong to the same company. Pinwing is not a 3D viewer: the output is a rendered video file plus a shareable video page. The app lives at https://dx.gl/studio; API keys (prefix pinwing_sk_) are created there and work across both the REST API and the MCP server.

## Studio

The browser studio edits the scene live on canvas — what you see is what renders:

- Scene: solid, multi-stop gradient, or image backdrops; HDRI environments (upload .hdr/.exr up to 8K) with grounded, floating, and lighting-only modes plus rotation, exposure, blur, and tint; a lighting rig with fill, spotlight, ambient, and gloss channels plus light colors; ground modes none / reflection / flat / contact / soft accumulated shadows.
- Editing: per-axis rotate (45° snap), scale (negative mirrors), move; decal stickers placed on the model surface; overlay frames (transparent PNG/WebP, left/center/right, opacity) baked over the render; effects: ambient occlusion, bloom, lens blur, vignette, cloth sway; nine color filters.
- Motion: turntable spins with easing, pan, tilt, and zoom, or keyframe camera paths (up to 64 keyframes, per-segment easing); shots run up to 120 s (60 on ProRes); GLB animation clips with direction and cycle controls; orbit and fly navigation with touch joysticks on tablets.
- Output: 16:9 / 1:1 / 9:16 at 540p, 1080p, 4K, or ProRes 4444 with alpha; 60 fps; every render is 2× supersampled, then downscaled with contrast-adaptive sharpening. Final renders load the original uploaded file — full geometry, full-resolution textures; the fast browser preview is a compressed twin used only for editing. Renders run on GPU nodes: nothing installs, and the user's device stays free while the job encodes.
- Assistant: an in-studio AI assistant drives the same controls through ~90 tools (scene readback, materials, motion, output, render submit) and always confirms credit costs before spending.
- Accounts: team workspaces with owner/manager/editor/viewer roles, workspace-bound API keys, TOTP 2FA with recovery codes. The model library works as a lightweight DAM: search, tags, grid/list views, instant-loading previews, public share links.

## Docs

- [API reference](https://pinwing.ai/docs/api/): every /v1 endpoint, render setting, credit rate, and error code
- [MCP server](https://pinwing.ai/docs/mcp/): hosted endpoint https://mcp.pinwing.ai/ (streamable HTTP, bearer auth; mcp.dx.gl also works), setup and tool reference
- [Full docs as raw markdown](https://pinwing.ai/llms-full.txt): the API and MCP references verbatim, in one file
- [OpenAPI 3.1 spec](https://pinwing.ai/openapi.json): machine-readable description of every /v1 endpoint
- [API catalog](https://pinwing.ai/.well-known/api-catalog) (RFC 9727)
- [MCP server card](https://pinwing.ai/.well-known/mcp/server-card.json)

## Site

- [Datasets](https://pinwing.ai/datasets/): render a model into multi-view training data — RGB, depth, normals, and masks per view, poses in instant-ngp + nerfstudio formats, seed point cloud; flat per-set pricing (100 views/40 cr, 196/160, 400/640)
- [ProRes 4444](https://pinwing.ai/prores/): the top video tier — 4K ProRes 4444 .mov with a real alpha channel for compositing (quality "pro", 64 cr/sec, 60s cap)
- [Features](https://pinwing.ai/features/): the full capability sheet with limits and numbers
- [Pricing](https://pinwing.ai/pricing/): credit packs and per-second video rates
- [FAQ](https://pinwing.ai/faq/): formats, limits, rendering, billing