Upscale Video

The node that makes your 480P clip presentable

Local video super-resolution in ComfyUI is a pain - chains of upscale models, frame interpolation, VRAM checks, and an hour of babysitting. Animon Upscale Video sidesteps all of it: hand it a VIDEO_ID, and Animon's hosted anisr model returns your clip at 1080P and 24 FPS, ready to drop into SaveVideo. It's a cloud upscaler with a very narrow, very clear job.

The trade-off is that it's not magic for every input. The README states the two hard limits up front: the input video must be smaller than 1080P, and it can hold no more than 360 frames. That's twelve seconds at 30 FPS, or about four and a half seconds of the default 81-frame generations this pack produces. Longer clips need trimming before you even think about upscaling.

How it works

Like the rest of the pack, this node talks to Animon's OpenAI-compatible API. It calls videos.create with the anisr model and your video's server-side ID, polls every ten seconds until the job completes, then downloads the result and decodes it with decord into a ComfyUI VIDEO tensor. Same rhythm as the generation nodes, one difference: the input is an ID, not a file or tensor.

That's the input that trips people up. video_id is an ANIMON_VIDEO_ID, and the only way to get one is an upload node. If your clip came from Animon's own Image to Video node, wire that node's VIDEO_BYTES output into Upload Video (from Bytes) to get the ID without re-encoding - the README's fast path, and one of the pack's neatest moves. If it came from anywhere else, Upload Video or Upload Video (from Tensor) first, then chain into here. The model and resolution widgets are fixed to anisr and 1080P; there's nothing to tune, which is either refreshing or a reminder that this node has exactly one setting that matters (your budget).

The workflow that actually makes sense

The flagship pattern, straight from the included AnimonI2V_Upscale example: Image to Video → its VIDEO_BYTES → Upload Video (from Bytes) → VIDEO_ID → this node → SaveVideo. One cloud call generates the base clip, a second upscales it, and you end up with a 1080P file without ever running a video model locally. For anime-style output where resolution matters more than fidelity debates, it genuinely works.

Installing and the honest catches

Install is the pack standard: ComfyUI Manager → search ComfyUI-Animon → restart, with pip install -r requirements.txt (just openai and decord) if you cloned manually. You need a recent ComfyUI for the io.ComfyNode definitions, an Animon API key, and an internet connection.

Two real catches before you commit. First, upscaling is another paid API call on top of whatever generated the source video - budget for two jobs per final clip, plus the 10-second polling loop on each. Second, remember that "smaller than 1080P" is enforced server-side; feeding it a 4K clip will just fail, and feeding it a 361-frame clip will too. Trim first, then upscale.