Send the scene (a .blend, a bpy script or a 3D file) and the render runs on a datacenter NVIDIA GPU with Blender 5.0, 5.2 or 5.3 and Cycles; nothing to install, no instance to keep. This page shows what actually comes back: one command on a public GLB returned four product shots in 13.9 GPU-seconds, a 4-frame preview comes back about 8 seconds after the request with a verdict on what to fix, and a 24-frame 720p clip took about a minute. AI agents connect at https://render.janction.jp/mcp (remote MCP with OAuth; connecting creates a free key).
Updated 2026-11-12
What it does Your own .blend, bpy script or 3D file (glTF/GLB, FBX, USD, OBJ...) rendered to images or video on a cloud NVIDIA GPU, from an AI app, the terminal or the browser. The first preview needs no payment.
What it does not do It does not speed up the Blender viewport on your machine, does not take non-Blender video, simulations or general GPU jobs, and an AI that is not connected cannot render on its own.
Try your own file for freeBuild it with an AI and try
Browser only; no connector or CLI needed. 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.
On 2026-10-10 we ran three commands on the production GPU against a public-domain GLB (the Khronos WaterBottle) and a sample script. The files they produced are in examples/one-command; more results with their scripts are on /examples.
uvx janction-render shots https://github.com/KhronosGroup/glTF-Sample-Assets/blob/main/Models/WaterBottle/glTF-Binary/WaterBottle.glb # product-shots.png: four angles (0, 90, 180, 270 degrees), studio light, plain backdrop. 13.9 GPU-seconds, about 1.4 JPY after the welcome credit. uvx janction-render turntable https://github.com/KhronosGroup/glTF-Sample-Assets/blob/main/Models/WaterBottle/glTF-Binary/WaterBottle.glb # turntable.mp4: one full turn, 24 frames at 720p. 85.2 GPU-seconds, about 8.5 JPY. uvx janction-render render https://render.janction.jp/samples/ex12_logo_ring_spin.py --frames 1-48 --size 1280x720 --engine eevee --output mp4 --wait # anim.mp4: a 48-frame logo animation. 54.1 GPU-seconds, about 5.4 JPY.
The first run creates a key and saves it on the machine; a new key starts with a welcome credit of up to 500 JPY (100 JPY with no card, 400 JPY more after registering a card, which is not charged), which covered all three. The CLI takes a local path or an https link (file pages on GitHub and Hugging Face are turned into raw links). Without a CLI or a connector, the same three things run in the browser at /start. From an AI app (Claude, ChatGPT, Cursor, Claude Code, Codex), the same service is the MCP server https://render.janction.jp/mcp: the agent calls render_preview, render_estimate and render_final as tools.
A preview renders up to four frames of the scene at a small size and returns them tiled in one image, with a verdict (ok, check or fix), a sentence on what is wrong, and bpy code that fixes it. Measured: a 4-frame preview of a rotating cube took 4.0 GPU-seconds and came back 8 seconds after the request (Claude.ai over MCP); single-frame previews of small scenes took 1.8 to 2.4 GPU-seconds and 6 to 8 seconds round trip. A real fix verdict, from a scene whose camera pointed at nothing: "nothing distinct is visible in the frame (the camera may point away from the objects, or the lighting is too flat to tell them apart)", with code that re-aims the camera at the objects. Previews are free within the daily preview time, so the loop preview, fix, preview costs nothing before the final.
A final job answers at once with a job id and an estimate. Its status goes queued, then running (frames done out of total, and an eta in words), then done. When the GPU is lent out or busy, the status says so with the wait, and an agent is told to stop polling and tell you. Measured finals: 24 frames at 1280x720 and 64 samples took 58.7 GPU-seconds and 65 seconds from submission to the file; 48 frames at 1920x1080 and 128 samples took 322 GPU-seconds (Claude Code over MCP). The result comes back as PNG frames or an MP4 through links that need no key and work for about 24 hours; while the files are kept (7 days, normally, for a key that has paid 500 JPY or more in total) asking again gives fresh links. The job is settled on the GPU seconds actually used, never above the ceiling shown before you pressed.
A .blend (files it references are collected and sent with it, or pack them into it), a bpy script as text, or a 3D file: glTF/GLB, FBX, USD, OBJ, STL, PLY, Alembic. Each job runs in a disposable container on a JANCTION GPU (NVIDIA RTX PRO 6000 Blackwell) with Blender and Cycles (EEVEE on request): a .blend opens with the version it was saved in (5.0, 5.2 or 5.3), and scripts and 3D files use 5.0 unless you pass blender='5.2' or '5.3'. The container has no network access and is destroyed after the job. Sizes, frame limits and how long results stay: /facts. Prices and the welcome credit: /pricing.
Not sure this is the right way for your case? How to render Blender without a GPU decides between your CPU, Colab, SheepIt, render farms and this, with the same measurements.
Blender 5.0, 5.2 or 5.3 (Cycles on NVIDIA GPUs). A .blend opens with the version it was saved in; scripts and 3D files use 5.0 unless you pass blender='5.2' or '5.3'. A bpy script is the most portable input. Blender 3.6, 4.2 and 4.5 are there too: pass blender='4.5' (or '4.2', '3.6'), and a script that stops on Blender 5 because it uses an older API is rendered again on 4.5, 4.2 or 3.6 automatically.
Yes: the Python package sends files referenced by the .blend automatically (blender-asset-tracer), and the HTTP API and MCP tools accept assets / asset_urls. Packing them into the .blend also works.
Yes: render_cancel from an AI app, or DELETE /v1/jobs/{job_id} over the API, or janction-render cancel on the command line. GPU seconds already used are charged; the rest of the amount held for the job is released.
