Enhance Ultra · 视频增强（大模型版）

Where the plain Enhance node cleans up, Enhance Ultra generates. It's the diffusion-model version of video enhancement: instead of just sharpening and denoising, the cloud model actively fills in missing detail and does generative restoration. For old film, heavy compression artifacts, and footage where the texture is simply gone, this is the heavyweight you reach for - and the one most likely to surprise you, in both directions.

The inputs (short list, honest limits)

• resolution - 720p or 1080p, and that's it. No 4K output here. That's not a bug; it's the honest ceiling of generative video restoration today. If you need to land on 4K, this isn't the tool.
• bitrate_level - low / medium / high target encode.
• fps - 0 keeps the source frame rate; if you specify a value it must be within 15–120 fps (the tooltip is explicit).
• video, plus the usual poll_interval_seconds (default 10) and max_poll_attempts (default 720).

The single output is video, wired into a Save Video node: Load Video → Enhance Ultra → Save Video.

How it works

Identical plumbing to every other node in the pack - materialize the video to a temp file, submit an async cloud task via mediakit-cli, poll, download, convert back to a native VIDEO. The difference is the capability under the hood: a diffusion-based generative model runs the restoration server-side, which is why it can put texture back where a classical enhancer would just smooth. It's also why it's the most expensive capability in the pack - you're paying for actual inference in the cloud.

Install and the cloud setup

Via ComfyUI Manager, search "MediaKit Toolkit" / mediakit-toolkit, fully restart ComfyUI - or:

cd ComfyUI/custom_nodes
git clone https://github.com/fudanll-sys/comfyui-mediakit-toolkit.git

No Python dependencies, no model downloads - the weights live on Volcengine's side, not yours. You need a Volcengine account with AI MediaKit activated, an API key, and the CLI installed and authed in the environment that launches ComfyUI:

npx @volcengine/mediakit-cli install -y
mediakit-cli init --mode cloud-first --api-key "YOUR_MEDIAKIT_API_KEY" --credential-store config --yes

The honest comparison and where people get burned

The open alternative for this exact job - generative video restoration - is SeedVR2, which is excellent but needs datacenter-grade hardware for real video; that's the gap this node fills, and it fills it from a Chinese cloud you may already have an account with. The failure modes are the ones generative models always have: hallucinated texture on faces and small text, and a bill that climbs with clip length. Generative restoration is the kind of thing you feel on old footage and barely notice on clean modern clips, so don't run it on footage that's already fine. Test a 10-second worst-case clip first, eyeball the faces and text, and only then commit the whole project.