On June 22, Alibaba released its newest video generation model, HappyHorse 1.1. This generation focuses on five capability upgrades: dynamic expressiveness, subject consistency, instruction following, visual texture, and audio. The technical specs are unchanged from 1.0 — 3 to 15 seconds per generation, with 720p / 1080p resolutions and free-form aspect ratios.

The model is now live on the Alibaba Cloud Bailian platform and the HappyHorse official site. Two practical changes stand out: first, the "greasy look" that got the previous generation roasted has visibly improved; second, the price of 1080P dropped 25%. This piece breaks down the hands-on results across the five dimensions so you can judge whether it's worth switching.

Pricing: 1.1 Is Cheaper than 1.0

Where to try it:

• Official site: www.happyhorse.cn
• API access: bailian.console.aliyun.com

Dimension One: Dynamic Expressiveness — The Slow-Motion Problem Is Fixed

One of the pain points that got HappyHorse 1.0 roasted was sluggish motion and weak pacing in some shots. Version 1.1 improves motion modeling and temporal consistency.

Tested with a motorcycle-riding case:

• HappyHorse 1.1: on-screen speed is normal and follows basic physics; pushed to a close-up, the scenery reflected in the windshield is also fairly logical. The only small blemish is that the motorcycle taillight's glow looks slightly unreal.
• HappyHorse 1.0: the same task shows obvious slow motion; the motorcycle rides against traffic in-frame, and the helmet reflection doesn't match the actual scene.

Verdict: dynamic modeling genuinely fixed the previous generation's "slow-motion feel."

Dimension Two: Subject Consistency — 9 Reference Images at Once

HappyHorse 1.1 supports 9 character reference images input simultaneously, flexibly combining product details, brand elements, characters, and scenes. Its grasp of popular tricks like multi-panel storyboards and N-grid image references has also improved.

We uploaded three reference images depicting a specific character's resignation and had 1.1 and 1.0 each generate a 10-second video:

• HappyHorse 1.1: faithfully reproduces the character's face and clothing; scene and character details stay stable and consistent across both shots, down to the corners of the frame
• HappyHorse 1.0: mostly maintains subject consistency, but more physics bugs show up in the frame

Verdict: in multi-reference-image scenarios, 1.1's stability clearly wins — well suited to continuous shots of branded IP characters.

Dimension Three: Instruction Following — Back and Forth with Seedance 2.0 Mini

We ran a comparison test with a surreal zero-gravity café scene, using this prompt:

Inside a modern-style café, gravity suddenly vanishes. Customers, tables and chairs, books, and various objects float slowly into the air. The barista keeps making coffee while floating, and liquid coffee spills from the cup, forming countless floating liquid spheres. An orange cat glides slowly through the space as if swimming in water. The camera keeps rotating and moving freely, showing the whole zero-gravity environment. All floating objects must obey real inertia and momentum, and the liquid motion must match fluid physics. The whole piece should convey extreme realism and complex physical simulation.

Comparison results:

• Both models reproduce the details one by one in prompt order
• HappyHorse 1.1: more continuity errors — characters' expressions go blank, and a chair springs out of the floor from nowhere
• Seedance 2.0 Mini: the way liquid floats in a vacuum-like state doesn't quite follow physics, but the characters' expressions fit the overall style better

Verdict: in extreme-surreal, physics-heavy scenarios, 1.1 hasn't opened up a lead and still has room to improve. For ordinary shots, its instruction following is solid.

Dimension Four: Visual Texture — The "Greasy Look" Clearly Improved

We asked for "the Chinese national team scoring a goal in the World Cup final" to stress-test texture in a large-crowd scene:

• On its main subjects, 1.1 shows much less of the previous generation's "greasy look" and "over-sharpening"
• But in the background of the frame, faces are already somewhat blurred, and the realism and sense of motion fall a bit short

Verdict: foreground texture has clearly improved; background-crowd realism is the shortcoming to fix next.

Dimension Five: Audio — No Obvious Improvement This Generation

Comparing 1.1 and 1.0 on an instrument-performance scene showed no obvious difference — the visuals of the performance and the audio still don't line up. This is the weakest item in this generation's upgrades.

Strengths and Where It Fits

Strengths:

• Cheaper: 1080P down 25%, with standout value after the discount
• Better visual texture: the greasy look and over-sharpening are greatly reduced
• Strong multi-reference-image support: 9 images at once, well suited to brand IP / advertising scenarios
• Complete ecosystem: dual entry via the official site + Bailian API, ready for engineering integration

Where it fits:

• Short-video creators: quickly producing 3–15 second story clips and camera-move material
• E-commerce / advertising: multi-reference-image product ads with a consistent protagonist and product
• Long-video teams: as a storyboard material generator, combined with editing tools into finished cuts

💡 Tip: HappyHorse 1.1's upgrade basically matches expectations for a minor-version iteration — solid optimization of the real problems the previous generation exposed, while audio sync, heavy physics simulation, and background-crowd realism still need polish. If your workflow is cost-sensitive and relies on multi-reference images for character consistency, 1.1 is worth switching to; if you mainly produce ultra-realistic long-form content, consider pairing it with Seedance 2.0 Mini and routing shots by type.