Prompt guides
How to prompt each AI video model
Each model reads a prompt differently: VEO 3.1 wants sound written out, Hailuo 2.3 takes bracketed camera commands, Seedance follows numbered shots. One guide per model we run, written from how our studio sends the prompt and from real generations, with ten prompts you can open in the studio.
VEO 3.1 prompt guide
VEO 3.1 Fast, VEO 3.1 Lite · from 3 credits a clip
VEO 3.1 writes its own soundtrack, so a good VEO prompt describes what you hear as carefully as what you see: put the spoken line, the sound effects and the room tone in their own sentences after the picture. On Viralance a 4-second VEO 3.1 Lite test costs 3 credits ($0.90) and an 8-second VEO 3.1 Fast clip 12; Fast finished 98% of 149 real user jobs, the highest rate in our generation data.
Read the guideKling 3.0 and 4.0 prompt guide
Kling 4.0, Kling 3.0 Pro, Kling 3.0 Standard · from 8 credits a clip
Kling follows motion well when the prompt is written as a sequence of events: who is in the shot, then what happens first, next and last, then the sound. Kling 3.0 Standard starts at 8 credits for 5 seconds at 1080p, and Kling 4.0 runs up to 30 seconds (59 credits at 720p) and takes up to 8 keyframes between a start and an end frame. In our data Kling 3.0 Pro finished 28 of 28 jobs with no policy refusals.
Read the guideSeedance 2.x prompt guide
Seedance 2.0, Seedance 2.5, Seedance 2.0 Mini · from 5 credits a clip
Seedance reads a prompt like a short storyboard: a precise subject, its action and the scene first, then light, camera and style, and for longer clips a list of numbered shots with one camera move each. Seedance 2.0 Mini tests an idea from 5 credits for 5 seconds at 480p; Seedance 2.5 is the only model on Viralance with a 30-second single take (66 credits at 480p).
Read the guideHailuo 2.3 prompt guide
Hailuo 2.3 · from 3 credits a clip
Hailuo 2.3 is MiniMax's silent physics model: no audio track, 6 or 10-second clips (6 seconds only at 1080p), and one of the cheapest ways to animate a photo: 3 credits for 6 seconds at 720p (about $0.90). It reads MiniMax's bracket camera commands such as [Push in] and [Static shot], so the best prompts name one subject, one physical action and one bracketed move. In our data it finished 34 of 34 jobs with no content refusals, including all 22 jobs that started from a photo. The sample is small.
Read the guideMiniMax H3 prompt guide
MiniMax H3, MiniMax H3 Max, MiniMax H3 Max Turbo · from 2 credits a clip
MiniMax H3 (Hailuo 03) and its Max and Max Turbo variants make 5, 10 or 15-second clips with native sound, from 2 credits on Max Turbo at 480p to 8 credits for 5 seconds of 4K on H3. Write them a short scene with spoken lines in quotes and the sounds you want; all three expand your prompt before rendering, so a tight brief beats a long one. They refuse more inputs than most models in our data, mostly photos of real faces, so the fixes below matter.
Read the guideWan 3.0 prompt guide
Wan 3.0 · from 4 credits a clip
Wan 3.0 is Alibaba's newest video model on Viralance: 5, 10 or 15-second clips at 480p, 720p or 1080p with native sound, from 4 credits for a 5-second 480p test (about $1.20). Alibaba's own guide gives it a formula, entity plus scene plus motion, and writing in that order keeps a prompt from leaving gaps for the model to fill. It has no negative prompt field, so anything you do not want goes in the prompt as a plain sentence such as "No background music."
Read the guidePixVerse V6 prompt guide
PixVerse V6 · from 4 credits a clip
PixVerse V6 is the budget all-rounder for short, stylized clips: 5 or 8 seconds at 720p or 1080p, from 4 credits (about $1.20), with sound on every clip. It rewards prompts that read like an effect brief, one subject, one transformation, one style, and it powers many of our one-click selfie templates. In our generation data none of its 36 jobs was refused by the provider.
Read the guide
Prompt refused? See why AI video prompts get rejected. Comparing models first? See all AI video models.