WaveSpeedAI Google VEO 3.1 Text-to-Video

The standard tier of Google's Veo 3.1, as a ComfyUI node. If the "Fast" variant in this pack is the iteration loop, this is the keeper: the full-strength text-to-video model, prompt in, hosted clip out, from razvanmatei-sf/razv-wavespeed against the WaveSpeed AI API. You use Fast to explore, then this one to actually render the shot you're committing to.

Veo 3.1's text-to-video is the current public benchmark for "type a paragraph, get a coherent clip": camera moves that make sense, physics that mostly behaves, and a strong feel for cinematic language. It still can't run locally - no weights exist - so every option is hosted. This node is just the friendliest front end you'll get for it inside a graph.

How it works

WaveSpeedAI Client → prompt → POST to the Veo 3.1 text-to-video endpoint → poll until done → video_url string out. Nothing is written to disk automatically; grab the URL or pipe it to a downloader.

The dials are identical to the Fast tier, and you should know what each one buys you:

• aspect_ratio - 16:9 or 9:16.
• duration - 4, 6, or 8 seconds.
• resolution - 720p or 1080p.
• generate_audio - native synchronized audio, off by default.

The real difference from Fast is subtle and important: the standard tier spends more compute per clip, which shows up as better motion coherence and fewer weird artifacts. It also costs more and takes longer. That's the trade, and it's why you want both nodes - Fast to iterate, standard to deliver.

The inputs that matter

• prompt - the whole game. Video prompts reward motion verbs and camera grammar: "slow push-in, golden hour, dust motes in the light." Describe the shot, not just the scene.
• duration + resolution - the price/speed dials. 8s 1080p is the expensive end.
• generate_audio - flip on if the deliverable needs sound.
• negative_prompt - genuinely useful here; "no text, no watermark, no people" keeps the model from doing those things it loves to add.

Install

Shared pack:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/razv-wavespeed
# restart ComfyUI

Or ComfyUI Manager → search "razv-wavespeed" → Install. Add a WaveSpeed AI key via the WaveSpeedAI Client node, config.ini (copy from config.ini.tmp, [API]), or a WAVESPEED_API_KEY env var.

Common issues

• This is the pricey one. Standard-tier video is the most expensive thing in the pack per second of output. Iterate on Fast or 720p; commit to 1080p.
• Output is a URL, and URLs expire. Download results promptly.
• Queue time is real at peak hours; async mode (the default) is the right call so your graph isn't frozen.
• Guardrails. Veo is censored at the API. Not a bug - a policy.