🎨 Hrio Design｜单视频生成

Hrio_Design_Single_Video_Node (🎨 Hrio Design｜单视频生成) is the video sibling of the pack's single image node, and it behaves exactly like a cloud API node should: you ask for a clip, the node blocks while the backend generates it, and you get one thing back. In this case that one thing is a STRING holding a local video path - the tooltip tells you it's directly previewable, which is a nice touch for a video output.

There's no GPU involved and no model download. The video is generated server-side through the same private Gemini-format backend the rest of the pack uses, with Google's veo3.1 as the only model option. This is a pay-per-call, inputs-leave-your-machine setup; it's a proposal tool, not a local sampler.

How it works

Video generation is async under the hood. The node kicks off a long-running job on the backend (a predictLongRunning-style operation), then polls it every few seconds (the config defaults to an 8-second interval) until the clip is done or a 30-minute timeout is hit. When the job finishes, the pack saves the video locally so the returned string is a real path you can preview, not just a URL.

That means one honest gotcha: this node is slow. A video job can legitimately sit there for minutes. Set it running, walk away, come back - it's not hung, it's rendering a Veo clip on someone else's GPU.

The inputs that matter

• prompt - your video prompt. The default is a reasonable Chinese brief (stable camera, restrained motion, natural light, material detail) for a design-proposal showreel; replace it with your actual brief.
• video_model - a dropdown with one choice right now: veo3.1. The tooltip notes it falls back to whatever the remote manifest/config specifies, but today it's Veo.
• api_key - leave blank and it reads [banana] api_key from config.ini. Same story as everywhere else in this pack: the README never explains how to obtain one, so the author is the gatekeeper.
• video_resolution - 1080p (default) or 720p. At 1080p, budget your time accordingly.
• aspect_ratio - 16:9 widescreen or 9:16 vertical/phone. For interior-design and poster proposals, 9:16 is often the deliverable that actually gets used.

There are ten optional reference-image slots (image_1…image_10), but read the tooltips carefully: the current backend actually sends at most four - the first image becomes the image, and the next three become referenceImages. Slot five and up are decorative until the backend grows.

Output

A single video output of type STRING - the local path to the rendered clip. That's it. If you want the raw URL and a machine-readable info payload as well, use the pack's Hrio_Design_Video_Node instead; this one is the minimal version.

Installing it

Same pack, same install. ComfyUI Manager search for Hrio-Design, or:

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

Restart ComfyUI afterward. Dependencies are just requests, numpy, and pillow; nothing heavy, nothing to download beyond the pack itself.

Where people get burned

The reference-image cap is the thing most likely to bite - you can wire in ten images and the backend silently only uses four. And remember this node's whole job is to phone home with your prompt and images: check verify_ssl = false in config.ini if the idea of a third-party server handling your traffic bothers you. There's no real community footprint for this pack on Reddit, so if a job times out, the knobs to turn are the veo_poll_interval_sec and veo_poll_timeout_sec values in the config.