Zhenzhen Google Veo3

Google Veo is the video model the open-source world keeps measuring itself against - and the one genuinely hard gap is native audio. Veo 3 generates synchronized sound, dialogue, and effects inside the video; there's no local Wan or LTX checkpoint that does that yet. Comfly_Googel_Veo3 is the Zhenzhen-pack wrapper that puts the whole Veo 3/3.1 lineup in your graph: twelve models across the dropdown, from the cheap veo3-fast to veo3.1-pro-4k, with image-to-video via up to three reference images. The video comes back as a VIDEO tensor, so it drops straight into a save or preview node - no URL-copying required.

This is the "closed model you literally cannot download" case, so the API wrapper isn't a compromise, it's the only door. The relay handles the Google quota and billing; you handle the prompt.

The inputs that matter

• prompt - your video prompt. enhance_prompt is the hidden gem: Veo prefers English prompts, so set this to true and it auto-translates and optimizes your Chinese (or messy) prompt into proper Veo English. Keep it false if your prompt is already clean English.
• model - the big dropdown. The base models (veo3, veo3-fast, veo3-pro) do text-to-video; the -frames and -components variants are the image-to-video ones (wire image1–image3), and the .1 family adds 4K tiers. veo3.1-components supports multi-image reference; veo3.1 handles first/last frame.
• aspect_ratio - 16:9 or 9:16. Landscape or portrait, that's it.
• enable_upsample - toggle the upscaler if your chosen model supports it.
• seed - reproducibility. Leave at 0 for random.

Outputs: video (VIDEO tensor → save node), video_url, and response (JSON).

Install and key

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/Comfyui-zhenzhen

Restart, or ComfyUI Manager → "Comfyui-zhenzhen". No model downloads - this is pure API over your Zhenzhen key from ai.t8star.org, pasted into apikey. Note the token-group thing: Veo rides the veo&grok token groups on the site, so make sure your key has one assigned or you'll get auth errors.

The honest part

Veo is not cheap, even relayed. The pack's changelog tracks the price war - veo3.1 dropped from a base of ~2 credits to ~0.3, and pro tiers run higher - but a batch of 4K clips is a real spend. The community's take on API video in general applies here: it's the right tool when you cannot run the model locally, and the wrong default if you could. Also, Google's video models are among the most heavily moderated - this node will not help you around that, and neither will any reseller, because the filter lives in the model. 500s from the relay are usually Google's capacity hiccups; re-run before you blame your graph.