🎉RH视频超清@炮老师的小课堂

Every video workflow hits the same wall eventually: the model spits out 720p and you need something that survives a big screen. Local upscalers exist and they're good, but they eat VRAM and a whole separate pass. DapaoRHVideoEnhanceNode ("🎉RH视频超清@炮老师的小课堂") is the cloud shortcut: feed it a video, pick a target resolution, and RunningHub's video-upscaler model does the upscaling on their GPU. Optionally it chains a second pass - video-fps-increaser - to bump the frame rate on top. Your card never breaks a sweat, and you never install a thing beyond the pack.

What it does, concretely

The two controls that matter are right at the top:

• 🧩 超分目标 - 不启用 (off), 720p, 1080p, 2k, or 4k. If you leave it at 不启用, the upscale stage is skipped entirely.
• 🎞️ 帧率增强 - 不启用 or 开启增强. When on, it runs after the upscale, so the order is always upscale first, then frame-rate boost.

Input is a 🎬 视频 socket, or if your source is hosted somewhere, the 🌐 视频公网URL string - nice for clips that are already on a server. 📋 额外参数JSON lets you push arbitrary fields through, and the 🚫 跳过错误 toggle keeps a failed clip from killing a whole batch queue.

Outputs are the pack's standard quartet: 🎬 视频 (a real VIDEO you can connect to a save node), 🆕 任务ID, 📋 响应信息, and 🔗 视频URL. Because this is an async job, you'll also see the poll controls (🔄 最大轮询秒数, ⏱️ 轮询间隔, ⌛ 请求超时) - video upscaling genuinely takes minutes, so don't set the poll window to 60 seconds and wonder why it times out. Defaults of 1800s max / 5s interval are sane.

Install

Same as every node in this pack. ComfyUI Manager → search dapaoAPI → Install → restart, or:

cd ComfyUI/custom_nodes/
git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git
cd ComfyUI-dapaoAPI
pip install -r requirements.txt

Restart, then drop in your RunningHub API key (国内版/国外版 channels use different keys - keep them matched). There are no model downloads, because there's nothing local to download.

Where this fits - and where it doesn't

Honest framing: if you already have a solid local upscale chain (and there are genuinely good ones), this node is a convenience, not a leap. Where it earns its keep is (1) no GPU headroom for a local upscale pass on top of video decode, (2) 4K targets that local chains would take forever to hit, and (3) batch jobs where you'd rather pay a few cents per clip than babysit VRAM. The tradeoffs are the usual cloud ones: per-clip cost, your video leaving the machine, and the quality is whatever RunningHub's upscaler gives you - you don't get to tune it the way you tune a local model. For a quick "sharpen this before I upload it" job, it's about as frictionless as it gets.

One gotcha from the pack's design: this is a pay-per-call node and the README's own framing is that the author focuses maintenance on the RunningHub and first-party dapaoAI channels - other integrations have been left behind. So keep the pack updated (Manager handles it), and don't be surprised if an endpoint bumps its contract. If you hit an error, the 响应信息 output usually says exactly which stage failed, which beats staring at an empty preview.