🎨 Hrio Design｜视频生成

Hrio_Design_Video_Node (🎨 Hrio Design｜视频生成) is the "full receipt" version of the pack's video generation. Where its sibling Hrio_Design_Single_Video_Node hands you just a local path, this one returns three strings - info, video, and mp4url - which makes it the one you reach for when you want to wire the result into something that needs the metadata or the actual remote URL, not just a file on disk.

It's still a cloud API node through and through: nothing runs locally, Google's veo3.1 is the model, and your prompt and reference images go to the author's private Gemini-format backend. Pay per call, data leaves the machine, no GPU required. If that tradeoff already ruled out the single-video node, it rules out this one identically - they differ in output plumbing, not in mechanism.

How it works

Same async video pipeline as the pack's other video node. The node submits a long-running job to the backend, then polls it every veo_poll_interval_sec (default 8 seconds) until it completes or veo_poll_timeout_sec (default 1800 seconds - half an hour) elapses. On success it saves the clip locally so video is a directly previewable path, and exposes the hosted mp4 as mp4url. On failure it logs an error and returns accordingly. The upshot: this node is slow by design, and a 1080p Veo clip can sit there for several minutes.

The inputs that matter

• prompt - the video brief. Default is a Chinese design-proposal showreel prompt (stable camera, controlled motion, natural light); swap in your own.
• video_model - a dropdown with one entry, veo3.1, which the tooltip says falls back to whatever the manifest/config defines.
• api_key - blank falls back to config.ini's [banana] api_key. As with the rest of the pack, the README doesn't say how to get a key - the author controls that.
• video_resolution - 1080p (default) or 720p.
• aspect_ratio - 16:9 widescreen or 9:16 vertical.

Ten optional reference-image slots (image_1…image_10) are declared, and this node's tooltips claim support for all ten - though given how closely the two video nodes track each other, don't be surprised if the backend still only honors the first few. Test with one or two before you build a workflow that depends on ten.

Outputs

• info - a STRING payload with run metadata; useful if you're scripting or logging runs.
• video - a STRING local path to the rendered clip, directly previewable.
• mp4url - a STRING remote URL for the hosted file, handy when you need to hand a link to a client or push it somewhere.

Installing it

The whole pack installs the same way, since all five nodes ship together. ComfyUI Manager search Hrio-Design, or:

cd ComfyUI/custom_nodes
git clone https://github.com/moki-flank/Hrio-Design

Then restart ComfyUI. Only requests, numpy, and pillow are required, and there are no model files to fetch - the pack does everything over HTTP.

Where people get burned

The classic trap is treating this like a local video node and expecting instant results; budget for a long-running job and the half-hour poll timeout. And because it's an API node whose whole purpose is phoning home, remember verify_ssl = false lives in config.ini, and that your prompt and images are being processed on a server you don't run. The pack has effectively no Reddit presence, so don't expect community troubleshooting - the config knobs for polling are the main levers you have.