Concurrent Submit | zhenzhen-upscaler-lowprice

Here's the trick this node plays on you: it's an upscaler, and it runs in the video pool. Check the node description - "shared video pool (10)" - and the output type is COMFLY_VIDEO_FUTURE, not the image future. The pack's own categorization treats upscaling a video as video work, which is the right call and the easiest thing to get wrong.

zhenzhen-upscaler-lowprice sends a video to the Zhenzhen API and gets a higher-resolution video back: 720p, 1080p, 2k, or 4k. If you've generated a clip at 720p and want a 4k deliverable without rebuilding the whole scene, this is the node. Your GPU does none of the work - the upscale is a cloud model call, billed per job.

The upscaling-KB lens is worth keeping in mind: cloud video upscalers are mostly more-pixels jobs, and "more pixels" is the one category where local tools (NVIDIA VSR, ESRGAN-family) are genuinely competitive and free. What the cloud version buys you is convenience and zero VRAM. If you have the GPU headroom, this node is a convenience tax; if you're on a laptop or a shared box, it's the difference between shipping 4k and not shipping at all.

Inputs and outputs

• video_url - a public MP4 URL. The tooltip is explicit: leave empty when input_video is connected.
• input_video - a VIDEO tensor from your local graph. Connect this and skip the URL. Pick one - both connected is ambiguous and the node will take the tensor path.
• resolution - 720p, 1080p, 2k, 4k. Note there's no "up by X" - you pick a target tier. Upscaling to 720p from a 480p source is pointless; you're here for the 2k/4k tier.
• api_config - the standard Seedance2 lowprice settings node (domestic key).
• skip_error - the pack-wide toggle: on, the node returns a blank instead of raising when a job fails.

Output is a task (COMFLY_VIDEO_FUTURE). Wire it into Concurrent Collect Videos (10) - and this is where the pool size catches people. Ten video workers, not thirty. The upscaler queues with your video generations, and video jobs are long. A full 10-slot queue of 4k upscales is a long wait, by design.

Install

ComfyUI Manager → search "Comfyui-zhenzhen" → install → restart, or:

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

No model downloads. API key from the domestic store (api.seedance.nz) required.

Where people get burned

• Wrong collector. Feed a video task into the Image collector and you get a type error. This node pairs with Concurrent Collect Videos (10), period.
• Upscale isn't free and isn't magic. Video jobs are the pack's most expensive category, and 4k costs more than 1080p. Upscale the clip you're actually keeping, not every draft.
• Input video size matters. Feeding a 480p clip to a 4k upscale asks the model to invent a lot. In the upscaling-KB terms: "more pixels" over a soft source can amplify artifacts. If the source is genuinely low-res, clean it up first, or accept the painterly result.
• Downloads can be big. 4k video across the border was flaky enough that the author added OSS settings (an object-storage endpoint picker on the site) for exactly this problem. If a large output download keeps failing, check the ComfyUI terminal for the actual download URL and grab it directly.

The honest read: this is the "I have no GPU headroom and need a 4k deliverable" node, and it does that job cleanly. If you do have the VRAM, run SeedVR2 or an ESRGAN tile pass locally and keep the credits for something the cloud genuinely can't be replaced on.