🎉RH全能视频X-video3@炮老师的小课堂

DapaoRHAllVideoXVideo3Node is the third video node in the RunningHub section, and it fills a different role from its siblings: where Seedance 2.0 is the headline generator and V3.1 is the five-mode utility, X-video3 leans editing - video editing, multi-image reference, and video continuation (续写). Same RunningHub plumbing (async submit + poll, 国内版/国外版 channel split), but the mode set and the up-to-seven reference-image inputs tell you this one is about changing what you already have, not just making it.

The inputs that matter

• 🌐 API渠道 + 🔑 API密钥 - the standing RH rule: domestic/international keys don't cross.
• 🤖 模型 - two entries, X-video3 default. 🏷️ 渠道 - 官方稳定版 default.
• 🎛️ 功能 - five modes: 文生视频, 图生视频, 视频编辑, 多图参考 (multi-image reference), 视频续写 (video continuation).
• 📐 视频比例 - five options, 16:9 default. 🧩 分辨率 - two options, 720p default. ⏱️ 时长(秒) - 18 options (the widest duration menu in the RH video set), 6s default.
• 🚫 出错时跳过 - a skip-on-error toggle at the required level (the other RH nodes bury it in options or omit it) - a hint that this node expects batch-ish use.
• 🎲 随机种 - default −1, may reach the API.
• Optional inputs: 🎬 首帧图 (for i2v), 🎞️ 输入视频 (for 编辑/续写 - the clip being edited), 🖼️ 参考图1–7 (the multi-image reference set), public-URL variants, 🎞️ 参考视频时长(秒) (default 6 - how long the reference video's source context is), 📋 额外参数JSON, polling trio (1800s max, 5s interval, 120s timeout).

Outputs

🎬 视频, 🆔 任务ID, 📋 响应信息, 🔗 视频URL - the standard RH video quartet.

How it works

Identical async machinery to its siblings: build the right endpoint from 功能, submit, poll, decode. The distinguishing bits are the mode surface and the 多图参考 path - feed several reference images (up to 7) plus a 输入视频, and the model edits the video to match the visual references. That's a "character-consistency across a clip" or "restyle this footage" workflow, which is genuinely the kind of thing video editors keep wanting and API video nodes rarely expose this directly.

The honest read

X-video3 is the most utility-shaped video node in the RH set - less glamorous than Seedance, but the mode set (especially 视频续写 with 参考视频时长) is what you actually use when you're assembling something longer than a single prompt's worth of video. The 18-option duration menu and the skip-on-error toggle say the author expects this to run in loops, which matches how video pipelines actually get built: generate a few takes, keep the good ones, extend them.

The genre's standing caveats all apply - metered per-job billing where video bills the hardest, your clips leaving your machine to RunningHub's servers, availability riding on both the platform and the author's maintained endpoints, and an ad-heavy README that wants you to use an invite code (the author's referral, kol01-rh138). external-api-nodes.md puts the community's split well: API video is the right tool for models you cannot run, and the wrong default for ones you can - X-video3 isn't runnable locally, so you're on the right side of that line if you're here.

Gotchas

• Key mismatch between channels - still the #1 RH support question.
• 编辑/续写 need the input video wired. The mode list tempts you to click 视频续写 and forget 输入视频; the request then fails or runs without context.
• 参考视频时长 governs how much source the model sees - a long reference with a short duration setting can truncate the source silently.
• Seed default −1 may be passed through - don't assume cache-only.
• Chinese-only labels.

Install: ComfyUI Manager → "dapaoAPI", or git clone https://github.com/paolaoshi/ComfyUI-dapaoAPI.git → pip install -r requirements.txt → restart. Needs a RunningHub account + API key.