Protoface Veo 3.1 Lite · Generate Video

Cheap Veo is the useful Veo

The knock on hosted video is that you can't afford to be bad at it. Every experiment is a charge, so people write one careful prompt, queue it, and then sit still - which is the opposite of how you get good footage out of any model. Iteration is where the quality comes from.

Lite exists to fix that. It's Veo 3.1 at a lower price point, still with image control and native audio, still with the prompt-following that makes Veo worth using. Fewer pixels, fewer seconds, less money: a draft tier.

This is the general Generate Video task in Protoface, the beta pack that turns hosted model APIs into native ComfyUI nodes. Paid per run, remote, no weights. The workflow it invites is obvious and correct: block out the shot on Lite until the motion is right, then re-render the keepers on full Veo 3.1 or a local model.

How it works

Schema-derived controls → validated payload → uploaded media → journaled request → submit → poll → download as native VIDEO. Same shared adapter as every node in the pack, which is why a Google node and a Kuaishou node have the same rhythm on the canvas.

generation_id is the paid-request identity, distinct from seed. Leave it fixed to reuse or resume the run you already paid for; increment to spend on a new one. Editing the prompt makes it a new request whatever the id says.

The inputs that matter

prompt - the shot. Lite's realism budget is smaller, so be concrete: one subject, one action, one camera move. Cram four ideas into a Lite prompt and you get a soft mush of all four.

reference is the image control, and it's optional in practice (media-style, defaulting to omitted). Connect a native IMAGE from your graph, or use References mode with an asset id. Starting from a frame is the single biggest quality jump available on this node.

quality defaults to 720p. The tooltip hints at what's going on: some tiers are upscaled rather than native, and the output descriptions tell you which - worth reading once, because paying for a bigger number that's really an upscale is a boring way to waste money.

duration_seconds is coupled to quality. The author's own constraint is listed right in the field: 720p accepts 4, 6 or 8 seconds; 1080p is 8 only. Mixing them up is the most common validation error on this node.

enhance_prompt (off by default) rewrites your prompt into the provider's preferred form. Handy when you're stuck; annoying when you've crafted a line and it gets "improved" into something else.

generate_audio, aspect_ratio (16:9 default), negative_prompt and seed round it out. Audio on by default; the neg prompt and seed are media-style controls where Default means the field isn't sent at all.

Outputs

video (native VIDEO, list), files (PROTOFACE_FILE list of originals), and run id. Save the id in your notes when you're comparing takes - it's also the only way back into a job that finished after your session ended, via Protoface Resume Run.

Install

Manager → search Protoface → install protoface-nodes → restart → reload the browser. Manual:

git clone --branch v0.1.0 --depth 1 https://github.com/protoface-ai/protoface-comfyui.git custom_nodes/protoface
python -m pip install -r custom_nodes/protoface/requirements.txt

ComfyUI 0.41.0+ and Python 3.10+, installed with the interpreter your ComfyUI actually runs on. Then click Protoface in the lower-right corner and enter a key from your Protoface account - separate from Comfy credits, which is the thing people email support about most.

Where people get burned

Duration/quality mismatches. 1080p only offers 8 seconds on this node. If the duration combo refuses to hold your value, that's why.

Expecting Lite to be Veo. It's the same family, not the same render. Big motion, fine hands and text-on-screen are where the gap shows. Judge it as a draft tier and you'll be happy; judge it against the flagship and you won't.

Queueing a reroll without bumping the id. You'll get the recorded job back, which feels like the node ignoring your change. It isn't - increment generation_id for a new generation.

Forgetting the key after a restart. Interactive keys live in server memory only. Set PROTOFACE_API_KEY_FILE on a headless machine, and remember that anyone who can queue workflows on a shared instance can spend on it.