comfyui-uberllm

comfyui-uberllm

ComfyUI nodes that make videos and pictures with your own UberLLM API key: Veo 3.1, Kling, Seedance, Wan, Hailuo, Vidu, LTX and more for video; Flux, Seedream, Nano Banana, Z-Image and more for pictures. One key and one prepaid balance for all of them, no GPU needed.

No UberLLM fee on requests. Adding credit by card costs 5.5% + $0.50.

Nodes

| Node | What it does | Main inputs | Outputs | |---|---|---|---| | UberLLM Text to Video | A video from a prompt | model, prompt, duration, resolution, shape, sound, max cost | VIDEO, file path, cost in USD, job id | | UberLLM Image to Video | A video that starts from a picture (or uses it as a reference) | image, optional end image, and the same settings | VIDEO, file path, cost in USD, job id | | UberLLM Image | Pictures from a prompt | model, prompt, size, how many | IMAGE, cost in USD | | UberLLM Image Edit | Changes a picture from a prompt | image, optional reference image, model, prompt | IMAGE, cost in USD |

• The model lists are read from UberLLM's public model list when ComfyUI starts (4-second limit), so new models show up without updating the pack. Offline, the list shipped in models_cache.json is used. Veo, Kling and Seedance are listed first.
• auto (and a length of 0) leaves a setting to the model. A length, size or shape a model does not make is refused before anything is charged, with a message that lists the ones it does.
• max_cost (US dollars, 0 = no limit): if the request can cost more, nothing is made or charged.
• The video nodes save the file to ComfyUI/output/uberllm/ and show it in the node. Connect video to any node that takes VIDEO (for example Save Video). UberLLM keeps the file for about 3 hours, so the local copy is the one to keep.
• The console shows each job's id, its listed price and the most it can cost before it runs, then what it cost.

Install

ComfyUI-Manager: search for UberLLM in Custom Nodes Manager and install.

By hand:

cd ComfyUI/custom_nodes
git clone https://github.com/PSA-Team-source/comfyui-uberllm

The pack needs nothing beyond what ComfyUI already has (Python standard library, torch, numpy, Pillow). ComfyUI 0.3.31 or newer.

Your API key

1. Sign in at uberllm.dev (email code), add credit, and make a key under Dashboard → API keys.

2. Start ComfyUI with the key in the environment:

   export UBERLLM_API_KEY=ull_...      # Windows: set UBERLLM_API_KEY=ull_...
   python main.py
   

Each node also has an api_key field. It is for quick tests only: a key typed there is saved inside the workflow file and in the metadata of the files ComfyUI makes, so anyone you share them with gets your key.

Example workflows

Drag a file from examples/ onto the ComfyUI canvas.

| Workflow | Models | Listed price (checked 2026-10-30) | |---|---|---| | text_to_video_veo31_fast.json | Veo 3.1 Fast, 4 s, 720p, no sound | $0.30 | | image_then_video_kling.json | Z-Image Turbo picture, then Kling 3 Standard image to video, 3 s, no sound | about $0.008 + $0.189 |

Prices change with the market; each model's current price is on its page at uberllm.dev and in the console when a job starts.

What you pay

• What the provider charged for what was made, never more than the most-it-can-cost shown when the job starts. See every job in Dashboard → Activity.
• A video job that fails or expires costs nothing.
• Stopping a run in ComfyUI asks UberLLM to stop the job. A job the provider has already started can still finish and be billed.

API details: uberllm.dev/docs/media.

Settings

| Environment variable | Default | Use | |---|---|---| | UBERLLM_API_KEY | (none) | Your key | | UBERLLM_BASE_URL | https://api.uberllm.dev | API address | | UBERLLM_MODELS_URL | https://api.uberllm.dev/api/v1/public/models | Where the model lists are read from |

Refresh the shipped model list: python3 uberllm_models.py --write-cache. Offline checks: python3 tests/test_offline.py.

Publishing (maintainers)

The pack is published to the Comfy Registry, which feeds ComfyUI-Manager. pyproject.toml names the publisher uberllm; that publisher has to exist on the registry first.

1. Sign in at registry.comfy.org and create the publisher with id uberllm (it can't be changed later; if that id is taken, put the one you get in [tool.comfy] PublisherId).
2. On the registry, open the publisher and create an API key. Copy it; it is shown once.
3. Either run pip install comfy-cli && comfy node publish in this folder and paste the key, or publish from GitHub Actions: move .github/publish_action.yml to .github/workflows/publish_action.yml (pushing a workflow file needs a GitHub login with the workflow permission), add the key as the Actions secret REGISTRY_ACCESS_TOKEN (Settings → Secrets and variables → Actions), and run Publish to Comfy registry. It then also runs on every push to main that changes pyproject.toml.
4. For a new release, raise version in pyproject.toml and push.

License

MIT