zhenzhen-FlashVSR-video-upscale-lowprice

You just generated a video, and it's 480p. Every local video upscaler you've tried either needs a 24GB card or takes an hour per clip - or both. Comfly_fashvsr_video_upscale_lowprice is the shortcut: it sends the clip to a FlashVSR video super-resolution model through Zhenzhen's China-facing API and hands you back a higher-res VIDEO tensor, ready to drop straight into ComfyUI's video save node. It's the "lowprice" family, so it bills through the domestic seedance relay at a discount, and the whole point is that your GPU isn't doing the heavy lifting.

FlashVSR is a video super-resolution method that's fast enough to feel real-time on the server side, which is why it slots so neatly into a relay node - you submit, it upscales, you download. No local model weights involved.

The one input that decides everything

The required input is video_url, and it comes with a tooltip that is basically the whole contract: "Optional public video URL. Leave empty when input_video is connected. Source video must be 480P and 3-15 seconds." So:

• Feed it either a public video_url or an input_video VIDEO tensor - never both (the node validates this and will tell you off if you do).
• The source really does need to be 480P and 3–15 seconds. Feed it a 1080p clip or a 30-second monster and it will refuse or give you a mediocre result.

On the optional side: api_config is the ZHENZHEN_SEEDANCE2_CONFIG input - wire in the pack's Seedance2 low-price settings node (it carries your API key and the domestic base URL). seed is purely a ComfyUI cache seed; the tooltip is explicit that it is not sent to FlashVSR, so it controls caching/re-runs, not the output. skip_error, as everywhere in this pack, decides whether a failed job throws or silently returns an empty video.

Outputs: video (the upscaled VIDEO tensor - connect it to ComfyUI's native save/preview), video_url (the hosted result), task_id (so you can find the job in your Zhenzhen task history), and response (the raw JSON for debugging).

Installing and setup

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, and use ComfyUI Manager next time - search "Comfyui-zhenzhen". There are no model files to download; the node is pure HTTP. You need a Zhenzhen key from api.seedance.nz (the domestic low-price site; the overseas one at ai.t8star.org works too, just pricier). Plug it into the Seedance2 settings node rather than the node itself.

The honest caveats

This is a pay-per-use upscale, so it only makes sense when local is a pain - if you've got the VRAM for a local video upscaler, that's still the free-after-electricity option. And remember the usual relay gotchas: a 500 is often upstream and a re-run fixes it, the video leaves your machine, and "low price" is relative to other API upscalers, not to free. One nice touch: because the result comes back as a video tensor, you can chain it straight into another node - upscale, then run a local sharpening or a color pass, all in one graph.