Keep what you render
Finished renders and their scenes stay, with no expiry while you are subscribed.
No high-end GPU? Turn your 3D into images and video.
A Blender file or a scene an AI wrote: preview, fix and finish it on cloud GPUs. No sign-up, free to start.
JANCTION Render is a cloud GPU render farm for Blender that AI agents call as an MCP server or HTTP API. Send a .blend file, a bpy script or a 3D file (glTF / FBX / USD), preview in seconds, and get frames or an MP4 back from JANCTION's NVIDIA GPUs. Every preview also returns a verdict on what is wrong, such as too dark, blown out, no light or an object out of frame, and the bpy code to fix it, so the agent can correct the scene before the final render. No local Blender or GPU is needed. AI agents connect at https://render.janction.jp/mcp.
Your 3D model or Blender file, rendered on a cloud GPU. The first preview needs no payment.
No sign-upFirst 100 JPY with no cardNo GPU or Blender needed.blend .glb .gltf .fbx .usd .usdz .obj .stl .spz …, bpyPreview in secondsThen 0.1 JPY per GPU-second, priced before it runsUse it from your AI (Claude, ChatGPT, Codex)
A small Blender scene from this service renders in a few seconds; then try your own file below.
or
No connector in your AI? Ask it: "Write a Blender 5.0 bpy script that builds <your scene> with a camera and a light; do not call render. Reply with the code only." Then paste the code here.
A 100 JPY welcome credit for a new key (14 days, no card; one per network every 30 days, further keys on the same network share what is left), and 400 JPY more for saving a card that is not charged (once per card); after that, previews stay free up to 2 GPU-minutes a day and finals cost 0.1 JPY per GPU-second.
From an AI app: connect https://render.janction.jp/mcp and say: “Using JANCTION Render, build a glossy red cube rotating on a gray floor and show me a 4-frame preview. If it looks right, render 24 frames at 1280x720 as an MP4.”
Start from your problem: what is really free · render without a GPU · cloud rendering, step by step
Drop a .blend, a GLB / FBX / USD or a bpy script into the browser. No sign-up; the first preview needs no payment.
Try your own file for free ↑Add the MCP URL to ChatGPT, Claude, Codex or Cursor. Your AI writes the scene and renders it here.
How to connect ↓No file yet? A small scene from this service renders in a few seconds; then try your own.
Render the sampleStorage (beta) · Phone scans · draft video · the whole turn in one sheet
Finished renders and their scenes stay, with no expiry while you are subscribed.

Send what your scan app exports: Blender 5.3 renders it, from a preview to a 360° video.

Before rendering every frame, up to 24 small frames played over the real length, with the estimate for all of them.

Before a 360° video, 24 frames every 15° in one sheet, back side included.
Live usage of render.janction.jp · external use only
Updated · source /v1/stats, refreshed every 5 minutes
| Renders | From AI apps | Success | GPU-min | Median round trip | |
|---|---|---|---|---|---|
| 24 hours | 182 | 98 | 100% | 164 | 7 s |
| 7 days | 1,513 | 737 | 100% | 1,526 | 10 s |
| 30 days | 3,014 | 1,035 | 99% | 3,824 | 10 s |
Completed renders grouped by the AI app that submitted them over MCP. The plain HTTP API and the CLI are counted in the totals but not here. The operator's own tests, benchmarks, monitoring and testers' keys are excluded. Success counts failures on JANCTION's side; failures caused by what was sent, such as an error in the script, are left out.
Track record (last 30 days, external use only): agents connect from Codex, Claude, ChatGPT, Cursor and Claude Code; 99% of renders succeed; median 10 s from submit to result. 3,959 renders since launch. Status
What JANCTION Render is
JANCTION Render is a cloud GPU render farm for Blender that AI agents call as an MCP server or HTTP API. Send a .blend file, a bpy script or a 3D file (glTF / FBX / USD), preview in seconds, and get frames or an MP4 back from JANCTION's NVIDIA GPUs. Every preview also returns a verdict on what is wrong, such as too dark, blown out, no light or an object out of frame, and the bpy code to fix it, so the agent can correct the scene before the final render. No local Blender or GPU is needed.
From an AI app, connect the MCP URL and say:
“Using JANCTION Render, build a glossy red cube rotating on a gray floor and show me a 4-frame preview. If it looks right, render 24 frames at 1280x720 as an MP4.”
Start from your problem: what is really free · render without a GPU · cloud rendering, step by step.

Send · preview · estimate · final
A .blend (textures alongside), a bpy script, a 3D file (glTF / GLB, FBX, USD, OBJ, STL, PLY, Alembic), or a phone scan (a Gaussian splat .spz or a 3DGS .ply). Blender 3.6 / 4.2 / 4.5 / 5.0 / 5.2 / 5.3, Cycles or EEVEE. A missing camera or light is added for you
1 to 4 frames tiled in one image in a few GPU seconds (up to 1280x720), with a verdict (too dark, blown out, no light, object out of frame) and the bpy code that fixes it. Lighting presets night / overcast / studio / sunset (environment="compare" shows all four). Also a long animation as a draft video at its real length; a 360° as the whole turn in one sheet

GPU seconds, wall time and the price before anything runs. A paid final waits for approval. Spending caps per key and per job
PNG / EXR frames or MP4 / WebM / ProRes / GIF / WebP. Up to 3,000 frames per instruction (run as parts under one job id), 1080p on free time, 4K on prepaid credit. Every final comes with a delivery check (all frames, size, black frames, flicker). Before the full range, a 10-frame test with the same settings. A finished render can become a public share page (/r/<id>)
Real renders on the production GPU · 25 cases and counting
Nodes sized by degree and coloured by cluster, edges to the two nearest neighbours.
Each letter scales 0 → 1.25 → 1 and rises from the floor, three frames after the one before.
An open cylinder with Solidify, a torus handle and a dark disc of coffee. Glossy glaze at roughness 0.15.
A sheet of 1 to 4 frames comes back in seconds with a verdict (too dark, blown out, no light, out of frame) and the bpy code that fixes it.
GPU seconds and yen for real jobs on one RTX PRO 6000, from a 4-frame preview to Blender's Classroom scene.
POST /v1/me/limits sets a per-job and per-day cap for the key; a job over the cap is refused (403) before anything is reservedIdempotency-Key makes a retried request render once; Retry-After says when to come back/v1/outcomes turntable / product-shot) never charge more than the quoted price_yenMCP URL: https://render.janction.jp/mcp
| from | how |
|---|---|
| Claude.ai | Plugins (older menu: Customize → Connectors) → Add custom connector → the URL above. "Connect" opens a page that creates a free key |
| ChatGPT | Plugins → Add → Create plugin → the URL above. A .blend attached to the chat reaches the tools |
| Claude Code | claude mcp add --transport http janction-render https://render.janction.jp/mcp |
| Codex | codex mcp add janction-render --url https://render.janction.jp/mcp |
| Cursor / VS Code | Add to Cursor Add to VS Code Add to LM Studio The button opens the editor; confirm and it is added. By hand: Streamable HTTP at the URL above |
| Terminal AIs | Gemini CLI / Antigravity · Kimi Code · Qwen Code · OpenCode · GitHub Copilot CLI · DeepSeek Harness (dsh-mcp-client). One line each (a plugin or mcp add); OAuth issues a free key |
| Grok · Perplexity · Le Chat · Notion | Connectors (plugins) → custom MCP → the URL above. In Grok, send a local file through an /upload link |
| Windsurf · Cline · Goose · others | Streamable HTTP at the URL above. OAuth, or Authorization: Bearer <API key>. llms-install.md |
| Local files as they are | claude mcp add janction-render -e JANCTION_RENDER_SERVER=https://render.janction.jp -- uvx --from janction-render janction-render-mcp (stdio; sends a large .blend in parts), or uvx janction-render preview <file> |
| Blender · Maya · Houdini · Cinema 4D | Blender add-on (preview, estimate and final from the Render properties panel), Maya / Houdini / C4D submitters |
| Plain HTTP | curl -X POST https://render.janction.jp/v1/keys for a key (no sign-up) → OpenAPI |
Every app, step by step · Package: PyPI · Code: GitHub
| tool | what it does |
|---|---|
scene_info | read without rendering: cameras, frame range, resolution, missing files |
render_preview | fast 1 to 4 frame preview (up to 720p); returns the image, a verdict and fix code |
render_estimate | minutes and JPY before rendering, and whether it fits the free time |
render_final | final render (PNG / EXR frames, or MP4 / WebM / ProRes video), frames rendered in parts |
render_status / render_download / render_cancel | status and eta, download (links valid 24 h, only='mp4' for the video alone), cancel |
billing / render_info | balance and free time left, worker availability, accepted formats and presets |
render_share / render_unshare | make a finished render a public page (/r/<id>) or take it down; listed=true asks for the gallery |
asset_search | search Poly Haven (CC0) models, textures and HDRIs by name; put 'polyhaven:<id>' in asset_urls |
render_import_file | take a file attached to the chat (.blend, glTF, FBX, USD, OBJ, STL, .py) and make it a scene_id (apps where attachments reach tools, such as ChatGPT) |
| free key | key that has paid 500 JPY or more | monthly plan (1,500 JPY+) | |
|---|---|---|---|
| Previews | within the welcome credit, then free up to 2 GPU-minutes a day | same | same |
| Finals | up to 1080p, within the welcome credit | up to 4K, 0.1 JPY per GPU-second (10 GPU-hours per job) | same |
| .blend size | up to 1 GB (other 3D files and scripts 512 MB) | up to 5 GB | up to 10 GB |
| Kept | 24 hours after last use | normally 7 days | 30 days after last use |
Never used for training. Download links last 24 hours and are renewed on request while the files are kept.
0.1 JPY per GPU-second (6 JPY per GPU-minute) · measured on the production GPU
| Measured job | GPU s | Price | How many |
|---|---|---|---|
| 4-frame preview, 720p | 4.0 s | 0.4 JPY | ≈ 1,250 |
| 1080p still, 128 samples (six material balls) | 7.0 s | 0.7 JPY | ≈ 714 |
| 24-frame 720p turntable MP4 | 58.3 s | 5.8 JPY | ≈ 85 |
| 5-second 720p animation, 120 frames | 138.2 s | 13.8 JPY | ≈ 36 |
| 1440² glass product still, 128 samples | 166.5 s | 16.7 JPY | ≈ 30 |
Counts are credit ÷ the measured price of each job on the production GPU. Your own scene may take more or less; render_estimate quotes it before anything runs. After the welcome credit, previews stay free up to 2 GPU-minutes a day (about 30 four-frame previews).
Credit goes to this browser's key. For the key your AI app uses, pay from the link the AI gives you.
Callable from MCP or HTTP

A .blend, a bpy script or a 3D file in; frames (PNG / EXR) or a video (MP4 / WebM / ProRes) out. Preview first, estimate, then the final.
metered: 0.1 JPY per GPU-second; a new key's welcome credit of up to 500 JPY covers the first renders. A 24-frame 720p MP4 is about 6 JPY.

A GLB / FBX / USD / OBJ (for example an AI-generated asset) in; four stills at 0 / 90 / 180 / 270 degrees out, studio lighting, optional transparent background. Thumbnails and QA.
fixed ceiling: 6 JPY per asset at 1080p or square (0 while the free time covers it).
The same 3D file in; one full rotation as an MP4 (24 fps, 24 / 48 / 96 frames, 720p or 1080p) out. Product pages and portfolios.
fixed ceiling: 9 to 59 JPY per video depending on size and frames (0 while the free time covers it).
POST /v1/try JSON {scene_script | scene_url} first 720p preview without a key + a key to continue, in one response (1 per network per day)
POST /v1/keys temporary key (no sign-up)
POST /v1/files (multipart file) .blend / .py / .glb ... -> scene_id
POST /v1/files/{id}/assets assets (textures, glTF .bin)
POST /v1/estimate {kind, frames..} time estimate, quota check (no render)
POST /v1/jobs {scene_id, kind: info|preview|final, environment?, blender?, ...} (Idempotency-Key header makes retries safe)
POST /v1/outcomes/turntable | product-shot fixed-price outcomes: price_yen quoted first and never exceeded (GET /v1/outcomes)
POST /v1/me/limits {job_yen, day_yen} spending caps per key; over-cap jobs are refused (403) before anything is reserved
GET /v1/jobs/{id} status, eta, artifacts, cost
GET /v1/jobs/{id}/artifacts/{name}
POST /mcp remote MCP (Streamable HTTP)Benchmarks (measured GPU seconds) · OpenAPI · Status · llms.txt (the guide agents read first) · llms-full.txt · Security · FAQ · GitHub
Classified by what you are trying to do
A cloud GPU rendering service for Blender whose interfaces are MCP tools, an HTTP API and a command line, so AI agents (Claude, ChatGPT, Claude Code, Codex, Cursor) and pipelines can render without a local GPU.
Every monthly plan includes JANCTION Storage (beta): renders and scenes are kept for as long as you are subscribed (Basic 200 GiB, Standard 1 TiB for 3 people, Pro 5 TiB for 5 people). .blend uploads go up to 10 GB. After cancelling, saved work can still be opened and downloaded for 90 days after the paid period ends.
Yes, on a monthly plan. A final's reply carries a one-time link to open it in storage, where each version lists its results and source scene. Ask your AI to open the earlier render and render it again with the same settings, or open it from the browser workbench.
Every monthly plan uploads .blend files up to 10 GB. Scenes with packed textures can be sent as they are and rendered to the final.
No. Smart Render (SmartRender) is a separate service from the same operator, JasmyLab Inc.: a browser upload service that renders .blend files on distributed GPUs, in development. JANCTION Render is the MCP server and HTTP API for AI agents and is live, with a welcome credit of up to 500 JPY per new key. Numbers, partners and dates published about one do not describe the other; the fact sheet at /facts is the primary source for JANCTION Render.
Claude.ai, ChatGPT, Claude Code, Codex, Cursor and any MCP client; curl or any HTTP library; the Python package janction-render.
No high-end GPU? Turn your 3D into images and video.
A Blender file or a scene an AI wrote: preview, fix and finish it on cloud GPUs.