Wan 3.0 review: every mode tested in 31 clips

Run 2 · 12/12

Run 3 · 12/12

Both lines came out word for word in all three runs, in order, with the right mouth moving each time.
No background music, as requested.
The mix is very loud: −7 to −9 LUFS, with peaks at full scale.

Close-ups of the woman and the man in all three runs: her mouth moves during her line, his during “Watch me” — Who speaks when. Her mouth moves during her line (0–2.5 s), and his moves during “Watch me” (about 3.5–4.2 s), in all three runs.

Run 1 · 12/12 · 56 s · −9.0 LUFS
She says her line at 0.0–3.0 s; he laughs and answers “Watch me” at 3.6–4.1 s. Word for word, right speaker, no music. He reaches for the croissant but leaves it.
Run 2 · 12/12 · 55 s · −7.1 LUFS
Her line at 0.0–2.4 s, his laugh at 2.5–3.2 s, “Watch me” at 3.5–4.3 s, then he picks up the croissant. Word for word.
Run 3 · 12/12 · 56 s · −8.0 LUFS
He laughs during her line (0.0–2.6 s), takes the croissant at 2.2 s, says “Watch me” at 3.6–4.0 s and bites into it. Word for word.

Criterion	Run 1	Run 2	Run 3
Prompt	2	2	2
Subject	2	2	2
Physics	2	2	2
Camera	2	2	2
Detail	2	2	2
Sound	2	2	2
Total	12/12	12/12	12/12

See ten frames from every run

Mode 2

Reference images

Reference images · one face

One photo, a new scene · 32/36

Reference mode builds a new scene from tagged pictures instead of animating a first frame. First test: one face in an entirely different place.

A young woman at a sunny cafe table holding an iced latte — [Image 1] face and hair

The prompt, word for word

“[Image 1] is the woman's face and hair. The woman from [Image 1] rides a bicycle along a seaside promenade at sunset, laughing as the wind moves her hair. The camera tracks alongside her.”

5 s · 720p · 9:16 · sound on · reference mode · 25 credits per run

Run 1 · 12/12

Run 2 · 11/12

Run 3 · 9/12

The face and hair matched the photo in all three runs.
She also kept the café T-shirt and necklace in every run: a reference carries over the outfit unless you describe another one.
Run 3 added a wave and allowed the background to drift.

Run 1 · 12/12 · 79 s · −25.6 LUFS
Same face and hair as in the café photo, now on a seaside boardwalk at sunset. The camera rides alongside, and the wind blows her hair across her face at 2.5–3.0 s. Waves and gulls.
Run 2 · 11/12 · 82 s · −18.7 LUFS
A sunset promenade lined with palms. The face holds, but the camera rides in front of her instead of alongside.
Run 3 · 9/12 · 84 s · −27.1 LUFS
She waves at 3.8–4.0 s, which we didn't ask for, and the background drifts from a pier with sailboats to a palm-lined boardwalk between 2.2 and 4.2 s. Camera in front of her.

Criterion	Run 1	Run 2	Run 3
Prompt	2	2	1
Subject	2	2	2
Physics	2	2	2
Camera	2	1	1
Detail	2	2	1
Sound	2	2	2
Total	12/12	11/12	9/12

See ten frames from every run

Reference images · face, product, room

Three references in one shot · 33/36

The harder version: a face, a product and a room, each tagged, combined into a single action.

The LUMEA serum bottle on a marble counter — [Image 2] product

A bright bathroom with a round mirror, an oak vanity and plants — [Image 3] room

The prompt, word for word

“[Image 1] is the woman's face and hair, [Image 2] is the LUMEA serum bottle, [Image 3] is the bathroom. In the bathroom from [Image 3], the woman from [Image 1] picks up the [Image 2] bottle, turns the label to the camera and smiles.”

5 s · 720p · 9:16 · sound on · reference mode · 25 credits per run

Run 1 · 11/12

Run 2 · 10/12

Run 3 · 12/12

All three runs combined the woman, the LUMEA bottle and the bathroom, with the label remaining readable in her hand.
Runs 1 and 3 opened on the counter from the product photo, then revealed the bathroom.
Quiet soundtracks: −32 to −48 LUFS.

Run 1 · 11/12 · 1 min 43 s · −40.1 LUFS
Opens on the bottle on a marble counter; a hand picks it up, and the camera pulls back to reveal her in the bathroom with the round mirror. She turns the LUMEA label toward the camera and smiles. Very quiet.
Run 2 · 10/12 · 94 s · −48.3 LUFS
Opens wide on the bathroom from the reference. She steps in, picks up the bottle and holds it toward the camera, smiling. Near-silent soundtrack.
Run 3 · 12/12 · 2 min 12 s · −32.2 LUFS
Like run 1: a hand lifts the bottle from the counter, she is revealed by 2.5 s and holds the label toward the camera from 2.8 s. Face, bottle and room all match the references.

Criterion	Run 1	Run 2	Run 3
Prompt	2	2	2
Subject	2	2	2
Physics	2	2	2
Camera	2	2	2
Detail	2	2	2
Sound	1	0	2
Total	11/12	10/12	12/12

See ten frames from every run

Mode 3

First and last frame

First + last frame

A rose that has to open on cue · 31/36

Give Wan 3.0 a first and last frame, and it fills in the motion between them. We asked a closed bud to become a bloom we made with the Image Editor.

First frame: a closed red rose bud in a glass vase on a sunny windowsill — First frame

Last frame: the same rose fully open, made with the ClipRush Image Editor — Last frame

The prompt, word for word

“Time-lapse: the rose bud slowly opens into a full bloom, its petals unfolding one by one. Static camera, soft morning light.”

5 s · 720p · sound on · first + last frame · 25 credits per run

Run 1 · 10/12

Run 2 · 10/12

Run 3 · 11/12

Every run started on our bud and ended on our bloom: the average pixel difference from the inputs was 4 at the start and 5–6 at the end, versus 37 between the two images.
The opening was finished by 3 seconds, and the rose held still for the last two.
Run 1's soundtrack was close to silent (−61 LUFS).

The bud and bloom inputs above the three runs: each starts on the bud, opens between 1 and 3 seconds and ends on the bloom — Our two inputs (top) and all three runs. Each starts on the bud, opens between 1 and 3 seconds, and ends on the bloom.

Run 1 · 10/12 · 64 s · −61.4 LUFS
Starts on the bud and ends on the bloom. The petals open between 1.0 and 3.0 s, then the rose holds still. Near-silent.
Run 2 · 10/12 · 79 s · −45.5 LUFS
The same path, almost frame for frame. Faint birdsong.
Run 3 · 11/12 · 61 s · −39.8 LUFS
Same again, with a slightly slower opening. Quiet ambience.

Criterion	Run 1	Run 2	Run 3
Prompt	2	2	2
Subject	2	2	2
Physics	2	2	2
Camera	2	2	2
Detail	2	2	2
Sound	0	0	1
Total	10/12	10/12	11/12

See ten frames from every run

Mode 4

Image to video

Image to video · product

A product shot with printed text · 32/36

The bread-and-butter job: animate a product photo without ruining the label.

Starting image: a frosted serum bottle labelled LUMEA NIGHT SERUM on a marble bathroom counter — Starting image

The prompt, word for word

“The camera slowly orbits the serum bottle from left to right while soft morning light glides across the frosted glass and the LUMEA label. A single golden drop slides down the dropper.”

5 s · 720p · sound on · 25 credits per run

Run 1 · 11/12

Run 2 · 9/12

Run 3 · 12/12

The label never broke: LUMEA and NIGHT SERUM remained readable in every frame of all three runs.
Run 2 placed the drop inside the bottle.
Two of three soundtracks were very quiet (−36 and −43 LUFS).

Run 1 · 11/12 · 60 s · −36.0 LUFS
A slow orbit brings the mirror and sink into view from 2.5 s. A golden drop forms on the bulb at 0.5 s and runs down the cap and glass from 2.2 s. The label remains readable throughout. Very quiet.
Run 2 · 9/12 · 58 s · −43.3 LUFS
The clearest orbit of the three: new bottles appear on a shelf at 4.2 s as the camera swings around. But the drop forms inside the bottle, below the label, and hangs in the serum from 1.2 s, which a real drop can't do. Very quiet.
Run 3 · 12/12 · 62 s · −28.4 LUFS
A golden drop beads on top of the bulb at 0.5 s, runs down the collar and glass from 1.8 s, and pools on the marble by 3.8 s. Subtle orbit, label readable throughout. Soft liquid sounds.

Criterion	Run 1	Run 2	Run 3
Prompt	2	1	2
Subject	2	2	2
Physics	2	1	2
Camera	2	2	2
Detail	2	2	2
Sound	1	1	2
Total	11/12	9/12	12/12

See ten frames from every run

Image to video · animal

A dog running at the camera · 31/36

Fast motion, fur and a moving camera all at once.

Starting image: a golden retriever sitting on a park lawn with a red ball in front of its paws — Starting image

The prompt, word for word

“The dog grabs the red ball in its mouth and runs toward the camera, ears flapping, as the camera tracks backward low to the ground.”

5 s · 720p · sound on · 25 credits per run

Run 1 · 10/12

Run 2 · 12/12

Run 3 · 9/12

The motion and camera were right in all three runs: the grab, the run, and a low camera backing away.
In runs 1 and 3, the dog's floppy ears briefly turned upright and pointed, making it look like a different breed.

The dog's head from 2.0 to 3.25 seconds in all three runs; in runs 1 and 3 the ears switch between floppy and pointed — The dog's head between 2.0 and 3.25 s. Runs 1 and 3 switch between floppy and pointed ears; run 2 remains a retriever.

Run 1 · 10/12 · 59 s · −22.6 LUFS
Grabs the ball at 0.8 s and runs toward the camera from 1.5 s while the camera backs away low. Its floppy ears turn upright and pointed at 2.0, 2.5 and 3.0 s, so it briefly looks like a different breed.
Run 2 · 12/12 · 59 s · −23.6 LUFS
The clean run: ball in mouth at 0.8 s, a steady run toward the low, retreating camera, with ears flapping and drool flying. It remains the same retriever throughout.
Run 3 · 9/12 · 60 s · −15.4 LUFS
An odd, overly wide-open jaw at 0.5 s before the grab, then the same ear problem as run 1: pointed ears at 2.75 and 3.25 s.

Criterion	Run 1	Run 2	Run 3
Prompt	2	2	2
Subject	0	2	0
Physics	2	2	1
Camera	2	2	2
Detail	2	2	2
Sound	2	2	2
Total	10/12	12/12	9/12

See ten frames from every run

Image to video · landscape

An aerial move over a lake · 30/36

A compound camera move: glide, then rise.

Starting image: a calm alpine lake at sunrise seen from just above the water — Starting image

The prompt, word for word

“The drone glides forward low over the lake, then rises to reveal the valley beyond. Mist drifts over the water and two birds fly across the frame.”

5 s · 720p · sound on · 25 credits per run

Run 1 · 9/12

Run 2 · 9/12

Run 3 · 12/12

Two of three runs drew an actual drone on the water: “the drone glides” was interpreted as an object, not as the camera.
Only run 3 rose to reveal the valley.

Five frames from each lake run: runs 1 and 2 show a drone on the water, run 3 rises over the forest — Runs 1 and 2 place a drone on the water and never rise. Run 3 makes the aerial move we intended.

Run 1 · 9/12 · 56 s · −28.5 LUFS
Took “the drone” literally: a small drone sits on the water and creates ripples at 0.5–1.5 s. Three or four birds instead of two appear from 2.5 s, and the camera never rises.
Run 2 · 9/12 · 68 s · −14.0 LUFS
Again, a drone appears in frame, now skimming toward the camera with a wake at 0.5–2.5 s. The two birds are right, with reflections in the water at 3.5 s, but there is no rise.
Run 3 · 12/12 · 56 s · −19.2 LUFS
What we requested: two birds take off from the water at 0.5 s and cross the frame, while the camera glides forward and rises from 2.5 s to reveal the forested valley.

Criterion	Run 1	Run 2	Run 3
Prompt	0	0	2
Subject	2	2	2
Physics	2	2	2
Camera	1	1	2
Detail	2	2	2
Sound	2	2	2
Total	9/12	9/12	12/12

See ten frames from every run

Mode 5

Text to video

The same product, from words alone · 28/36

No starting image this time: the bottle, label and room all have to come from the prompt.

The prompt, word for word

“A frosted glass serum bottle with a gold dropper and a label that reads "LUMEA" with "NIGHT SERUM" underneath stands on a white marble bathroom counter in soft morning light. The camera slowly orbits the bottle while a single golden drop slides down the dropper. Premium product commercial.”

5 s · 720p · 9:16 · sound on · 25 credits per run

Run 1 · 9/12

Run 2 · 12/12

Run 3 · 7/12

LUMEA was spelled correctly in all three runs.
Only run 2 got the physics right. Run 3 invented a second dropper and printed NIGHT SERUM twice.

Four frames from each text-to-video run: run 2 is clean, run 3 has a second dropper sticking out of the bottle — Run 2 is the clean one. Run 1 keeps the drop inside the bottle; run 3 grows a second dropper and repeats NIGHT SERUM.

Run 1 · 9/12 · 53 s · −16.4 LUFS
The label and orbit are right, but the “drop” forms on the pipette inside the bottle, in the serum, and a pump bottle appears in the background at 4.0 s. The L of LUMEA hides behind the label's edge. Soft background music.
Run 2 · 12/12 · 1 min 46 s · −23.7 LUFS
The run to copy: a drop beads on the bulb, runs down the collar and glass, and pools on the marble by 4.5 s while the camera orbits. Clean label.
Run 3 · 7/12 · 53 s · −8.4 LUFS
An impossible bottle: a second glass pipette sticks out of the collar at an angle and drips. The label prints NIGHT SERUM twice, and the camera pushes in rather than orbiting.

Criterion	Run 1	Run 2	Run 3
Prompt	1	2	1
Subject	2	2	2
Physics	1	2	0
Camera	2	2	1
Detail	1	2	1
Sound	2	2	2
Total	9/12	12/12	7/12

See ten frames from every run

Mode 6

30-second takes

30-second take

Seven actions in one 30-second shot · 19/24

Wan 3.0 is the only engine in ClipRush that goes past 15 seconds. We packed a 30-second clip with seven actions and asked for one continuous take.

Starting image: a young woman at a sunny cafe table holding an iced latte — Starting image

The prompt, word for word

“She reads a paperback and sips her iced latte. A waiter sets a croissant on her table; she thanks him and takes a bite. Her phone buzzes, she reads the message and laughs, then she packs up her book and walks out of the frame. One continuous take while the camera slowly drifts closer.”

30 s · 720p · sound on · 150 credits per run

Run 1 · 10/12

Run 2 · 9/12

Her face stayed the same for the full 30 seconds in both runs.
Run 2 cut to a new camera angle at exactly 15.0 s and skipped the bite.
Run 1 remained one take, but a phone appeared on the table from nowhere at 21.5 s.
Both runs turned “she thanks him” into a spoken “Thank you”, timed to her looking up.

Frames from 14.7 to 15.2 seconds: run 2 jumps to a new angle at 15.0 s, run 1 continues smoothly — 14.7–15.2 s. Run 2 jumps to a new angle at 15.0 s: new background, new light, and a phone on the table. Run 1 continues.

Run 1 · 10/12 · 3 min 43 s · −19.0 LUFS
Reads and sips (0–10 s); a waiter's hand sets down a croissant (11–12 s); she looks up and says “Thank you” (13.1 s); bites (16–18 s); a phone appears on the table from nowhere and buzzes (21.5 s); reads it and laughs (23–25 s); packs up and walks out (26–29 s). One continuous take, but the camera barely moves closer.
Run 2 · 9/12 · 4 min 2 s · −10.0 LUFS
Reads and sips; the croissant arrives at 5 s, and she looks up to thank him at 7 s. At exactly 15.0 s, the clip cuts to a new camera angle, despite “one continuous take”. She never takes a bite. Reads the phone (18–20 s), laughs, packs a tote and walks out (27–28 s).

Criterion	Run 1	Run 2
Prompt	2	1
Subject	2	2
Physics	1	2
Camera	1	0
Detail	2	2
Sound	2	2
Total	10/12	9/12

See ten frames from every run

Mode 7

480p, 720p and 1080p

The product prompt again, with one run each at 480p and 1080p, shown beside the 720p runs above. One run per resolution, so these clips aren’t scored.

480p

720p

1080p

Resolution	Credits (5 s)	Render	Label sharpness	What we saw
480p	15	97 s	0.19×	NIGHT appears as NICHT in several frames.
720p	25	60 s	1.0× (baseline)	Label clean in all three runs.
1080p	45	97 s	2.3×	The sharpest label in the test.

At 480p, NIGHT appeared as NICHT in several frames.
At 1080p, the label was 2.3 times sharper than at 720p (edge sharpness at the same display size), while 480p was only a fifth as sharp.
The 480p and 1080p runs both took 97 seconds, so one run each can't show which is faster.

The LUMEA NIGHT SERUM label at 480p, 720p and 1080p; at 480p NIGHT reads as NICHT — The label at 1.0 and 3.0 s, scaled to the same size. At 480p, the G of NIGHT loses its bar.

A/B test

Does ClipRush’s prompt writer help Wan 3.0?

In our image-to-video test, ClipRush expanded the short café prompt before Wan 3.0 saw it. Here we sent the same prompt as written: starting it with “ACTION:” tells the studio the prompt is already structured, so it reaches Wan 3.0 untouched. Scores use that test's nine criteria (not the six used elsewhere on this page), 54 points per three runs.

The prompt, word for word

“ACTION: She glances out the window, then turns back to the camera, takes a sip of her iced latte through the straw, lowers the cup and waves at the camera with her free hand, smiling. Slow push-in, people walk past outside.”

10 s · 720p · sound on · 50 credits

As written · 51/54

Run 1 · 17/18

Run 2 · 17/18

Run 3 · 17/18

With the studio’s rewrite · 52/54 · from our image-to-video test

Run 1 · 17/18

Run 2 · 18/18

Run 3 · 17/18