The AI Video Maker takes a single sentence and gives you back a finished shot. This guide walks through every control in the composer — the ones you should usually touch, and the ones to leave alone until you know what you want.
The model picker
Lumen ships with five video models, each with a personality:
- Veo 3.1 Lite — the cheap, fast default. Great for drafting until your prompt feels right.
- Veo 3.1 Pro — cinematic 1080p with native synced audio. Pick this for hero shots.
- Sora 2 — the strongest physical-world simulation. Pick this when objects need to behave correctly.
- Runway Gen-4 — best when you want granular motion control and editor-style workflows.
- Kling 3 — the best price-per-second for long takes.
If you don't know which to pick, draft on Veo 3.1 Lite, then re-render the keepers on Veo 3.1 Pro or Sora 2 once your prompt is right. Drafts are roughly 5× cheaper.
Writing prompts that land
A useful AI video prompt has three parts:
- Subject — what is in the frame. Be specific. Not “a cat,” but “a tabby kitten with one folded ear.”
- Action — what they're doing. “Yawning slowly” reads completely differently from “stretching its back.”
- Cinematography — the look. Lens (“35mm”), light (“golden hour, soft rim light”), motion (“slow dolly in”), and style (“painterly, soft Ghibli watercolour”).
An example prompt
A tabby kitten with one folded ear, yawning slowly on a windowsill at golden hour, soft rim light, dust motes drifting, 35mm shallow depth of field, painterly Ghibli watercolour style.
Aspect, duration, resolution
- 9:16 for TikTok, Reels, Shorts.
- 16:9 for YouTube and landscape ads.
- 1:1 for feed posts.
- 21:9 for cinematic intros (Veo 3.1 Pro only).
Duration affects coherence: 4-second clips are nearly always coherent. 8 seconds becomes risky on prompts with many moving parts. 12 seconds is where you need to lean on Kling or Sora 2.
Resolution maps cleanly to credits: 720p costs roughly half what 1080p does. Draft in 720p, finalise in 1080p.
Start and end frames
Tap the Start frame slot to pin the first frame of your video to an image from your camera roll. This is the single best trick in Lumen — uploading a still image essentially gives you image-to-video, which is far more controllable than text alone.
End frame is rarer but powerful: combine start and end frames to specify exactly where the camera needs to begin and end.
Audio mode
Veo 3.1 ships with native audio — ambience, music, or even dialogue if your prompt suggests it. The audio toggle in the composer turns this on or off:
- Off — silent video, mix sound later (recommended for ads).
- Ambient — natural soundscape (wind, room tone).
- Music — generates a fitting score in the model's voice.
- Dialog — synthesises speech for any quoted text in your prompt.
Seed and variations
If you find a generation you like and want close variations, tap the dice to lock the seed and bump variations to ×2 or ×4. Same prompt, same seed, slight differences — like negatives in film photography.
A 60-second workflow
- Open Image & Video.
- Tap the green Video tab.
- Type a 1-sentence subject with one cinematography note.
- Pick 9:16, 4 seconds, 720p, Audio off.
- Tap the up-arrow.
- Iterate the prompt 2-3 times until you like it.
- Lock the seed, bump to ×4, re-render on Veo 3.1 Pro at 1080p.
That's it. Five minutes from idea to shareable.