🌸 OVERTLI Pollinations Video Generator

Video generation in ComfyUI usually means one thing: a local model eating your entire VRAM for minutes at a time. GZ_VideoGen skips that. It's the Overtli Studio Suite's video generator, and like the rest of the Pollinations cloud nodes it sends your prompt somewhere else and gets media back - in this case a real video, delivered on a native ComfyUI VIDEO output. If you've never generated video locally because your card can't hold a video model, this node is the version of the job that fits in a text box.

How it works

Pick a video-capable model from the dropdown (61+ entries tagged [video-gen], most [free], fetched live from Pollinations' catalog), type a prompt, run. The node calls the provider's video endpoint, waits for the render, and downloads the result - an mp4 that lands in a temp file - then returns it as a native VIDEO output the ComfyUI canvas can preview and save. The default model is a Qwen video-gen entry on the free tier, which keeps the cost story the same as the image node: free-tier models work with no key, and [paid] ones need gems.

The mechanism matters for two behaviors you'll notice immediately. First, this is synchronous: generation blocks until the provider finishes, which the README says is expected - the UI freezes, then resumes when the clip is ready. Second, there's a timeout_seconds knob (default 300, range 60–600) because provider-side video renders are slow, and the node ships temp-file housekeeping so old downloads don't pile up in your temp directory.

Inputs that matter

• prompt - describe motion, not just a scene. That's where mode_preset earns its keep: presets like Cinematic Video Prompt, Loop Video Prompt, or the Image-to-Video Motion Script give the model the structure it needs.
• model - the video-capable model. If a run fails, trying a different model is the first lever, since capability varies by provider entry.
• image - optional. Wire a frame or a generated IMAGE in and the video becomes an image-to-video job rather than text-to-video.
• width/height - 256–1280, default 512². These are caps; the provider may not honor large sizes on every model.
• style_preset_1..3 and additional_styles - style layers as everywhere in the suite.
• seed, enhance_prompt, api_key.

Output

One video socket. It's a real VIDEO tensor in current ComfyUI builds - you can preview it natively or save it like any video output. One genuine compatibility note: the native VIDEO output type needs a reasonably current ComfyUI core. If you're on an old build and the node shows up but its output socket looks wrong, update ComfyUI before debugging anything else.

Install and where it gets you

Same pack install as the other nodes: search "Overtli" in ComfyUI Manager, or clone into custom_nodes + pip install -r requirements.txt, restart. Featherweight deps - no video models, no weights, nothing to download beyond the pack.

The honest caveats mirror the image node's. Free-tier Pollinations video is a hosted service with a hosted service's reliability - slow renders and the occasional failure are the price of free, and the generous timeout_seconds default exists for that reason. Renders you make are your prompts and any reference image leaving the machine, subject to the provider's moderation. And for real, controllable, repeatable video work, a local model (Wan, LTX-Video) still gives you control this node structurally can't - this is a draft-and-ideate tool or a no-GPU workflow's video door, not a replacement for a video checkpoint.