# Veo 3.1 Lite vs. Fast — Comparison Checklist

Companion to Chad Sheppard's video: https://www.youtube.com/watch?v=BHwtppZHRHk
Guide: https://chadesheppard.com/veo-lite-vs-fast

## Match the inputs
- [ ] Save the same prompt for both models.
- [ ] Match duration, resolution, aspect ratio, and audio settings where available.
- [ ] Use the same starting image for image-to-video.
- [ ] Record the provider, model, date, and price shown before generating.
- [ ] Decide how many attempts to allow; keep failures in the record.

## Inspect each full clip with sound
- [ ] Cinematic realism: wheels, reflections, suspension, camera movement, road contact.
- [ ] Human/object interaction: hands, mug placement, object continuity, action order.
- [ ] Dialogue: exact words, correct speaker, mouth synchronization, accessories, audio.
- [ ] Image-to-video: identity, starting-image consistency, requested movement, unwanted additions.
- [ ] Restaurant stress test: body contact, tray stability, action order, speaker attribution.

## Record your own results
| Test | A or B preference | Model revealed | Usable? | Cost | Mistakes |
| --- | --- | --- | --- | --- | --- |
| Cinematic realism | | | | | |
| Human movement | | | | | |
| Dialogue and audio | | | | | |
| Image to video | | | | | |
| Restaurant stress test | | | | | |

Total spend: ____  Usable clips: ____
Cost per usable clip = total spend divided by usable clips. If none are usable, record that instead of dividing by zero.

## What Chad observed
One generation per model per test, no rerolls or prompt tweaks. Chad knew the outputs; the A/B reveal was blind only for viewers.
Recorded costs: Lite $0.40; Fast $0.80 per generation. Not a guarantee of current prices across providers or settings.
Chad preferred Fast for realism, barista interaction, dialogue (close), and the restaurant scene (both poor). Image-to-video was a tie: Lite followed instructions better; Fast looked better but morphed.

Companion prompts in the guide reconstruct the script's descriptions; they are not verbatim original prompts. Use your own starting image for the fourth test.