Wan 3.0 Video

long clip prompt structure

Stage/end-state prompting hit the stated endpoints more clearly than a run-on sentence in PixelDojo's long-clip guide.

Reported result: Stage/end-state version hit stated endpoints; run-on version was muddier and re-handled the object.

Prompt: 20s bicycle-wheel truing, three stages/end states vs run-on sentence

Settings: matched pairs: 2 · duration seconds: 20

timestamp adherence

A timestamped 12-second plate-to-glass-to-green-vase prompt completed its three four-second beats in order.

Reported result: All three beats played in order and cleared before the next.

Prompt: 12s locked-off plate→glass→green vase in three 4s ranges

Settings: duration seconds: 12 · matched pairs: 1

first frame drift

The opening frame matched a subject leaning at a post, but the later subject left the post and the camera moved behind him beyond the request.

Reported result: Opening frame matched; later subject left post and camera moved behind him beyond request.

Prompt: First-frame subject should stay leaning at post and only turn head

Settings: matched pairs: 1

camera language matched pair

A bare rack-focus term produced no pull, while explicit visible-focus-change wording produced a clean transfer.

Reported result: Bare term produced no pull; explicit visible-change wording produced clean focus transfer.

Prompt: Bare rack focus vs explicit visible focus transfer

Settings: matched pairs: 2

audio text syntax

Bracket syntax produced on-screen text in 0/2 matched attempts while plain sentences produced it in 1/2; every clip carried audio, but dialogue wording was not verified by ear.

Reported result: Bracket form produced on-screen text 0/2; plain sentences 1/2; every clip carried audio, but dialogue wording was not verified by ear.

Prompt: Bracket syntax vs plain-language on-screen text/audio instruction

Settings: matched pairs: 2 · audio: true

actual charge

In a one-shot cinematic test, Wan preserved some props and instructions while the source judged Seedance stronger on atmosphere.

Reported metric: 1.35 USD

Reported result: Wan preserved some props/instructions; source judged Seedance stronger on atmosphere in this one-shot test.

Prompt: Cinematic character sequence with timed beats, prop handoff, dialogue, reveal

Settings: resolution: 720P · duration seconds: 15 · aspect ratio: 16:9 · audio: true · runs per model: 1 · rerolls: 0

actual charge

In a vertical perfume ad test, the source reports Wan covered more of the requested shot list at lower cost.

Reported metric: 1.35 USD

Reported result: Source reports Wan covered more requested shot list at lower cost.

Prompt: Vertical perfume ad testing geometry/text/hand interaction/lighting/audio

Settings: resolution: 720P · duration seconds: 15 · aspect ratio: 9:16 · audio: true · runs per model: 1 · rerolls: 0

actual charge

In a sunscreen product image-to-video test, both models preserved packaging; the source reports Wan performed more requested motion while Seedance had calmer sound.

Reported metric: 1.35 USD

Reported result: Both preserved packaging; source reports Wan performed more requested motion, Seedance had calmer sound.

Prompt: Sunscreen product I2V testing reference/text/hand motion/background

Settings: resolution: 720P · duration seconds: 15 · aspect ratio: 9:16 · audio: true · runs per model: 1 · rerolls: 0