Wan 3.0 prompt guide: what worked for us
A good Seedance prompt already works on Wan 3.0. Our first Wan clips used our Seedance prompts unchanged, pictures, voices and previous clip included, and Wan accepted every one of them. Wan's own guide asks for three small changes: name files as "Image 1" and "Audio 1", write "One continuous shot" when a clip has no cut, and write "No dialogue" when nobody talks.
29 October 2026 6 min read
Alibaba's Wan 3.0 now sits next to Seedance 2.5 and MiniMax H3 in ShowMaker. Before adding it, we re-shot three real clips from our test episodes on it, with the exact prompt, pictures, voice samples and previous clip that Seedance had been given.
Three clips, prompts unchanged
A wildlife scene with a narrator, a fantasy dinner with five characters that carries on from the clip before, and a night street with explosions. Every clip came back at the length we asked for, with sound and spoken lines.
This is the prompt the kelp clip was shot from, exactly as sent. It was written for Seedance.
Wildlife documentary cinematography, photographic realism, real camera optics, long-lens and underwater housing shots, natural sunlight through kelp canopy, fine grain, physically accurate water and fur, neutral natural colour grade. Late-afternoon sun over Kelp-forest surface. Pebble drowses on her back beside the loose Kelp strand. 0-3 seconds: Pebble bobs on her back, paws tucked on her golden belly, eyes half-closed. Water laps softly in a faint breeze. Set pieces as this beat begins — Kelp strand: floating loose on the surface among the canopy, trailing in a lazy curve. 3-10 seconds: Juniper floats in from the right, calm and intent. She rolls Kelp strand twice around the pup's belly and tucks the end under, then lets go, kelp rustling and dripping. Pebble watches wide-eyed and still. Narrator (off screen, voice-over), warm and unhurried: "Before she hunts, Juniper ties her pup to the forest." During this beat Kelp strand changes to: wound twice around the pup's belly with the end tucked under, the rest trailing back into the canopy. @Image1 is Kelp-forest surface. @Image2 is the character design for Juniper in this episode's outfit: natural wet dark-brown fur, nothing worn. Render Juniper from this design, exactly as drawn — same build, same hair, same colouring. @Image3 is the character design for Pebble in this episode's outfit: natural fluffy golden-brown natal fur, nothing worn. Render Pebble from this design, exactly as drawn — same build, same hair, same colouring. @Audio1 is Narrator's voice: use it only for how Narrator sounds. Music: soft sustained strings with a gentle harp, warm and slow. Do not add subtitles. Avoid: illustration, cartoon, 3D render look, outlines, cel shading, oversaturation, people, humans, hands, boats, divers, buildings, text or lettering of any kind, signage, logos
What Wan's own guide asks for
We read Alibaba's Wan 3.0 prompt guide line by line against our Seedance prompt. They ask for nearly the same thing: a style line, what each picture is for, timed beats, the spoken lines, the sound. Where they differ:
| Wan 3.0, from its guide | Seedance 2.5, what we do | |
|---|---|---|
| Naming files | "Image 1", "Audio 1", in plain words, counted per kind in the order sent | "@Image1" works |
| A clip with no cut | Write "One continuous shot" first, or Wan may cut on its own | Say nothing |
| A clip where nobody talks | Write "No dialogue", or Wan may add speech | Say nothing |
| Voices | "Voice timbre references Audio 1", once per character, and "Lip sync" | Say which audio is whose voice |
| Things to avoid | A real list of what you don't want works | Describe what you want instead |
| Beat length | 2 to 5 seconds a beat | At least 3 seconds a beat worked for us |
The "Avoid" line at the end of our prompt, which Seedance's guide doesn't ask for, is exactly the kind of list Wan's guide wants.
The same prompt, tuned for Wan
The kelp prompt with three changes from the table: "One continuous shot" first, files named "Image 1" and so on, and the voice line in Wan's words. Nothing else changed. Wan's guide would also split the 7-second beat in two; we left it, to show only the naming changes. We haven't shot this version; it's here to show how small the difference is.
One continuous shot. Wildlife documentary cinematography, photographic realism, real camera optics, long-lens and underwater housing shots, natural sunlight through kelp canopy, fine grain, physically accurate water and fur, neutral natural colour grade. Late-afternoon sun over Kelp-forest surface. Pebble drowses on her back beside the loose Kelp strand. 0-3 seconds: Pebble bobs on her back, paws tucked on her golden belly, eyes half-closed. Water laps softly in a faint breeze. Set pieces as this beat begins — Kelp strand: floating loose on the surface among the canopy, trailing in a lazy curve. 3-10 seconds: Juniper floats in from the right, calm and intent. She rolls Kelp strand twice around the pup's belly and tucks the end under, then lets go, kelp rustling and dripping. Pebble watches wide-eyed and still. Narrator (off screen, voice-over), warm and unhurried: "Before she hunts, Juniper ties her pup to the forest." During this beat Kelp strand changes to: wound twice around the pup's belly with the end tucked under, the rest trailing back into the canopy. Image 1 is Kelp-forest surface. Image 2 is the character design for Juniper in this episode's outfit: natural wet dark-brown fur, nothing worn. Render Juniper from this design, exactly as drawn — same build, same hair, same colouring. Image 3 is the character design for Pebble in this episode's outfit: natural fluffy golden-brown natal fur, nothing worn. Render Pebble from this design, exactly as drawn — same build, same hair, same colouring. Narrator's voice timbre references Audio 1. Music: soft sustained strings with a gentle harp, warm and slow. Do not add subtitles. Avoid: illustration, cartoon, 3D render look, outlines, cel shading, oversaturation, people, humans, hands, boats, divers, buildings, text or lettering of any kind, signage, logos
Limits worth knowing before you plan
- Length: 2 to 30 seconds a clip on the API (ShowMaker offers 4 to 30).
- References: up to 10 pictures, up to 5 videos totalling 15 seconds, up to 5 voice samples totalling 15 seconds. To join clips, ShowMaker sends only the last 3 seconds of the previous clip, well inside that.
- Prompt rewriting: Wan's API rewrites your prompt by default ("prompt expansion"). Turning it off is faster and keeps your exact words; it can also lower quality, so test both.
- Start and end frames: they can't be used together with references. Send frames and your character pictures and voices are dropped. We wrote up why in Why we don't use start and end frames.
- Price: the list price is 5 cents a second at 480p, 10 cents at 720p and 20 cents at 1080p. A reference video's seconds can be billed too.
Wan 3.0 in ShowMaker
In ShowMaker, Wan 3.0 is one of the three models you pick per episode, at 480p, 720p or 1080p. The app writes the prompt for each clip and sends the right pictures, voices and previous clip with it.