HappyHorse 1.1:
lock a look across nine reference images
HappyHorse 1.1 is Alibaba's video model, and its place in our lineup is references: it takes up to nine images on reference to video, tied with Seedance 2.0 for the most we accept. It renders 720p or 1080p at 56 or 72 credits a second, 3 to 15 seconds a clip, in nine aspect ratios including 4:5. It also edits existing video and speaks dialogue you quote in the prompt.
720p / 1080p · 3–15 s · 9 ratios · 9 references · video edit · speech from quoted dialogue · rates in credits
When HappyHorse 1.1 is the right pick
HappyHorse earns its rate when a brief arrives with a pile of references, or needs a feed shape no other model we run will render. When the job is cheap drafts or 4K, it is the wrong tool.
Reach for it when
- You have product angles, a character sheet and a palette, and want all of them in one request.
- The delivery is 4:5 or 5:4 for a feed, or 9:21 and 21:9. HappyHorse is the only 4:5 video model we run.
- A character needs to say a line and you have no recording. Quote it in the prompt.
- You need to edit a clip longer than Wan 2.7 accepts. HappyHorse takes a source of up to 60 seconds and edits its first 15.
Look elsewhere when
- You need 4K or a cheap 480p draft. HappyHorse renders 720p and 1080p only.
- You rely on negative prompts. HappyHorse rejects them.
- A recorded voice track has to drive the performance. That is Wan 2.7.
- You need extension or a first-and-last-frame transition. HappyHorse has neither.
HappyHorse 1.1 pricing in credits
56 credits a second at 720p and 72 at 1080p, on text, image and reference shots. If a job fails or is cancelled, the hold is reversed.
Video edit bills the source seconds plus the output seconds at the same two rates. We treat the edit rate as provisional, so it may change once the provider's edit pricing is confirmed.
What a month of credits buys on HappyHorse
Hobby's 500 credits cover one 5-second 720p shot with 220 credits to spare, so regular HappyHorse work starts on Pro. Packs of 600, 2,500 and 5,000 credits never expire. The Playcut pricing page lists what each plan includes, and the AI video credit calculator works out how many clips a plan covers.
Edit billing, worked through
A 10-second source edited at 720p comes back as a 10-second output and bills (10 + 10) × 56 = 1,120 credits. The output matches the source length up to 15 seconds, and a longer source is edited from its first 15 seconds only.
Credits are held for the maximum up front, and the difference is refunded once the provider reports the exact billed duration. Trim the source to the segment you want before uploading, so the edit lands on the part you care about.
Reference to video with HappyHorse 1.1, step by step
HappyHorse runs inside Playcut on your Playcut credits, with no Alibaba Cloud account to open. Here is the reference-to-video path, which is what the nine-image ceiling is for.
- 1. Switch the model and the mode. In the video studio at app.playcut.ai, change the picker from Veo 3.1 to HappyHorse 1.1 and choose reference to video.
- 2. Add up to nine images, in order. JPEG, PNG or WEBP, with the shortest side at least 400 pixels and 20 MB at most per image.
- 3. Name each image in the prompt. Use [Image 1], [Image 2] and so on, with brackets, in the order you added them. Describe the subject next to its token: "the green glass bottle in [Image 1]" holds better than a bare token.
- 4. Pick ratio, resolution and length. Nine ratios, 720p or 1080p, and 3 to 15 seconds.
- 5. Quote any spoken line. Write it into the prompt, for example a barista saying "Careful, it's hot." The model voices it; custom voices are not accepted.
A reference prompt that follows HappyHorse's rules
Slow orbit around the green glass bottle from [Image 1] on the stone plinth from [Image 2], label turning to face the camera by the end, soft daylight matching [Image 3]. No text overlays and no people in frame.The last sentence is where a negative prompt would go on other models. HappyHorse rejects the negative-prompt field with an error, so exclusions live in the main prompt. Keep the whole prompt under 5,000 characters, because anything past that is cut off without warning.
More prompts written to the same rules, with spoken lines and exclusions kept in the main text, are in the HappyHorse prompts for text, photos, casts and edits.
4:5 and the other feed shapes
HappyHorse is the only video model we run that renders 4:5, and it also covers 5:4 and the tall 9:21. Ratios apply to text and reference shots; image to video and edit follow the shape of the source. The 4:5 aspect ratio spec lists the pixel sizes and the platforms that use it.
For a feed campaign, a 4:5 cut rendered natively avoids cropping a 16:9 or 9:16 master. The Instagram ads page covers the Reels and feed workflow in Playcut end to end.
From code, over MCP or REST
Pass model: "happyhorse-1.1" to the text, image, reference and video-edit tools. Keys and client setup are on the MCP server and REST API page.
What HappyHorse 1.1 rejects, truncates or leaves out
- Edit runs on HappyHorse 1.0
- There is no 1.1 edit model upstream, so the edit mode runs on Alibaba's HappyHorse 1.0 edit model under the same happyhorse-1.1 name on Playcut. Its credit rate is provisional.
- Negative prompts are rejected
- Sending one returns an error instead of being quietly ignored. Fold exclusions into the prompt itself.
- Long prompts are truncated silently
- The cap is 5,000 characters for non-Chinese text, and the overflow is dropped with no error. Put the details that matter first. The guide to fixing a weak AI video prompt covers this and the other rules that quietly change a prompt.
- No audio input
- Speech comes from dialogue quoted in the prompt. Audio files and custom voices are rejected, so a clip that has to follow your own recording belongs on Wan 2.7.
- 720p and 1080p only
- There is no 480p draft tier and no 4K. Playcut always sends the resolution and defaults to 720p, so you are never billed at 1080p by accident.
- No interpolation or extension
- Both are rejected. Seedance 2.0 bridges a first and last frame, and Veo 3.1 extends a clip.
- Image to video takes no ratio
- The output follows the source frame. To get a 4:5 animate shot, start from a 4:5 still.
HappyHorse vs Seedance: same ceiling, different jobs
Both take up to nine reference images, the most we accept on any video model. After that they split on almost every axis. Rates are as of 30 September 2026.
How to choose between them
Pick HappyHorse when the delivery shape or an edit decides it: 4:5 feed cuts, 9:21, or reworking a clip you already have. Pick Seedance when cost or resolution decides it, or when you want a first-and-last-frame transition.
Seedance's face rule is the other deciding factor for image-driven shots. The Seedance 2.0 page explains what it blocks.
And the other edit model, Wan 2.7
Wan 2.7 also edits by prompt, at 40 credits a second at 720p, but caps the source at 10 seconds and does not take reference images on edit. It animates a still to an audio track you upload, which HappyHorse cannot. See the Wan 2.7 page.
Veo 3.1, at 160 credits a second, takes three references and two ratios. The Veo 3.1 page covers when its extension mode is worth that rate.
The full price list for every model is on the AI models comparison hub, and the AI video generator page covers working across all seven video models in one studio.
Who makes HappyHorse, and which version we run
HappyHorse is Alibaba's. CNBC reported in April 2026 that Alibaba had been revealed as the company behind HappyHorse 1.0, and version 1.1 followed.
Playcut runs HappyHorse 1.1 through Alibaba's DashScope international service for text, image and reference shots, and the 1.0 edit model for edits. Alibaba's own HappyHorse API reference covers the provider side.
The provider stamps a watermark by default, and Playcut switches it off on every call. HappyHorse 1.1 also powers the Plus tier of Act, the part of Playcut that performs your AI actors in a scene, at that app's own rates.
Building a reference set that holds up
Order is load-bearing. The first image you add is [Image 1], and a prompt that names the tokens out of order points the model at the wrong picture.
Give each image one job, such as the product, the setting or the light, and say which job in the prompt. Nine is a ceiling, not a target, so leave out any image that does not change the result.
Save the prompt to your prompt library once it lands. Typing / in an empty prompt box inserts it next time, token order and all, which keeps a campaign's shots consistent across sessions.
HappyHorse 1.1 for developers
The same modes are open over MCP and the REST tools API. Field names, error semantics and request examples live in the HappyHorse 1.1 API reference on docs.playcut.ai.
First time in the studio? The guide to using Playcut covers sessions, the model picker and credits.
Frequently asked questions
Who made HappyHorse 1.1? +
Alibaba. It was revealed as the company behind HappyHorse 1.0 in April 2026, and HappyHorse 1.1 is the version that followed. On Playcut it runs through Alibaba's DashScope service and bills in Playcut credits.
How much does HappyHorse 1.1 cost on Playcut? +
56 credits a second at 720p and 72 at 1080p. A 5-second 720p shot is 280 credits and a 10-second 1080p shot is 720. Video edit bills source plus output seconds at the same rates, which are provisional for edit. Paid plans start at $9 a month for 500 credits.
How many reference images does HappyHorse 1.1 take? +
Up to nine on reference to video, tied with Seedance 2.0 for the most we accept on video. Video edit takes up to five for swaps, and image to video takes one first frame. Name them in the prompt as [Image 1], [Image 2] and so on, in the order you added them.
Can HappyHorse 1.1 generate spoken dialogue? +
Yes. Quote the line in the prompt and the model speaks it in the clip. There is no audio input and no custom voice, so if a clip has to follow a recording you already have, use Wan 2.7's image to video with driving audio instead.
Should I use HappyHorse or Seedance? +
Choose HappyHorse for 4:5 or 9:21 delivery, video edit, or dialogue spoken from the prompt. Choose Seedance for lower rates at 720p, the 480p and 4K tiers, or a first-and-last-frame transition. Both take nine references, and Seedance rejects input images that contain human faces.
Can I use a negative prompt with HappyHorse 1.1? +
No. HappyHorse rejects the negative-prompt field with an error rather than ignoring it. Put what you want to exclude into the main prompt instead.
Ready to make something cinematic?
Pick a plan. Every plan bills at checkout — cancel anytime.