🎉RH全能视频V3.1@炮老师的小课堂

DapaoRHAllVideoV31Node wraps RunningHub's "全能视频 V3.1" (all-capable video V3.1) model family - the more recent sibling of the Seedance node in this pack. Where the Seedance node is one model with a few modes, V3.1 is a five-mode node: text-to-video, image-to-video, first-frame-to-last-frame (首尾帧), reference-based video generation, and video extension. It shares the RunningHub plumbing with its siblings (same 国内版/国外版 channel split, same async poll loop), but it's the one you reach for when the job is "extend this existing clip" rather than "make a clip from scratch."

The inputs that matter

• 🌐 API渠道 + 🔑 API密钥 - the eternal RH rule: domestic vs international accounts, keys don't transfer.
• 🤖 模型 - three entries, V3.1-FAST default, plus the standard/quality variants.
• 🏷️ 渠道 - 官方稳定版 (default here; the README's framing is quality-first for video, since a bad video costs a full re-render) vs 低价渠道版.
• 🎛️ 功能 - the mode dial: 文生视频, 图生视频, 首尾帧, 参考生视频, 视频扩展.
• 🧩 分辨率 - three options, 720p default. ⏱️ 时长(秒) - three options, 8s default. 📐 视频比例 - 16:9 default. 🔊 生成音频 - false by default (unlike the Seedance node, which defaults true).
• 🎲 随机种 - default −1, passed through in some flows like its Seedance sibling.
• Optional inputs are mode-dependent: 🎬 首帧图 and 🏁 尾帧图 (for 首尾帧), 🎞️ 输入视频 (for 视频扩展 - the clip you're extending), 🖼️ 参考图1–3 (for 参考生视频), public-URL variants of each, a 🚫 反向提示词, plus 📋 额外参数JSON and the polling trio (max 1800s, 5s interval, 120s request timeout).

Outputs

🎬 视频 (VIDEO tensor → save directly), 🆔 任务ID, 📋 响应信息, 🔗 视频URL. No 尾帧图 output here - that's the Seedance node's party trick.

How it works

Submit the appropriate endpoint based on 功能, poll the task, download, decode. The video-extension mode is where this node earns its place in a real workflow: you generate a 5s clip, feed it back in as 🎞️ 输入视频, and get a longer take - the "video in, longer video out" pattern that a growing number of users want but few one-node wrappers actually expose. It's the difference between a wrapper that mirrors a vendor's product page and one that matches how people actually chain generations.

The honest read

This is a thin-but-competent client: the five modes are genuinely there, the channel/quality split is a real choice, and the defaults (FAST model, official-stable channel, 720p/8s) are a sensible "don't make me think" starting point. The tradeoffs are the whole genre's - metered billing, data leaves your machine, availability depends on both RunningHub and the author's maintained endpoints. And because video is where costs bite hardest (external-api-nodes.md is blunt about it), the "what mode did you actually pick" discipline matters: 视频扩展 with a long input can bill like a much bigger job than a fresh 5s clip.

Gotchas

• Key mismatch between channels - the recurring RH trap, still true here.
• Generate-audio defaults differ per node. This one defaults off; the Seedance node defaults on. If you silently expect audio and get silence (or vice versa), check the toggle, not the platform.
• Mode-dependent inputs are easy to wire wrong - 首尾帧 wants both frames; 视频扩展 wants the video. Miss one and the request fails or silently drops the input.
• Seed is −1 by default and may reach the API; don't assume the cache-only rule.
• Chinese-only labels.

Install: ComfyUI Manager → "dapaoAPI", or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git → pip install -r requirements.txt → restart. RunningHub account + API key required (README invite code kol01-rh138 is the author's referral).