HappyHorse 1.1 vs Seedance 2.0: Same-Prompt Test and Price

We gave HappyHorse 1.1 and Seedance 2.0 the same three 5-second 720p prompts on October 26, 2026, and Seedance 2.0 won all three: the skate jump, the noodle stall and the product shot. HappyHorse 1.1 rendered about twice as fast. At list price, a 5-second 720p clip costs $0.70 on HappyHorse and $0.76 on Seedance.

Need 4K, a last frame or video references? Only Seedance 2.0 has them. HappyHorse 1.1 stops at 1080P and takes one first frame or up to 9 reference images.

Both models are on OmniaKey: HappyHorse 1.1 and Seedance 2.0.

How we tested

• Models and IDs: happyhorse-1.1-t2v and happyhorse-1.1-i2v for HappyHorse, dreamina-seedance-2-0-260128 for Seedance 2.0, all through the OmniaKey API.
• Settings we sent, exactly: HappyHorse text to video {"duration": 5, "resolution": "720P", "ratio": "16:9"}, HappyHorse image to video {"duration": 5, "resolution": "720P"}, Seedance text to video {"duration": 5, "resolution": "720P", "generate_audio": true, "ratio": "16:9"}, Seedance image to video {"duration": 5, "resolution": "720P", "generate_audio": true, "ratio": "adaptive"}.
• We ran everything on October 26, 2026: 3 prompts, 1 clip per model per prompt, 6 clips in all, and every task succeeded on the first try.
• The prompts are the same three we used for Kling 3.0 vs Wan 3.0, and the HappyHorse clips here are the same ones in HappyHorse 1.1 vs Kling 3.0.
• The first frame for test 3 was made with Nano Banana 2.1 (2752 × 1536, 16:9), so neither model animated its own image.
• Time is from sending the job to the task showing succeeded, queue time included.
• We watched every clip and compared frames at the same moments. We asked whether the action and the camera move match the prompt, whether the sign and label text is right, and whether the first frame holds. Sound was checked by loudness, two AI audio models and a spectrogram.
• OmniaKey sells API access to both models, so we ran the tests through it. Every price below is Alibaba's or BytePlus's own list price, not ours.
• Sources: Alibaba Cloud Model Studio's pricing page and the HappyHorse text-to-video API reference, plus BytePlus ModelArk's pricing page and the Seedance 2.0 page, each checked October 26, 2026.

That is one clip per prompt per model, and video models vary from run to run.

Three tests, same prompts

Test 1: Skate ollie over a puddle (text to video, 16:9)

This test checks whether the board clears the water or lands beside it, and how much of the skater the camera keeps in frame.

Prompt:

Tracking shot at golden hour on an empty concrete plaza. A teenage skateboarder in a mustard hoodie rolls toward a shallow puddle, pops an ollie over it and lands cleanly on the far side. The rear wheel clips the edge of the puddle and throws a small arc of water. Low warm sun, long shadows, handheld camera following at hip height. Sound: wheels rumbling on concrete, the clack of the board landing, a splash.

• HappyHorse 1.1 (97.6 s): The hip-height camera frames the skater from the waist up for the first 1.5 s, so the board only comes into view near the puddle. He pops a small ollie at about 2.7 s beside the puddle and lands on dry concrete at about 3.1 s, his head now out of frame. Near the end his wheels kick up a little water from a wet streak.
• Seedance 2.0 (144.8 s): A wide tracking shot keeps the whole skater in frame. He pops at about 2.0 s right at the puddle, and the board slaps the water, throws a big splash and turns over under him in the air. He lands past the puddle at about 4.0 s and rolls toward the camera.

Verdict: Seedance 2.0. It did the move the prompt describes: over the puddle, a splash, and a clean landing on the other side. HappyHorse jumped next to the puddle and kept most of the skater's body out of frame. For action shots, Seedance followed the script.

Test 2: Night noodle stall with a sign (text to video, 16:9)

This test checks the sign text, the slow push-in, and whether the cook takes the paper ticket.

Prompt:

Night street food stall, medium shot with a slow push-in. An elderly cook in a white apron tosses noodles in a flaming wok; steam and sparks rise under a single hanging bulb. Behind him a red neon sign reads NOODLES in capital letters. A customer on the right hands over a paper ticket. Wet pavement reflects the neon. Sound: sizzling oil, the clang of the wok, distant street chatter.

• HappyHorse 1.1 (100.7 s): The "NOODLES" sign is correct. A white-haired cook in a white T-shirt and apron works a wok with flames and steam under a hanging bulb, with a red, wet-looking ground reflecting the neon and a wall behind him. The camera pushes in slowly, and the cook stirs the noodles but never tosses them. A customer comes in from the right with a paper ticket at about 1 s and hands it over, and by 4.9 s the cook is holding it and reading it.
• Seedance 2.0 (319.0 s): The "NOODLES" sign is correct, with a second neon sign above it in made-up lettering. The cook tosses noodles high out of the flaming wok at 1.0 s, at 4.9 s a big flame throws sparks, and a wet street with red reflections sits behind the stall. A customer's hand holds out a paper ticket from about 2 s, and the cook keeps cooking and does not take it.

Verdict: Seedance 2.0, narrowly. It gave the noodle toss, the sparks and the wet street from the prompt. HappyHorse finished the ticket handover and added no extra text, but its cook only stirs. If you need clean signs, add "no other signs" to a Seedance prompt.

Test 3: Perfume bottle orbit from a still photo (image to video, 16:9)

This test checks whether a product photo survives a camera move with the cap, the label and the small type intact. Nano Banana 2.1 made the frame first, then both models animated it.

First-frame prompt:

Studio product photograph of a square glass perfume bottle with a gold cap standing on wet black slate. The front label is cream paper with the word AURELIA in black serif capitals and "eau de parfum · 50 ml" in small type underneath. Water droplets on the glass and the stone, a soft key light from the left, dark background, wide 16:9 frame with the bottle in the center.

Motion prompt:

The camera slowly orbits to the right around the perfume bottle. Water droplets slide down the glass and a soft beam of light sweeps across the label. Keep the bottle, the gold cap and the label text exactly as they are. Sound: a quiet drip of water.

• HappyHorse 1.1 (91.7 s, output 1286×716): Keeps the bottle, the gold cap and the "AURELIA" label, and orbits to the right so the bottle is seen at an angle by 4.9 s. A soft band of light crosses the label at about 2.5 s. It adds drifting smoke behind the bottle, the liquid turns a deeper amber, and the small "eau de parfum · 50 ml" line is a little soft by the end.
• Seedance 2.0 (184.7 s, output 1280×720): Keeps the bottle, the cap, both label lines and the pale liquid unchanged and sharp through the clip, and orbits to the right, so the bottle is seen at an angle by 4.9 s. The label brightens as light passes at about 1.5 s, and drops of water fall around the bottle.

Verdict: Seedance 2.0. Both moved the camera as asked and kept the label readable. Seedance changed nothing else, while HappyHorse added smoke and tinted the liquid, which matters for a product shot.

Results by test

Sound, speed and price per clip

Sound

Every clip came back with a sound track, and none had speech. Both perfume clips have the dripping water the prompt asked for, and both skate clips have the clack of the board. Our checks disagreed on what the noodle-stall tracks contain, so for those we give only the loudness.

Mean loudness per clip, where closer to 0 is louder:

Alibaba's docs don't mention sound for HappyHorse, but all three of its clips had it. Seedance's generate_audio is on by default, and BytePlus's 2.0 price table has one price, with no separate rate for sound.

Watermark

Alibaba's watermark setting is on by default and stamps the mark in the lower-right corner, which is why every HappyHorse frame shows a small "Happy Horse". Adding "watermark": false to the parameters removes it, and we confirmed that on a separate 3-second clip. The Seedance clips had no mark.

Speed

Time from sending the job to succeeded, queue included:

HappyHorse finished every clip first, and its slowest clip was still faster than Seedance's quickest.

Price

Alibaba bills HappyHorse per second of output video, and input is free, per Alibaba Cloud Model Studio's pricing page checked October 26, 2026. We quote its International (Singapore) prices. The text-to-video, image-to-video and reference-to-video IDs share one price.

BytePlus bills in tokens and prints a per-video example for a 5-second 16:9 clip, per its ModelArk pricing page checked October 26, 2026.

What a whole clip costs at those rates:

At 480P a 5-second clip is $0.35 at both list prices, at 720P the two are close, and at 1080P HappyHorse costs about half. BytePlus also sells Seedance 2.0 Fast at $0.60 and Seedance 2.0 Mini at $0.38 per 5-second 720p video, which we did not test.

For OmniaKey's current rates, see the HappyHorse 1.1 and Seedance 2.0 model pages.

What about Seedance 2.5?

ByteDance has a newer video model, Seedance 2.5, which makes clips up to 30 seconds. We tested 2.0 because that is the pair people search for against HappyHorse. Our Seedance 2.5 review covers the newer model.

Pros and cons

HappyHorse 1.1

• Pros: fastest on every clip (91.7–100.7 s), half Seedance's list price at 1080P, the same price at 480P, clean signs, finished the ticket handover, nine aspect ratios.
• Cons: jumped beside the puddle, no noodle toss, added smoke and changed the liquid color in the product shot, watermark on by default, one first frame only, no last frame and no video references, tops out at 1080P, sound not mentioned by Alibaba.

Seedance 2.0

• Pros: the most exact action, left the product shot untouched, sound on by default and in BytePlus's docs, 4K, first and last frame, image, video and audio references, edit and extend.
• Cons: slower, with the noodle clip at 319.0 s, made-up text on a background sign, no input images or videos with real faces, $1.87 per 5 s at 1080P.

Which one for which job

Run the same test through one API

This request sends the noodle stall prompt to HappyHorse 1.1, with the watermark switched off:

curl https://api.omniakey.com/v1/media/tasks \
  -H "Authorization: Bearer $OMNIAKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: noodle-stall-001" \
  -d '{
    "model": "happyhorse-1.1-t2v",
    "type": "video",
    "operation": "text_to_video",
    "input": {"prompt": "Night street food stall, medium shot with a slow push-in. An elderly cook in a white apron tosses noodles in a flaming wok; steam and sparks rise under a single hanging bulb. Behind him a red neon sign reads NOODLES in capital letters. A customer on the right hands over a paper ticket. Wet pavement reflects the neon. Sound: sizzling oil, the clang of the wok, distant street chatter."},
    "parameters": {"duration": 5, "resolution": "720P", "ratio": "16:9", "watermark": false}
  }'

Leave "watermark" out to get the default mark, as in our test clips.

1. The reply is 202 with a task id.
2. Fetch https://api.omniakey.com/v1/media/tasks/{id} every few seconds until status is succeeded. A video job can take several minutes, so keep polling the same id instead of resubmitting.
3. Download the clip from assets[0].url.

For Seedance 2.0, set model to dreamina-seedance-2-0-260128 and parameters to {"duration": 5, "resolution": "720P", "generate_audio": true, "ratio": "16:9"}. To animate a photo with HappyHorse, set model to happyhorse-1.1-i2v, operation to image_to_video, and send the frame in input.media with type set to first_frame and a url that is an https link or a data URL. Drop ratio for HappyHorse image to video, because the frame sets the shape.

Open the HappyHorse 1.1 and Seedance 2.0 model pages to start a run, and the video API docs for every parameter.

Common questions

Is HappyHorse 1.1 better than Seedance 2.0?

For our three prompts, no. Seedance 2.0 followed the action more closely each time: it jumped over the puddle, tossed the noodles and left the product shot alone. HappyHorse 1.1 was faster, cost about half at 1080P and added no made-up text.

Which is cheaper, HappyHorse 1.1 or Seedance 2.0?

At 480P a 5-second clip is $0.35 at both list prices, and at 720P they are close: $0.70 on HappyHorse against $0.76 on Seedance. At 1080P HappyHorse is about half, at $0.90 against $1.87 for 5 seconds.

Does HappyHorse 1.1 generate sound?

Every HappyHorse clip in our test came back with a sound track, including the dripping water we asked for in the perfume prompt. Alibaba's docs list no audio parameter, default or price for it. So treat the sound as something the model returned in our run, not as a setting you can control.

How do I remove the HappyHorse watermark?

Add "watermark": false to the request parameters. The mark is on by default and sits in the lower-right corner, and we confirmed the switch on a separate 3-second clip that came back clean. In our test clips the mark reads "Happy Horse".

Is HappyHorse 1.1 open source?

Not that we could find. Alibaba offers HappyHorse 1.1 as an API on Alibaba Cloud Model Studio, and a Hugging Face search for "happyhorse" and "happy horse" on October 26, 2026 found no HappyHorse model.

Is it HappyHorse or Happy Horse?

Alibaba's docs write it as one word, HappyHorse. The watermark on our clips reads "Happy Horse", and plenty of people search that way, so both spellings point to the same model.

Update log

• October 27, 2026: first version; the tests ran on October 26, 2026.