All posts
comparisons

Wan 3.0 vs HappyHorse 1.1: 20 Prompts on Four Alibaba Video Models

We gave four Alibaba video models, Wan 3.0, Wan 3.0 Prime, HappyHorse 1.1 and HappyHorse 1.0, the same 20 prompts on Flixly and kept the first result each time. On these prompts Wan stayed in one continuous take and followed camera directions, while HappyHorse cut between angles. Wan 3.0 Prime was the fastest by far.

By Flixly TeamNovember 13, 2026
Wan 3.0 vs HappyHorse 1.1: 20 Prompts on Four Alibaba Video Models

TL;DR

There is no single winner among Alibaba's four video models. In 20 same-prompt tests on Flixly at 1080p and 8 seconds, Wan 3.0 and Wan 3.0 Prime stayed in one continuous take on all 20 prompts and followed camera directions literally, while HappyHorse 1.1 and HappyHorse 1.0 cut between angles and took more liberties. Wan 3.0 Prime was fastest, with a median of 110 seconds per clip.

Wan 3.0 and Wan 3.0 Prime stayed in one continuous take on all 20 of our prompts and followed camera directions literally, while HappyHorse 1.1 and HappyHorse 1.0 cut between shots like an editor and take more liberties with the prompt; Wan 3.0 Prime was by far the fastest. We gave all four Alibaba video models the same 20 text prompts on Flixly, at 1080p, 8 seconds and 16:9, and kept the first result every time. That is 80 clips. Eight of the prompts are below as 2×2 grids, and every prompt is in the results table at the end.

Each grid shows Wan 3.0 top left, Wan 3.0 Prime top right, HappyHorse 1.1 bottom left and HappyHorse 1.0 bottom right. The grids are silent: all four models made a soundtrack for every clip, but four soundtracks at once is noise, so the sound section below sticks to what we measured.

What are Wan 3.0, Wan 3.0 Prime, HappyHorse 1.1 and HappyHorse 1.0?

They are four video generation models from Alibaba, in two families.

All four generate sound with the picture. On Flixly they are all in the video generator.

How do the four models compare on Flixly?

Wan 3.0 Wan 3.0 Prime HappyHorse 1.1 HappyHorse 1.0
Text to video Yes Yes Yes Yes
Image to video (first frame) Yes Yes Yes Yes
References Up to 10 images, 5 videos, 5 audio clips Up to 10 images, 5 videos, 5 audio clips Up to 9 images 1 image
Video editing No No No Yes
Length 2 to 30 seconds 2 to 30 seconds 3 to 15 seconds 3 to 15 seconds
Resolution 480p, 720p and 1080p 480p, 720p and 1080p 720p and 1080p 720p and 1080p
Aspect ratios Auto, 16:9, 9:16, 1:1, 4:3, 3:4 Auto, 16:9, 9:16, 1:1, 4:3, 3:4 16:9, 9:16, 1:1, 4:3, 3:4, 4:5, 5:4, 21:9, 9:21 16:9, 9:16, 1:1, 4:3, 3:4
Sound Generated with the video, can be switched off Generated with the video, can be switched off Generated with the video Generated with the video
Frame rate of our clips 30 fps 30 fps 24 fps 24 fps
Median time in our test 384 seconds 110 seconds 171 seconds 161 seconds

Times are wall-clock on Flixly from pressing generate to a finished 8-second 1080p clip, so they include queueing. Wan 3.0 slowed sharply in the middle of the afternoon on our test day, which inflates some of its times.

How we tested

  • Same prompt, same settings. Every model got the same 20 text prompts at 1080p, 8 seconds and 16:9.
  • First result only. No rerolls and no editing. The grids only add the layout and the labels, and are compressed for the web.
  • Checked frame by frame. We looked at every one of the 80 clips as a grid of frames taken twice a second, and at full-size frames for anything we describe below.
  • Some prompts are not shown. We left out the soccer, astronaut, city-skyline and harbor grids because they contain real brand names, flags, real-looking cities or garbled signs. They are still in the table.

Which model follows the prompt best?

It depends on what the prompt asks for. Wan 3.0 and Wan 3.0 Prime were more literal about camera directions; the HappyHorse models were more likely to act out the whole sentence.

Prompt: "A dragon made of autumn leaves rises from the forest floor and swirls into the sky, fantasy" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 375 seconds, Wan 3.0 Prime took 110 seconds, HappyHorse 1.1 took 183 seconds, HappyHorse 1.0 took 163 seconds.

All four built a dragon out of the leaves. Only HappyHorse 1.0 really sent it into the sky: the camera looks up through the trees as the serpent spirals upward and scatters into leaves against the sun. Wan 3.0's serpent rears up and then bursts back into leaves at tree-trunk height. Wan 3.0 Prime and HappyHorse 1.1 made winged dragons that take off up through the trees, and HappyHorse 1.1 added glowing eyes and a glowing chest the prompt did not ask for.

Prompt: "A humpback whale breaches in slow motion off a rocky coast, spray catching the sunlight" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 276 seconds, Wan 3.0 Prime took 132 seconds, HappyHorse 1.1 took 163 seconds, HappyHorse 1.0 took 161 seconds.

Every model made a full breach and splash. The difference is the animal: both Wan whales look like humpbacks, with knobbly heads and long white flippers, while both HappyHorse whales have patchy black-and-white markings closer to an orca. HappyHorse 1.1 also moved the scene to golden hour.

Prompt: "A surfer rides inside a barrel wave, the camera following from behind, water curling overhead" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 237 seconds, Wan 3.0 Prime took 92 seconds, HappyHorse 1.1 took 162 seconds, HappyHorse 1.0 took 162 seconds.

None of the four followed "from behind". Every camera faces the surfer: the Wan models from out in front of the wave, HappyHorse 1.1 from right in front of his face, and HappyHorse 1.0 from inside a tube that swirls more like a whirlpool than a wave. If a camera position matters, expect to reroll.

Which one handles camera moves and editing better?

On all 20 of these prompts, both Wan models stayed in one continuous take; we did not see a cut in any of the 40 Wan clips. That is a result of this test, not a rule: Wan can still cut between shots on its own, as both Wan models did on an anime prompt in our Wan 3.0 vs Wan 3.0 Prime test. Both HappyHorse versions often cut to a second or third angle, which can make a clip feel edited but also means you get less of the shot you described.

Prompt: "A spaceship descends through clouds above a desert city at dusk, lights blinking, epic scale" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 342 seconds, Wan 3.0 Prime took 110 seconds, HappyHorse 1.1 took 202 seconds, HappyHorse 1.0 took 161 seconds.

Both Wan models brought a huge ship down through the clouds over a lit-up city in one take, with dust rising at the end. HappyHorse 1.1 made three shots: the underside up close, the ship from above with its engines firing, then a wide view of a small ship over the city. HappyHorse 1.0 made two.

Prompt: "A slow aerial orbit around a modern glass house on a forested hillside at golden hour" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 734 seconds, Wan 3.0 Prime took 112 seconds, HappyHorse 1.1 took 188 seconds, HappyHorse 1.0 took 163 seconds.

The two Wan models did what the prompt says: a slow, high partial orbit, with the house turning a little in frame. HappyHorse 1.1 started tight on a corner of the house and pulled back to show the hillside. HappyHorse 1.0 slid sideways past pine trunks at roughly the height of the house, which reads less like an aerial shot. This is also where Wan 3.0's queue was worst: 734 seconds against 112 for Prime.

Prompt: "A cat knocks a pen off a desk, then looks directly at the camera, comedic timing" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 324 seconds, Wan 3.0 Prime took 76 seconds, HappyHorse 1.1 took 172 seconds, HappyHorse 1.0 took 161 seconds.

Here the editing pays off. Both HappyHorse models pushed the pen over the edge and then cut to a close-up of the cat's face for the stare, which is how you would cut the joke. Wan 3.0 Prime tipped the pen off in one wide shot and then turned to the camera. Wan 3.0's cat slides the pen toward the edge and out of frame, so you never see it fall, before it looks up into the lens.

Which one looks more natural?

All four look convincing at 1080p, and to our eyes none was clearly ahead on image quality across the set. The differences are in choices, and HappyHorse sometimes changed or dropped a detail the prompt asked for.

Prompt: "Close-up of a woman laughing, freckles, natural window light, emotional portrait" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 390 seconds, Wan 3.0 Prime took 74 seconds, HappyHorse 1.1 took 199 seconds, HappyHorse 1.0 took 162 seconds.

The two Wan models produced strikingly similar women, laughing from the first frame to nearly the last. Both HappyHorse versions gave the laugh an arc, starting from a smile, building to a laugh and settling again. Three of the four put clear freckles on her face; HappyHorse 1.1 gave her barely any.

Prompt: "A candle flame flickers in a dark room, then a gust of wind blows it out and smoke curls upward" Text to video, 1080p, 8 seconds, 16:9. Wan 3.0 took 1213 seconds, Wan 3.0 Prime took 165 seconds, HappyHorse 1.1 took 162 seconds, HappyHorse 1.0 took 161 seconds.

A two-step prompt that every model got right: flicker, out, smoke. Wan 3.0 Prime put the flame out earliest and spent the rest of the clip on the smoke; HappyHorse 1.1 cut to a wider shot of the candle on a table first. This prompt also had Wan 3.0's slowest render of the day, 1213 seconds.

How fast is each model?

Wan 3.0 Prime was the fastest model on 16 of the 20 prompts, with a median of 110 seconds for an 8-second 1080p clip. HappyHorse 1.0 was fastest on the other 4 and the most consistent, taking 144 to 163 seconds every time. HappyHorse 1.1 took 145 to 255 seconds. Wan 3.0 was the slowest on every prompt, from 237 to 1213 seconds with a median of 384, and that includes the slowdown in its queue that afternoon. If you are iterating on a prompt, start with Wan 3.0 Prime.

What about sound?

Every one of the 80 clips came back with a stereo soundtrack. We did not score dialogue or sound quality in this test; for that, see HappyHorse 1.1 vs 1.0, where we checked spoken lines. One measured difference is worth knowing when you cut clips together: the Wan files are mixed much louder. Median loudness was about −11 LUFS for both Wan models, −16 LUFS for HappyHorse 1.0 and −20 LUFS for HappyHorse 1.1, so expect to turn HappyHorse clips up or Wan clips down.

All 20 prompts: results

Times are in seconds, wall-clock on Flixly, including queueing.

# Prompt Wan 3.0 Wan 3.0 Prime HappyHorse 1.1 HappyHorse 1.0 What we saw
1 Earbud case on a pedestal 686 206 164 155 All four turn the case. HappyHorse 1.1 adds a warm rim light; HappyHorse 1.0 ends on the earbuds showing in the open case.
2 Ramen lifted with chopsticks 500 110 189 147 All four lift the noodles. Wan 3.0 Prime and HappyHorse 1.0 show the most steam.
3 Humpback whale breach 276 132 163 161 All four breach. Wan whales look like humpbacks; HappyHorse whales have orca-like markings.
4 Spaceship over a desert city 342 110 202 161 Wan: one continuous descent. HappyHorse: cuts between angles.
5 Man in a kitchen talks about his pancake 631 112 145 158 All four end on him facing the camera. HappyHorse 1.0 opens on the pancakes, then never shows him holding one. Dialogue not scored.
6 Soccer free kick 368 193 206 160 Not shown: a real brand name and garbled sponsor boards.
7 Red fox in a birch forest 423 174 170 158 The two Wan clips are near-identical side-on tracking shots. HappyHorse 1.1 walks the fox toward the camera and away; HappyHorse 1.0 ends closer on the fox.
8 Aerial orbit of a glass house 734 112 188 163 Wan: slow high orbit. HappyHorse 1.1 pulls back from the house; HappyHorse 1.0 slides past tree trunks.
9 Runway model in a silver gown 394 93 169 162 Wan keeps the whole gown in one shot. Both HappyHorse versions cut to close-ups of the fabric.
10 Dragon of autumn leaves 375 110 183 163 Only HappyHorse 1.0 sends it into the sky; Wan 3.0's bursts apart near the ground.
11 City timelapse, day to night 358 109 165 160 All four go from day to night with light trails. Not shown: the skylines look like real cities.
12 Potter's hands at the wheel 341 73 188 161 All four shape a pot. Wan chose terracotta clay, HappyHorse grey and white clay.
13 Cyclist in a lavender field 377 97 187 163 Wan: wide side tracking shots. HappyHorse 1.1 drifts to a front-on angle; HappyHorse 1.0 is a tight side shot with the lavender blurred.
14 Cat knocks a pen off a desk 324 76 172 161 HappyHorse cuts to the cat's face for the stare. Wan 3.0's pen slides out of frame before it falls.
15 Steam train on a viaduct 528 95 255 159 All four cross the viaduct. HappyHorse 1.1 starts close on the engine and pulls back; HappyHorse 1.0 cuts to a close-up of the engine.
16 Astronauts pass an apple 305 75 162 144 Not shown: flags and mission-style patches. HappyHorse 1.0 has the clearest sign of zero gravity, floating hair.
17 Surfer in a barrel wave 237 92 162 162 None films from behind as asked.
18 Laughing woman, freckles 390 74 199 162 Wan laughs throughout; HappyHorse builds to the laugh. HappyHorse 1.1 drops most of the freckles.
19 Tilt-shift harbor 426 95 163 162 Only HappyHorse gives the miniature look, with blurred top and bottom; Wan's look like ordinary drone footage. Not shown: brand-like text on containers.
20 Candle blown out 1213 165 162 161 All four get the flicker, the gust and the smoke.

Which one should you use?

If you want Use
The fastest draft Wan 3.0 Prime
One continuous shot that follows your camera direction Wan 3.0 or Wan 3.0 Prime (one take on all 20 of our prompts, but not guaranteed)
Clips longer than 15 seconds Wan 3.0 or Wan 3.0 Prime
A clip that cuts between angles on its own HappyHorse 1.0 or HappyHorse 1.1
A comic beat or a reaction shot HappyHorse 1.0 or HappyHorse 1.1
A tilt-shift miniature look HappyHorse 1.0 or HappyHorse 1.1
Vertical or ultra-wide formats (4:5, 21:9, 9:21) HappyHorse 1.1
Editing an existing video HappyHorse 1.0

Wan 3.0 and Wan 3.0 Prime gave very similar results in this test, so unless you are comparing the two yourself, Prime's speed makes it the easier pick.

How to run your own same-prompt test on Flixly

  1. Open the video generator with Wan 3.0 Prime selected and write your prompt.
  2. Set the same length, resolution and aspect ratio you will use for every model, for example 8 seconds, 1080p and 16:9.
  3. Check the price on the button, generate, and keep the first result.
  4. Switch the model to Wan 3.0, HappyHorse 1.1 and HappyHorse 1.0 with the same prompt and settings.
  5. Compare the first results side by side before you start rewriting the prompt for one model.

Model pages: Wan 3.0, Wan 3.0 Prime, HappyHorse 1.1 and HappyHorse 1.0.

Frequently Asked Questions

Which is better, Wan 3.0 or HappyHorse 1.1?▾

It depends on the shot. In Flixly's test of Alibaba's models with the same 20 prompts, Wan 3.0 and Wan 3.0 Prime stayed in one continuous take on all 20 prompts and followed camera directions such as a slow aerial orbit, though Wan can still cut on its own on other prompts. HappyHorse 1.1 and HappyHorse 1.0 cut between angles, which worked well for a comic beat but sometimes changed details the prompt asked for.

What is the difference between Wan 3.0 and Wan 3.0 Prime?▾

Wan 3.0 Prime is the faster tier of Alibaba's Wan 3.0, with the same inputs, lengths and resolutions on Flixly. In Flixly's test it took a median of 110 seconds for an 8-second 1080p clip against 384 seconds for Wan 3.0, and the two often produced very similar shots. HappyHorse 1.1 and HappyHorse 1.0 took a median of 171 and 161 seconds.

Which Alibaba video model is the fastest?▾

Wan 3.0 Prime. In Flixly's test of four Alibaba models it was the fastest on 16 of 20 prompts, with a median of 110 seconds for an 8-second 1080p clip. HappyHorse 1.0 was fastest on the other 4 and the most consistent, HappyHorse 1.1 took a median of 171 seconds, and Wan 3.0 was slowest every time. Times include queueing on Flixly.

Do Wan 3.0 and HappyHorse 1.1 generate sound?▾

Yes. All four Alibaba models, Wan 3.0, Wan 3.0 Prime, HappyHorse 1.1 and HappyHorse 1.0, generate a soundtrack with the picture, and all 80 clips in Flixly's test came back with stereo audio. On Flixly the Wan models let you switch sound off. The Wan files were mixed louder, so match levels when you cut clips together.

How long can Wan 3.0 and HappyHorse 1.1 videos be?▾

On Flixly, Alibaba's Wan 3.0 and Wan 3.0 Prime make clips from 2 to 30 seconds at 480p, 720p or 1080p. HappyHorse 1.1 and HappyHorse 1.0 make clips from 3 to 15 seconds at 720p or 1080p. HappyHorse 1.1 has the most aspect ratios, including 4:5, 21:9 and 9:21.

Which Alibaba model should I start with?▾

Start drafts on Wan 3.0 Prime: it was the fastest in Flixly's test and gave results close to Wan 3.0. Switch to HappyHorse 1.1 or HappyHorse 1.0 when you want a clip that cuts between angles, a tilt-shift miniature look or, with HappyHorse 1.1, vertical and ultra-wide formats. Video editing on Flixly uses HappyHorse 1.0 from Alibaba.

Tools mentioned in this post

Wan 3.0Wan 3.0 PrimeHappyHorse 1.1HappyHorseAlibabaAI Video GeneratorText to VideoModel ComparisonSame Prompt TestFlixly

Ready to create with comparisons?

Jump straight into Flixly's AI studio and try comparisons with 193+ models — free to start.