HitPaw Video Enhance

HitPaw Video Enhance is the "fix this low-res footage" node: upscale a muddy clip to high resolution, and clean up the artifacts and noise along the way. It's the video sibling of the image upscalers the KB covers in depth - the same "resolution is only half the job, the model has to invent the missing detail" dynamic, just applied across frames. And like every partner node here, it runs in the cloud through Comfy's API: you feed a VIDEO, HitPaw's enhancers process it, an improved VIDEO comes back, and you're billed per second of video. Your GPU stays idle.

Where this shines is exactly where local video upscaling hurts: local enhancement means chaining ESRGAN-style models frame-by-frame, babysitting VRAM, and fighting temporal consistency. HitPaw bundles the model choice and the per-frame consistency into a single call. The KB's upscaling essay is the right mental model - a good upscaler doesn't just interpolate pixels, it reconstructs plausible detail, and that's what you're paying for here.

The inputs

• video - the clip to enhance.
• model - the dynamic dropdown, and it's where the meaningful choices live. HitPaw offers several model families (standard, ultra-HD, generative) and, once you pick one, a resolution sub-option that ranges up to 8K depending on the model. In other words: the model dropdown is the settings panel - standard models for straight upscaling, ultra-HD for heavy detail recovery, generative for adding plausible texture back.

That's the whole input surface. No seed, no prompt, no duration knobs - drop a video, pick an enhancer, go.

What comes out

A single VIDEO output, upscaled and cleaned.

Gotchas

• Price is per second and scales with resolution. The node shows it right on the badge - the cost is quoted per second of video, and 4K/8K runs cost meaningfully more than 720p. A long clip at max resolution adds up fast, so upscale the clips that matter, not the whole raw dump.
• Some models cap their resolution range. The node only offers resolutions the chosen model supports - if 4K isn't in the list, it's not a bug, it's the model's ceiling.
• Garbage in, prettier garbage out. Enhancements amplify the source's problems along with its detail; a badly deinterlaced or heavily compressed source only gets so much better.
• It's a cloud call with real latency. Expect a wait proportional to clip length; that's the trade for not melting your own GPU.

The honest verdict: if you upscale video once in a while, this beats babysitting a local chain. If you upscale constantly, do the math on per-second pricing against your local setup - this node is convenience, and convenience has a meter running.