Render Studio on your GPU
By default, everything you make in webfaCe Studio renders on our studio GPU. A production company can render on a machine of its choosing instead: a workstation in the office, a DGX Spark, or a GPU rented by the hour. Studio then sends every picture, video shot, effect and song to that machine, and nothing waits in our queue.
You set the machine up once. After that, Studio works the same as before.
What the machine needs
Section titled “What the machine needs”| Minimum | Better | |
|---|---|---|
| GPU | NVIDIA with 24 GB of memory (RTX 4090, RTX 3090) | NVIDIA with 48 GB or more (RTX 6000 Ada, L40S, A6000), 80 GB (A100, H100), or a DGX Spark |
| System memory | 64 GB | 128 GB |
| Disk | 300 GB free | 500 GB NVMe |
| System | Linux | Ubuntu 22.04 or 24.04 |
| Network | Port 8288 open to the internet | An HTTPS address in front of port 8288 |
Macs, AMD GPUs and Intel GPUs can’t run Studio’s models; they need an NVIDIA GPU.
On a 24 GB card Studio works, but the biggest video models only partly fit in the GPU, so shots render more slowly and the longest ones may fail. From 48 GB everything fits.
No machine? Rent one
Section titled “No machine? Rent one”Cloud GPU services such as RunPod, Lambda and Vast.ai rent machines by the hour. Choose one with an NVIDIA GPU of 48 GB or more, Ubuntu, and at least 300 GB of disk, and open port 8288. On RunPod, use the pod’s HTTPS address for port 8288 when you connect.
Install only what you use
Section titled “Install only what you use”The full install downloads about 210 GB of models. To install less, put FEATURES="pictures video" (or any of the names below) in front of bash in your install line.
| Feature | What it makes | Download |
|---|---|---|
pictures |
Pictures, keyframes, picture edits | 22 GB |
video |
Video shots, storyboard animation | 45 GB |
follow |
Act it out, Turn it into on moving shots | 45 GB |
talk |
Talking shots and music video performances | 35 GB |
effects |
Replace, Remove, Add | 35 GB |
songs |
Songs and film music | 22 GB |
polish |
Sharpen to 1080p, 48 frames a second | 7.3 GB |
Set it up
Section titled “Set it up”1. Get a Hugging Face token
Section titled “1. Get a Hugging Face token”Two of the models (FLUX 3 Klein for pictures, LTX 2.5 for video) are free, but their makers ask you to accept a licence first.
- Make a free account at huggingface.co.
- Open FLUX 3 Klein and LTX 2.5, and accept each licence.
- Make a Read token in your token settings and copy it.
2. Make your install line in Studio
Section titled “2. Make your install line in Studio”- Open Studio and choose Use your own GPU.
- Paste your Hugging Face token into step 1 of the panel. It stays in your browser: it only goes into the line you copy, never to us.
- Choose Make my install line, then Copy the line.
The line looks like this, with your two tokens in it:
curl -fsSL https://docs.webfacemedia.com/studio-machine/install.sh | STUDIO_TOKEN=… HF_TOKEN=… bashSTUDIO_TOKEN is the key to your machine: only requests that carry it get through. Don’t share the line.
3. Run it on the machine
Section titled “3. Run it on the machine”Open a terminal on the machine (or the rented GPU’s web terminal), paste the line and press Enter.
It checks the machine first and stops with a plain message if something is missing: no NVIDIA GPU, too little GPU memory, not enough disk, or a licence you haven’t accepted yet. Then it installs everything and downloads the models. When it finishes, it starts the machine (it starts again by itself after a restart) and prints the address to use, such as http://203.0.113.5:8288.
4. Connect it
Section titled “4. Connect it”Back in Studio’s Use your own GPU panel, paste the address and choose Connect and check. Studio lists each feature as Ready or Missing models. Once it’s connected, the button reads Rendering on your GPU.
To go back to our studio GPU at any time, choose Use the studio GPU again.
If something goes wrong
Section titled “If something goes wrong”- “Missing models” for a feature: run the same install line again. It downloads only what’s missing.
- “The Studio node patch is not applied”: run the install line again.
- Studio can’t reach the machine: check that port 8288 is open in the machine’s firewall (and in the cloud provider’s network settings), and that the address starts with
http://or, on RunPod,https://. - “These models need a free Hugging Face account that has accepted their licence”: open the pages it lists, accept the licence with the same account your token comes from, and run the line again.
Still stuck? Email support@webfacemedia.com with the last lines the install printed.