# JANCTION Render Product Video

> Turns a product 3D model (.glb, .fbx, .obj, .usdz) or one product photo (.png, .jpg) into a finished product video, a 360 turntable or four product
> images, rendered on cloud GPUs (JANCTION Render). Priced in JPY before ordering; changes re-render only what changed.

## Product Video or JANCTION Render?
- Product Video: finished media to post as is (captions, end card, backdrop, 360 loop, shop images, listing pack, edits
  without re-rendering, saved projects). GPU at cost plus a production fee
- JANCTION Render: raw renders to finish yourself (your own .blend or bpy scene, transparent PNG product shots, a plain
  turntable, EXR, many frames). GPU at cost only: https://render.janction.jp/v1/outcomes

## Use it when
- someone has a 3D model of a product and wants a short product video, a 360 turntable loop or product images
- someone has only a product photo and wants a short captioned video (template photo, 5 or 15 s)
- the same product needs new captions, colours, camera angle or lighting later (projects keep every version)
- many SKUs need the same look (batch orders from a CSV, saved brands)

## Do not use it for
- videos from text alone with no photo or 3D model (try with a sample model instead: sample:water-bottle)
- people, acting, live action or photoreal generated scenes
- editing existing video footage

## Prices (JPY): GPU at cost + a production fee
- GPU: 0.1 JPY per GPU-second, the same rate as JANCTION Render. The quote shows a maximum; on delivery you pay the fee
  plus the GPU seconds actually used, up to the cap, and the rest is returned. Nothing is charged if it fails
- Video (hero, feature): 5 s: fee 40 JPY, about 62 JPY in all, 15 s: fee 90 JPY, about 142 JPY in all, 30 s: fee 160 JPY, about 299 JPY in all
- 360 turntable 5 s: fee 30 JPY, about 42 JPY in all
- Four product images: fee 20 JPY, about 22 JPY in all
- Store listing pack (the 4 images plus a 5 s 360 turntable, one approval): about 64 JPY in all
- From one photo (CPU, no GPU): 5 s 20 JPY, 15 s 50 JPY. the photo is zoomed and moved with captions on CPU; the product does not turn in 3D
- Uploads: up to 200 MB per file; larger scenes go to JANCTION Render
- Kept: free trials 7 days, paid 180 days
- Shots rendered before (or shared sample shots) cost no GPU
- Glass and heavy models use more GPU seconds (the cap in the quote covers it)
- Changes: captions, title, backdrop colour, logo, audio, order: free a number of times a day; camera, lighting,
  length: only the re-rendered shots, never more than a new one
- Sample models: free trials per account

## Measured (gpu004, 2026-10-09/10 and 10-13)
- 5 s video: about 2 minutes from order to delivery when the GPUs are free
- 15 s video: 4 min 35 s to 5 min 20 s from order to delivery
- caption and colour change: 10 s, 0 JPY
- camera change of one 5 s shot: 2 min 21 s, 176 GPU-seconds (about 28 JPY with the fee)

## Flow (MCP: https://render.janction.jp/video/mcp, Streamable HTTP, Authorization: Bearer <JANCTION Render API key>)
1. plan_video: offers, prices, sample models, budget
2. quote_video: price, minutes, deliverables, quote_id (30 minutes)
3. show the price to the user and get a yes
4. submit_video(quote_id, approved_price_yen)
5. get_video_job until done, then get_video_outputs (links open without a key for 24 hours)
6. review_video (does the user accept it?), revise_video for changes
Before ordering, storyboard_video(quote_id) draws one small image per shot (free, 10 a day).
For shop and asset-store pages: quote_listing_pack -> submit_listing_pack -> get_listing_pack (images and a turntable in one order)
Limits per order and per day are enforced by the server. A paid order from an AI assistant always needs the user to
press Approve on approval_url first; free trials do not.
Tested AI clients: claude_code: verified 2026-10-15 (MCP, order to delivery), chatgpt: not verified, codex: not verified, cursor: not verified. Machine-readable facts: https://render.janction.jp/video/v1/spec

## 3D workspace (no model yet, or the scene needs changes)
- Build a simple scene (product stage, can, room, simple character) or place your own model, then change it with typed
  operations (add, move, colour, light, camera, keyframes); no Python. Every change is a new version; going back makes
  another version, nothing is overwritten. Two agents on one scene get a conflict instead of a silent overwrite
- MCP: project_create, scene_inspect, scene_apply, revision_restore, scene_revisions, scene_preview (one GPU image,
  free 20 a day), asset_search and asset_import (CC0 models from Poly Haven), project_export (glb, obj, stl, blend),
  project_share (a view-only link to one version), scene_to_product (then quote_video)
- Pictures: upload a PNG/JPEG (POST /v1/files?role=logo) and put it on an object with the set_image operation (a can
  label, a box face, a standing photo panel). This is photo-based, not a 3D model rebuilt from a photo
- Browser: https://render.janction.jp/video/workspace
- Not offered: 3D models generated from text or photos (a photo makes a 2D video, see Prices)

## HTTP
- https://render.janction.jp/video/v1/catalog
- https://render.janction.jp/video/v1/tools (what each tool changes, what it can cost, how to undo it)
- https://render.janction.jp/video/v1/videos/quote, https://render.janction.jp/video/v1/videos, https://render.janction.jp/video/v1/videos/{id}, https://render.janction.jp/video/v1/videos/{id}/revise
- https://render.janction.jp/video/v1/projects

## Related
- JANCTION Render (Blender rendering for AI agents): https://render.janction.jp/