ComfyUI, without the GPU bill
LivePair nodes put hosted image and video models inside your ComfyUI canvas — the workflow UI you already know, backed by API generation. Pay per result with one prepaid balance; no local checkpoints, no CUDA, no separate provider account.
Running in three steps
Install the nodes
Search LivePair in ComfyUI Manager and click Install — the package is listed on the official ComfyUI registry as livepair. Or clone the repo into custom_nodes; it's stdlib-only, so there are no dependencies to resolve.
Add your lp_ key
Generate an API key from your LivePair billing page and set it as LIVEPAIR_API_KEY — or paste it into the node's key widget. Generations bill prepaid credits per result.
Wire it into a workflow
Drop LivePair Image or LivePair Video onto the canvas like any sampler node. Outputs are standard IMAGE/VIDEO tensors plus the result URL — composite, upscale, and chain with the rest of your graph.
Two nodes, the whole catalog
LivePair Image
IMAGE + urlText-to-image, instruction-based edit, upscale, and background removal. An optional input image routes to i2i edit automatically.
LivePair Video
VIDEO + urlText-to-video and image-to-video on the Wan and Seedance lines. Feed it a still for i2v source frames.
Model picker loads from the live catalog — pricing shown there per resolution and duration matches what your key is charged. New models land via the refresh node; custom accepts any current id.
The ecosystem worth knowing
ComfyUI's power is its node ecosystem — these are the community packs most workflows assume. LivePair nodes sit alongside them: we run the heavy generation in the cloud, they do the local graph work.
Essentials — install these first
The package manager for every node below — install, update, and resolve missing nodes from inside the canvas.
Detection, segmentation, and regional refinement — the standard detail-up and face-fix toolkit.
Quality-of-life nodes and graph helpers — reroutes, groups, fast image ops. Most power workflows assume it.
The missing primitives: masking, compositing, resizing, and simple utility nodes from cubiq.
Kijai's utility set — masking, conditioning, and misc nodes that many video workflows depend on.
Control & conditioning
Every ControlNet preprocessor — depth, pose, canny, lineart — as reusable graph nodes.
Schedules and weights for ControlNet across steps — finer control than the stock nodes.
Prompt formatting, wildcards, and graph conveniences used across shared workflows.
Video
Video load/save/frame ops — required by nearly every video workflow, pairs with our video node output.
Motion modules for animated diffusion — the maintained evolution of the original AnimateDiff nodes.
Kijai's Wan video wrapper — local Wan inference when you do have the GPU; our node covers when you don't.
Identity & finishing
Image-prompt conditioning — the local way to carry a character or style across generations.
Face-identity preservation. Our Instant Character / PuLID models run the same family of technique hosted.
Tiled diffusion upscaling for large outputs — chain it after a LivePair result or use our hosted upscaler.
The big utility grab-bag — image filters, text tools, masks, and hundreds of helper nodes.
Questions
Do I still need a GPU?
Not for the LivePair steps. The nodes call the hosted API — generation runs on LivePair's side. A machine that can run ComfyUI itself (for graph editing, VAE-free previews, and saving) is enough.
Which models are available?
The node dropdowns mirror the LivePair catalog — image models for generation, edit, upscale, and background removal, plus Wan and Seedance video lines for t2v and i2v. LivePairRefreshModels pulls new entries from the API, and a custom option accepts any current model id. See the full live list on the models page.
How is it billed?
Per result, from prepaid credits — the same ledger as the Studio and the API. There is no subscription and no separate provider account to configure.
Where do results go?
The node returns tensors directly into your graph plus the result URL. Hosted renders auto-delete after 48 hours — save locally if you want to keep them.