Skip to content

Free, open-source 3D sculpting for Linux, macOS and Windows: dense meshes, remeshing, painting, built-in AI (image/text to 3D), Send to Unreal, game-ready LODs, Gaussian splats and OpenUSD.

7.2 KB
Raw Blame

Generate 3D with your own ComfyUI

If you already run ComfyUI, SculptForge can use it as its image-to-3D and text-to-3D generator. ComfyUI can run on your own computer or on a GPU server at your studio. It runs any workflow you like: Hunyuan3D, TRELLIS, a custom node pack, or your own graph. SculptForge uploads the picture (or sends the description), queues the workflow, waits for it to finish, downloads the mesh it saved and opens it ready to sculpt, remeshed exactly like the other services.

Nothing goes anywhere except your ComfyUI server. There's no account, no key and no cost per model.

Setup

  1. Start ComfyUI. By default it listens on http://127.0.0.1:8188. If SculptForge runs on a different computer, start ComfyUI with --listen (for example python main.py --listen 0.0.0.0) so other machines can reach it. Only do that on a network you trust: ComfyUI has no login.
  2. Install the models your workflow needs. For the bundled Hunyuan3D 2.1 workflow, that's hunyuan_3d_v2.1.safetensors in ComfyUI/models/checkpoints. The workflows README (integrations/comfyui/workflows/README.md) lists every file, its size and its download link.
  3. In SculptForge, open Generate 3D (File > Generate 3D..., or drop a picture onto the window) and pick ComfyUI - your server.
  4. Enter the Server address and press Test connection. You should see the ComfyUI version and your GPU, for example Connected: ComfyUI 0.42.0 - NVIDIA GeForce RTX 4090 (24.0 GB).
  5. Pick a Workflow: one of the bundled ones, or Other file (API format)... for your own.
  6. Press Generate 3D model. The progress bar shows your place in ComfyUI's queue and how long the workflow has been running. Cancel removes the job from the queue, or stops it if it's already running.

Save settings stores the server address and your chosen workflows in ai.json, so they're still set next time you open SculptForge.

Bundled workflows

SculptForge ships these in integrations/comfyui/workflows/ (installed under share/sculptforge/comfyui/workflows):

Workflow Tab Needs
hunyuan3d-image-to-3d From image The Hunyuan3D 2.1 checkpoint (7.4 GB) and a GPU (Tencent quotes about 10 GB of VRAM)
text-to-3d From text FLUX.1 schnell, BiRefNet and Hunyuan3D 2.1 (about 25 GB of downloads) and a GPU
sculptforge-inflate From image The ComfyUI-SculptForge node pack (ComfyUI-Manager: Install via Git URL). No GPU: a quick inflated blockout, handy for checking your setup

When you haven't chosen a workflow, SculptForge uses the first bundled one that fits the tab.

From the command line, Python or an MCP agent

sculptforge-cli -c 'comfyui_check'                                     # is it up? version and GPU
sculptforge-cli -c 'image_to_3d path=hero.png service=comfyui workflow=my-hunyuan-api.json; export_glb path=hero.glb'
sculptforge-cli -c 'text_to_3d prompt="a carved stone owl" service=comfyui url=http://gpu-box:8188 seed=42'

workflow= and url= override your saved settings for that one run. seed= makes a run repeatable; without it, every run gets a fresh seed.

Workflows: API format and the sf_* contract

SculptForge runs workflows saved in ComfyUI's API format. That's not the file ComfyUI's ordinary Save writes. To export one:

  • Current ComfyUI: open the workflow, then use Workflow > Export (API).
  • Older ComfyUI: in the settings, turn on Enable Dev mode Options, then use Save (API Format).

If you choose an ordinary editor save by mistake, SculptForge says so and tells you how to export it.

To tell SculptForge where things go, give nodes these titles (right-click a node > Title):

Title Node What SculptForge does
sf_image Load Image Sets image to the uploaded picture. Every node with this title gets it. Without one, the first Load Image node is used.
sf_prompt Any node with a text, prompt or string input (CLIP Text Encode, String, Primitive String...) Sets that input to your description. Without one, the first CLIP Text Encode that isn't titled "negative" is used.
sf_seed A Primitive Int or similar Sets its value to the seed. Any input named seed or noise_seed gets it too.

Titles are not case-sensitive. Inputs that are wired to another node are left alone, apart from those on the titled nodes above.

The result: SculptForge imports the first mesh file (.glb, .gltf, .obj, .ply, .stl, .fbx, .usdz) that the workflow saves. Core Save GLB, Preview 3D and the SculptForge Export node all work, and so do most 3D node packs. End your workflow with one of them. If the workflow finishes without saving a mesh, SculptForge lists what it did produce.

The picture is prepared the same way as for the hosted services: the subject is cropped out of a transparent background, centred on a square and scaled to at most 1024 px. It's uploaded to ComfyUI's input folder as sculptforge_<random>.png.

Make your own workflow

  1. Build and test the graph in ComfyUI, starting from a Load Image node (or a text node for text to 3D) and ending with a node that saves the mesh.
  2. Title the Load Image node sf_image (or the text node sf_prompt).
  3. Export it with Workflow > Export (API) and choose it in SculptForge.

tools/comfyui-validate-workflow.py --server http://127.0.0.1:8188 my-workflow.json checks the workflow against your server before you run it: missing nodes and models, bad links, and which nodes the contract will use.

Troubleshooting

Message What to do
Nothing is answering at http://... ComfyUI isn't running at that address, or a firewall blocks it. Check the address and port (8188 by default). For another computer, start ComfyUI with --listen.
...answered, but not like ComfyUI Something else is on that port, or a proxy is in the way. Use the address ComfyUI prints when it starts.
This workflow was saved in ComfyUI's editor format Export it again with Workflow > Export (API).
This ComfyUI is missing the nodes ... The workflow uses custom nodes this server doesn't have. Install them (ComfyUI Manager > Install Missing Custom Nodes) and restart ComfyUI.
...names a model file this server doesn't have Download the model into the folder named in the message (for example models/checkpoints).
ComfyUI rejected the workflow: node N ... An input is out of range or not allowed. Open the workflow in ComfyUI: it shows the same error on that node.
The workflow finished but saved no mesh file End the workflow with Save GLB (or another node that saves a mesh). The message lists what it did produce.
... failed: CUDA out of memory The GPU ran out of memory. Try a lower resolution or octree size, or a smaller model.
The job disappeared from ComfyUI's queue Someone cleared the queue, or ComfyUI restarted. Generate again.

On a shared studio server, your job waits in the same queue as everyone else's, and the progress bar shows your place. Cancel only removes or stops your own job.