API reference contents
FrameThrower for ComfyUI
Search a library of film stills from inside your ComfyUI canvas. Pick a frame and it comes out as a picture, as depth, pose and lineart maps computed on your own machine, and as text.
No API key to paste, no pip install, and no cloud GPU. Everything the node imports already ships with ComfyUI, and the depth, pose and lineart maps are computed locally.
Install the node
You need ComfyUI 0.3.30 or newer. Both routes below end the same way: restart ComfyUI, and the node appears under Add Node → FrameThrower → Reference Node.
With git
The shortest path, and the one that depends on nothing but git.
cd ComfyUI/custom_nodes git clone https://github.com/framethrower-ai/comfyui-framethrower
Then restart ComfyUI.
With ComfyUI Manager
Open the Manager
In ComfyUI, click Manager, then Custom Nodes Manager.
Search for it
Type framethrower in the search box and press Install on FrameThrower Reference.
Restart
Restart ComfyUI so the node loads.
Connect your account
Drop the node on the canvas. Until it is connected it shows a sign-in panel where the grid will be, with two ways in.
Sign in with browser
Opens your API settings, where you create a token and paste it back into the node.
Or use a pairing code
The node shows a short code; enter it at framethrower.ai/link. This is the one to use when ComfyUI runs on a remote GPU and you are not browsing from that machine.
No account yet? Create one free — no card, and new accounts start with 2,000 credits.
ComfyUI/user/framethrower/config.json. It is never written into a workflow file, so sharing a workflow shares no credentials — whoever opens it signs in as themselves.Your first search
Type what you are looking for
Plain English — kung fu dojo, neon rain at night. The grid fills as you type; you do not have to queue anything to search.
Click a frame
That pins it, so the same frame is used on every run until you pick another one. Drag a frame onto the canvas instead and you get a plain Load Image.
Queue Prompt
Queueing is what produces pictures — the grid is only choosing. A preprocessor runs only while its output socket is wired, so unplug what you do not need.
What comes out of it
image— the frame itself, full size.depth— a depth map, computed locally with Depth-Anything-V2-Small.pose— a skeleton in OpenPose layout, computed locally with ViTPose. Empty when nobody is in the frame.lineart— a line drawing. The detail slider appears on the node once this socket is wired.prompt (text)— the frame’s scene description as plain text. Not conditioning: wire it into CLIP Text Encode to use it as a prompt.credit (text)— title, year and director, so attribution can travel with your work.
The starter workflow
The node ships with an annotated graph — every output wired, and a note beside each group explaining what it does. Open it from Workflow → Browse Templates → comfyui-framethrower → getting-started.
Or take the file directly: getting-started.json — drag it onto the canvas.
What it costs
Searching is metered against your FrameThrower credits: a search costs 2, an image or colour search 10. Depth, pose and lineart run on your own machine and cost nothing. The node also caches on its inputs, so re-queueing an unchanged graph does not spend anything.
$1 = 1,000 credits, and they never expire. Balance and top-ups live in billing.
Troubleshooting
The node does not appear after installing
Restart ComfyUI — custom nodes are loaded at startup. If it still is not there, check that the folder landed in ComfyUI/custom_nodes/comfyui-framethrower and look at the ComfyUI console for an import error.
The node loads but its body is blank
The node’s interface needs ComfyUI frontend 1.16 or newer, which ships with ComfyUI 0.3.30. Update ComfyUI and restart.
ComfyUI runs on a remote machine
Use the pairing code rather than browser sign-in — the browser flow expects to open a tab on the same machine.
“Not enough credits”
Searching spends credits; the local preprocessors do not. Top up in billing — the node links straight there when it hits an empty balance.
Links
- Source on GitHub — issues and releases.
- Comfy Registry listing
- API reference — the same library over REST.
- MCP server — for agents.
- How these frames may be used — worth a minute before you build with them.