Civitai Video Upscaler

Video upscaling is the least fun job in the AI pipeline and historically the worst supported. Local generative upscalers can barely hold a single image's identity together - do it across every frame and you get flicker, face drift, and a 45-minute job on a 5090. This node sidesteps all of it: it sends your clip to Civitai's fleet, runs the upscale there, and hands back a bigger VIDEO. Billed in Buzz, no local model, no VRAM.

Reach for it when you've got a generation you're happy with and you need it bigger - a 720p clip that needs to survive on a 1080 or 4K timeline, or a 540p output you want to rescue. The cloud angle matters more here than it does for most nodes, because video upscaling is exactly the job where your local card runs out of breath and every frame is a chance for the model to invent a new face. Offloading it to a fleet that can chew through the whole clip in one pass is a genuinely reasonable trade.

How it works

Standard pack plumbing: the node uploads your video, submits a videoUpscaler workflow to Civitai's orchestration API, long-polls until the job succeeds, and converts the result into a native ComfyUI VIDEO. Your source clip leaves your machine - that's the deal with every API-wrapper node, and it's worth remembering for footage you'd rather not ship anywhere.

The inputs that matter

This is a two-knob node, which is exactly right for the job:

• video (required) - the clip to enlarge. Wire in any VIDEO output, including the Civitai generation nodes.
• scale_factor - 2, 3, or 4 (default 2). Doubling is the sane starting point; 4x on a soft source just amplifies the mush.

Outputs: video (the upscaled clip), workflow_id (for cost/debug lookup), and raw_json (the full workflow response).

Install and auth

Install the whole pack once from ComfyUI Manager: Custom Nodes Manager → search Civitai Comfy Nodes → Install → restart. Or by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # just requests

Credentials resolve from a Civitai Auth node, the CIVITAI_API_TOKEN env var, a stored key/OAuth from the Civitai sidebar, or a browser login.

Where people get burned

Upscaling is "more pixels" or "more detail," and this node is firmly in the more-pixels camp - don't expect it to hallucinate pores and fabric weave the way a generative restorer would. The other trap is doing 4x on a source that's already soft: garbage in, bigger garbage out. Upscale once at 2x from your best render rather than trying to rescue a bad one at 4x, and remember the cost compounds with clip length - a 30-second 4x job is a lot of billed compute.