← ComfyTV Image

The Bridge family is how ComfyTV stops being a walled garden. Every ComfyTV stage passes images around in its own COMFYTV_IMAGE type - a reference into the project's asset store rather than a plain tensor - so a normal ComfyUI node can't consume it directly. BridgeFromImage is the adapter: it takes a COMFYTV_IMAGE and emits a standard IMAGE output that any conventional ComfyUI image node happily accepts.

One input (image, a COMFYTV_IMAGE), one output (IMAGE), nothing else. Mechanism-wise it resolves the reference to the file in the project store, loads it, and hands over a proper image tensor. That's the entire job, and it does it instantly.

Where it earns its keep: ComfyTV ships a huge built-in suite, but "huge" is not "everything." When your workflow needs a specific third-party upscaler, an IPAdapter pass, a detailer, or a custom node only you run - grab the ComfyTV image, drop it through this bridge, run it through the plugin, then send the result back through → ComfyTV Image (BridgeToImage) to re-enter the project. That round-trip is the pattern the README's bridges guide is built around: "subgraphs and third-party plugins just work; Bridge nodes connect any plugin into a ComfyTV pipeline."

Honest note: it's pure plumbing, so there's no tuning, no options, nothing to get wrong - the only failure mode is feeding it an empty or stale reference (error: "input is empty"), which usually means the upstream stage hasn't been Run yet in this session, or the reference died when the project reloaded. Re-run the upstream stage and it resolves.

One design detail worth knowing: unlike the FX stages, BridgeFrom* nodes are not output nodes in the ComfyTV sense - they're the raw handoff at the boundary, exactly where you'd expect the ecosystem to start. That means the canvas treats them as part of the standard graph flow, not as project stages with their own history.

Install is the pack-wide story: ComfyUI Manager search "ComfyTV", or git clone https://github.com/jtydhr88/ComfyTV into custom_nodes/, then a full backend restart (Desktop/macOS: clone by absolute path into the running instance). Zero extra pip dependencies - it uses PIL and torch, both already in ComfyUI, no models.