How to run Stable Fast 3D (image-to-3D) on a cloud GPU
Rent a GPU on Deploy, then open a terminal in JupyterLab or over SSH. Install Stable Fast 3D from the official Stability-AI/stable-fast-3d repository the way the README prints, accept the gated model, and run python run.py demo_files/examples/chair1.png --output-dir output/ or python gradio_app.py. Copy the GLB under output/ off the instance before you destroy it, because destroying the instance deletes its disk.
A single-image reconstruction whose README says the default options take about 6GB VRAM is How to run TripoSR image-to-3D on a cloud GPU. TRELLIS’s README says at least 16GB is necessary: How to run TRELLIS image-to-3D on a cloud GPU. InstantMesh’s README, app.py, and run.py print no GPU GB figure: How to run InstantMesh image-to-3D on a cloud GPU. Hunyuan3D-2.1’s README publishes 10 GB, 21 GB, and 29 GB: How to run Hunyuan3D on a cloud GPU. Rendering a .blend file is How to render a Blender scene on a cloud GPU. If the process runs out of memory: How to fix CUDA out of memory on a cloud GPU. The tunnel for the Gradio app is How to use SSH port forwarding on a cloud GPU.
LICENSE.md (commit d5aea7418cce16ed7e841fc3ebf242e6af74e496, 2024-08-01) is the Stability AI Community License Agreement, last updated July 5, 2024. The GitHub license field on 2026-11-13 was NOASSERTION. The model API the same day reported cardData.license other, license_name stabilityai-ai-community, gated auto, pipeline_tag image-to-3d, lastModified 2025-04-08T16:31:08.000Z, sha f0c9a8ffd62cb1bbc8a7a53c9f87a0be1b6be778. A fetch of the model card README returned HTTP 401, so this page does not quote the card body. The safetensors total in that API response was 1006027436, all F32. This page does not convert that count into GB.
The commands below are copied from the upstream files named in each step. They were not run on a CloudGPU instance for this page. Anything this page adds around them is marked.
What the README says about memory
The README (commit 87cd0b88703fbee348b6fe011c2d191d899f353b, 2025-01-03; repository HEAD is ff21fc491b4dc5314bf6734c7c0dabd86b5f5bb2, committed 2025-01-22T14:55:42Z) says the default options take about 6GB VRAM for a single image input. It does not print a second CUDA GB figure. The MPS section says support was tested on an M1 max with 64GB and the latest PyTorch nightly, recommends PyTorch 2.4.0 as of writing, and says MPS currently consumes more memory compared to the CUDA PyTorch backend. It recommends the CPU version if the system has less than 32GB of unified memory. That 32GB sentence is about unified memory in the MPS section. This page does not treat it as a CUDA vram_gb comparison. This page did not measure the 6GB sentence.
GET /v1/public/compute-capabilities does not list GPUs. It returns feature flags. The memory column is vram_gb from gpu-prices.json, updated_at 2026-11-13T21:06:02.664Z. GET /v1/public/gpu-catalog the same day (generated_at 2026-11-13T21:06:03.185Z) listed the same names and the same vram_gb values, plus Ascend 910B at 64, which is not in gpu-prices.json and is not listed here. The catalog response also includes an availability field. This page does not copy it. Every prices row has gpu_count 1. Every catalog row has min_gpu_count 1. Every printed vram_gb is greater than 6. That comparison is arithmetic against the README’s about 6GB. This page did not run Stable Fast 3D and does not mark a row as enough. Which listing you can start is on Deploy.
The catalog template list had a template named ComfyUI, id flux-1-dev. It had no template whose id or name contained stable-fast or sf3d.
| Catalog name | vram_gb |
|---|---|
| RTX 2080 | 8 |
| RTX 3080 | 10 |
| RTX 2080 Ti | 11 |
| RTX 3080 Ti | 12 |
| V100 16G | 16 |
| A30 | 24 |
| RTX 3090 | 24 |
| RTX 4090 | 24 |
| RTX 4090D | 24 |
| RTX 4080 SUPER 32G | 32 |
| RTX 5090 | 32 |
| RTX 5090D | 32 |
| V100 32G | 32 |
| A100 40G | 40 |
| A800 40G | 40 |
| A40 | 48 |
| RTX 4090 48G | 48 |
| RTX 4090D 48G | 48 |
| RTX A6000 | 48 |
| NVIDIA RTX PRO 5000 72G | 72 |
| RTX PRO 5000 72G | 72 |
| A100 80G | 80 |
| A800 80G | 80 |
| H800 | 80 |
| RTX 6000D | 83 |
| H20 96G | 96 |
| RTX PRO 6000 96G | 96 |
| RTX PRO 6000 Max-Q | 96 |
| H20-3e 141G | 141 |
Install and gated access
On Deploy, choose Raw GPU. The README says Python >= 3.8, optional CUDA or MPS, and PyTorch installed from the local install page it links, with the PyTorch CUDA version matching the system. How to install a specific PyTorch CUDA version is the index check against the CUDA Version in the nvidia-smi header. The README does not pin a torch wheel. This page did not pick one.
The install section does not print git clone. requirements.txt and run.py are files in that repository. The clone below is this page’s, and it comes first so those files exist. It was not run.
# The README install section does not print a clone. These two lines are this page's. They were not run.
git clone https://github.com/Stability-AI/stable-fast-3d.git
cd stable-fast-3d
The dependency lines, copied:
pip install -U setuptools==69.5.1
pip install wheel
pip install -r requirements.txt
requirements.txt (commit 9a65f9f0c3cfb962446326e5c048194f99a199f2, 2025-01-03) pins huggingface-hub==0.23.4, transformers==4.42.3, trimesh==4.4.1, and numpy==1.26.4. It does not pin torch. For the Gradio demo the README adds pip install -r requirements-demo.txt. That file (commit 872c000d5328569b864b0d5e92657621cbb0b4c5, 2024-08-12) pins gradio==4.41.0 and gradio-litmodel3d==0.0.1.
The README’s access steps are: log in to Hugging Face and request access at the model page it links, create a read token, then run huggingface-cli login and enter the token. requirements.txt pins huggingface-hub==0.23.4. The Hugging Face CLI docs read on 2026-11-13 use hf and do not print huggingface-cli. This page keeps the README’s login line and does not upgrade the pin.
huggingface-cli login
The model id in that access link, and the default of --pretrained-model in run.py, is stabilityai/stable-fast-3d. run.py (commit 18eeb2ca799879289c7cf402f609491576aae9b6, 2025-01-03) passes that string to SF3D.from_pretrained. This page does not add a separate download command.
The README also documents an experimental MPS path, an experimental Windows path that names Visual Studio 2022, and SF3D_USE_CPU=1 to force the CPU backend. This page did not run those paths.
run.py
The README command, copied:
python run.py demo_files/examples/chair1.png --output-dir output/
The README says this saves the reconstructed 3D model as a GLB file to output/. run.py writes output/ plus the image index plus input.png, then mesh.glb in that same index directory. --texture-resolution defaults to 1024. --remesh_option accepts none, triangle, or quad, and the default is none. --target_vertex_count defaults to -1. The help says -1 does not perform a reduction. The README says you may pass more than one image path separated by spaces.
Gradio
The README command, copied:
python gradio_app.py
gradio_app.py (commit b48137202c85a2f57e6ebd437407f78cba658aac, 2025-01-12) calls SF3D.from_pretrained with stabilityai/stable-fast-3d, exports a GLB through NamedTemporaryFile with suffix .glb, and ends with demo.queue().launch(share=False). That call does not pass a host or a port. This page does not name one. Forward the URL the process prints with how to use SSH port forwarding on a cloud GPU. This page did not start the app.
ComfyUI
The README documents a ComfyUI extension in this repository. The install it prints is:
cd ComfyUI/custom_nodes
git clone https://github.com/Stability-AI/stable-fast-3d
cd stable-fast-3d
pip install -r requirements.txt
It says to restart ComfyUI after that, and it links demo_files/workflows/sf3d_example.json. Installing ComfyUI itself is How to install ComfyUI and ComfyUI-Manager on a cloud GPU. This page did not check whether the catalog’s ComfyUI template already contains this node.
When it fails
CUDA out of memory. The README’s CUDA figure for the default options is about 6GB VRAM for one image. It does not print an offload flag. If the process raises that error, use how to fix CUDA out of memory on a cloud GPU.
“CUDA error: no kernel image is available for execution on the device” on an RTX 5090. The README does not pin a torch wheel and does not name compute capability 12.0. The RTX 5090 guide says that card is compute capability 12.0 (sm_120) and that PyTorch 2.7 and later, on a build for CUDA 12.8 or newer, is the line that announced support. This page did not install a torch wheel and did not re-test Stable Fast 3D on an RTX 5090. Catalog RTX 5090 and RTX 5090D are 32. That is the catalog number only.
Gated download. The README says the model is gated and that huggingface-cli login must be run after access is granted. This page did not request access.
Save the mesh, then destroy
Copy output/ from the documented run.py command before you destroy the instance. Save your work before you destroy a cloud GPU is the checklist. The terms say data on local instance storage is permanently deleted when the instance is destroyed.
The pricing page FAQ states the billing rule in these words:
Per-minute options have a 1-minute minimum; unused prepaid minutes are refunded when you destroy the instance. Some configurations use whole-hour billing with a 1-hour minimum; unused time within a charged hour is not refunded on voluntary stop. The billing period, minimum charge and initial prepayment are shown before confirmation. No subscription.
Frequently asked questions
How much VRAM does the Stable Fast 3D README publish?
The README says the default options take about 6GB VRAM for a single image input. The MPS section says MPS currently consumes more memory than the CUDA backend and recommends the CPU version if the system has less than 32GB of unified memory. That 32GB sentence is about unified memory in the MPS section. It is not a CUDA catalog comparison. The model API safetensors total on 2026-11-13 was 1006027436, all F32. This page does not convert that count into GB. A fetch of the model card README returned HTTP 401. Every vram_gb in gpu-prices.json (updated_at 2026-11-13T21:06:02.664Z) is greater than 6. This page did not run Stable Fast 3D and does not mark a row as enough. Which listing you can start is on https://cloudgpu.app/deploy.
Which license applies?
LICENSE.md is the Stability AI Community License Agreement, last updated July 5, 2024. Section III says that if at any time You or Your Affiliate(s), either individually or in aggregate, generate more than USD $1,000,000 in annual revenue (or the equivalent thereof in Your local currency), regardless of whether that revenue is generated directly or indirectly from the Stability AI Materials or Derivative Works, any licenses granted to You under this Agreement shall terminate as of such date. That dollar figure is the upstream license threshold, not a CloudGPU price. The GitHub license field on 2026-11-13 was NOASSERTION. The model API cardData.license was other, license_name was stabilityai-ai-community, and gated was auto. This page does not quote the card body. Read LICENSE.md before you redistribute the program or the weights.
Where does run.py save the mesh, and what does the Gradio app write?
The README command is python run.py demo_files/examples/chair1.png --output-dir output/. It says this saves a GLB file to output/. run.py writes output/ plus the image index plus input.png, then mesh.glb in that directory. --remesh_option accepts none, triangle, or quad, and the default is none. --texture-resolution defaults to 1024. gradio_app.py exports a glb through NamedTemporaryFile and calls demo.queue().launch(share=False) with no host and no port. This page did not start it.
Does the README document ComfyUI?
The README says to clone this repository into ComfyUI/custom_nodes, pip install -r requirements.txt, and restart ComfyUI. It links demo_files/workflows/sf3d_example.json. The catalog template list had a template named ComfyUI and no template whose id or name contained stable-fast or sf3d. This page did not check whether that image already contains the node.
Sources (read 2026-11-13, Asia/Shanghai)
- stable-fast-3d README (Python >= 3.8, setuptools pin, gated login, about 6GB VRAM,
run.pycommand, Gradio command, ComfyUI clone, MPS 32GB unified-memory sentence): https://github.com/Stability-AI/stable-fast-3d commit87cd0b88703fbee348b6fe011c2d191d899f353b. Repository HEADff21fc491b4dc5314bf6734c7c0dabd86b5f5bb2 LICENSE.md(Stability AI Community License Agreement, last updated July 5, 2024, annual-revenue termination sentence): commitd5aea7418cce16ed7e841fc3ebf242e6af74e496. GitHub license field NOASSERTIONrun.py(--pretrained-modeldefaultstabilityai/stable-fast-3d,output/<index>/mesh.glb,--texture-resolution1024,--remesh_optiondefaultnone): commit18eeb2ca799879289c7cf402f609491576aae9b6gradio_app.py(launch(share=False), GLB viaNamedTemporaryFile): commitb48137202c85a2f57e6ebd437407f78cba658aacrequirements.txt(huggingface-hub==0.23.4): commit9a65f9f0c3cfb962446326e5c048194f99a199f2.requirements-demo.txt(gradio==4.41.0): commit872c000d5328569b864b0d5e92657621cbb0b4c5- Model API
stabilityai/stable-fast-3d(licenseother,license_namestabilityai-ai-community,gatedauto, safetensors total 1006027436 F32, shaf0c9a8ffd62cb1bbc8a7a53c9f87a0be1b6be778). Card README fetch returned HTTP 401 - Public GPU memory: https://cloudgpu.app/gpu-prices.json (
updated_at2026-11-13T21:06:02.664Z) and https://cloudgpu.app/v1/public/gpu-catalog (generated_at2026-11-13T21:06:03.185Z).GET /v1/public/compute-capabilitiesreturned feature flags and no GPU list. The catalog had no stable-fast template
Cards you can start are on the Deploy page. Prices are on https://cloudgpu.app/pricing and are not repeated here.
Last updated: 2026-11-13 (Asia/Shanghai).