veo3

Veo 3 is Google's closed video model, and the one thing no open model had matched for a long time: it generates sound natively with the picture - dialogue, footsteps, ambience, synced. This node puts that behind a ComfyUI canvas via Civitai's cloud. You type a prompt, optionally hand it a first frame and an audio choice, and the job runs on Civitai's fleet, billed in Buzz. Nothing runs on your GPU, nothing gets downloaded.

That native-audio capability is the honest reason to reach for Veo over the other video nodes in the civitai-comfy-nodes pack. If your clip needs synchronized audio and you don't want to bolt a separate TTS/foley pass onto a silent video, this is the node. Everything else in this pack makes silent movies; this one doesn't have to.

The inputs that matter

Two are required:

• prompt - the multiline prompt. Veo responds well to concrete camera and action language.
• generate_audio - and here's the interesting part: it's a required checkbox, default false. You must deliberately switch it on to get sound. It's the one required input in this pack that's a choice, and flipping it on is the whole reason to pick this node. (Audio also costs more, so the default-off makes sense.)

The optional stack:

• version - 3.0 or 3.1 (empty = current default). Veo 3.1 is the improved release with better image-to-video and finer audio control; if you're here for the audio, try 3.1 first.
• mode - standard, fast, or lite (plus empty = default). These trade quality for speed/cost the way you'd expect.
• duration - 4, 6, or 8 seconds. Short by design; Veo's clips are premium-per-second.
• aspect_ratio - 16:9, 9:16, or 1:1.
• images - wire a LoadImage in for image-to-video.
• fast_mode - a separate quick toggle, off by default. Different axis from mode; flip it when you're iterating and want results sooner.
• negative_prompt - useful for steering away from artifacts.
• enable_prompt_enhancer - on by default; it rewrites your prompt server-side. Turn it off when you've written exact language.
• seed, api_config - the usuals.

Outputs

Standard pack trio: video (VIDEO) into a Save Video / preview node, workflow_id (STRING) for sidebar tracking, raw_json (STRING) for the response and per-run cost. If you turned audio on, it's in the video file - the node returns a single VIDEO regardless.

Install and auth

Pack-wide, one-time. ComfyUI Manager → search Civitai Comfy Nodes → install, restart:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # just `requests`

Auth: Civitai Auth node, CIVITAI_API_TOKEN, or the sidebar sign-in. Env var is the headless route.

Troubleshooting and takes

• "Why is my video silent?" - generate_audio defaults to false. It's the single most common gotcha on this node, and the fix is one checkbox.
• Cost. Veo is premium-tier: per-second billing, and adding audio raises the meter. Draft at fast/lite and 4 seconds, then spend on the keeper.
• Google's filter applies. Veo's content policy is enforced at the model - the node can't and won't bypass it. If a prompt refuses, that's Google's call, not a fault in your graph.
• Slow queues. The pack launched June 2026 during Civitai's GPU crunch; long waits and the occasional timeout were the norm. The default timeout is 30 minutes - raise it via the Civitai Auth node's timeout_minutes if you're running long jobs.
• Early preview. Nodes are generated from Civitai's API spec and "may change without notice." After a pack update, re-check the fields - version and mode lists tend to grow as Google ships.

Reach for this node when the clip needs to sound real, or when you want Google's top-tier motion on a still you already love. When you just need a quick silent draft, the cheaper engines in this pack will do.