Happy Horse 1.1: What Changed From 1.0
Happy Horse 1.0 reached the top of the Artificial Analysis Video Arena in April 2026, a month after its launch on April 27. Happy Horse 1.1 is the first major refinement since then. It is not a ground-up rebuild. The team behind the model, Alibaba's ATH AI Innovation Unit, took the five areas that most often caused re-renders under production deadlines and targeted each one.
Every capability in 1.1, including text-to-video, image-to-video, and reference-to-video, is exposed through the API with full parity to the web experience. The upgrade is available to creators and developers at the same time.
Happy Horse 1.1 is available now on vega.chat and through the api.vega.chat API.
The five upgrades
1. Stronger motion expressiveness
Happy Horse 1.1 improves motion modeling and temporal consistency. In the fight-scene example from the official comparison, the protagonist counters an opponent's punch using their own momentum, with body rotation and centrifugal force that follow a clear cause-and-effect chain. Where 1.0 could let motion drift or stall mid-action, 1.1 keeps body rotation, momentum transfer, and impact timing physically grounded throughout. The gap is most visible in shots with more than one overlapping motion, which is where 1.0 was most likely to look sluggish or incoherent.
This matters most for action shorts, dynamic social content, and cinematic sequences with continuous camera movement.
2. Better subject consistency and multi-reference fusion
Happy Horse 1.1 strengthens how the model reads and fuses multiple reference images in R2V tasks, keeping subjects closer to the reference assets across the full clip.
In the product-livestream example using two reference images (skincare and a lipstick), 1.1 holds product details and brand elements steady through the clip. In 1.0, those details could soften or shift toward the middle of a clip. In a two-character narrative example with a princess confronted by a dragon-man, 1.1 maintains each character's costume and facial identity without blending them across the shot.
This matters most for e-commerce product video, storyboard-driven production, and multi-character narrative content.
3. Improved instruction following
Happy Horse 1.1 improves long-context semantic understanding, scene planning, and character-relationship modeling. It produces more on-direction results under complex multi-scene, multi-character prompts.
In the minimalist living-room example, two characters move through a quiet emotional exchange with the restrained pacing and short-drama look the prompt called for, instead of the looser staging 1.0 sometimes produced. The biggest difference shows up in prompts that define a relationship dynamic. Happy Horse 1.1 reads "character A is confronting character B" as a motion cue, while 1.0 occasionally treated such lines as purely descriptive.
One prompt tip from the comparison material: when writing R2V prompts that describe interaction between references, name the relationship directly. "[image 1] and [image 2] start fighting" works better than "two characters engage in combat." Keep prompts focused. Over-constrained prompts can suppress the motion engine even in 1.1.
This matters most for short drama, multi-scene ad scripts, and storyboard execution.
4. Higher visual quality and more realistic detail
Happy Horse 1.1 refines character rendering, particularly facial detail generation. The over-sharpening, oily highlights, and smeared skin textures that could appear in 1.0 close-ups are reduced.
In the cinematic classroom example, 1.1 renders skin with more natural texture and pore-level detail while warm afternoon sunlight rakes across the desks and fine dust drifts through the light. In 1.0, the same shot could look over-sharpened or glossy. Close-up character shots gain expressiveness, and 1.1 shows a steadier grasp of professional camera language such as shot-reverse-shot and tracking shots, so multi-shot transitions and pacing feel more coherent.
One limitation: in very high-motion close-ups (fast head turns, rapid expression changes), both versions can still produce mild facial warping. Happy Horse 1.1 handles these better than 1.0, but the issue is not fully eliminated.
This matters most for short drama, advertising with hero talent shots, and brand content where faces fill the frame.
5. Upgraded audio expression
Happy Horse 1.1 improves audio-visual synchronization and prompt-based audio interpretation. Dialogue delivery pace and tone shift to match scene context, and sound-related prompt descriptions are followed more reliably.
In the kitchen-at-dusk example, the soft simmer of a pot, the rhythm of a chopping board, and the restrained dialogue line up with the on-screen action. The chop of a knife lands with the cutting motion rather than slightly ahead of or behind it, as could happen in 1.0. Happy Horse 1.1 also keeps ambient layers tracking the picture more consistently and introduces fewer irrelevant audio artifacts not referenced in the prompt.
This matters most for short drama, lifestyle brand content, and any clip where ambient sound is part of the narrative.
Quick comparison
| Capability | Happy Horse 1.0 | Happy Horse 1.1 |
|---|---|---|
| Motion expressiveness | Good, but drifts in fast action | Smoother, physically grounded |
| Subject consistency | Solid single-subject | Stronger multi-reference fusion |
| Instruction following | Reliable on simple prompts | Better on complex multi-character prompts |
| Visual quality | Can over-sharpen skin | More natural skin and richer detail |
| Audio expression | Synced but occasionally offset | Tighter AV sync, fewer artifacts |
| Generation modes | T2V / I2V / R2V | T2V / I2V / R2V (all upgraded) |
| API support | Yes | Yes, full capability parity |
| High-motion close-up faces | Some warping | Improved but not fully resolved |
In real production
Reading the differences through actual production jobs makes the value concrete. For short drama creation, the motion, instruction-following, and audio gains keep multi-shot dialogue scenes coherent and emotionally readable, so fewer takes get thrown out mid-production. For e-commerce advertising, the reference-to-video consistency in 1.1 protects product labels and brand color values that 1.0 could soften by the mid-clip. For brand marketing and character CG, the higher visual fidelity and steadier camera language push 1.1 output closer to broadcast-ready without extra compositing passes.
Across every category, Happy Horse 1.1 reduces the re-renders that 1.0 occasionally forced. On a real production schedule, that is where the upgrade pays for itself.
Should you upgrade?
Upgrade if you make short dramas, character-driven stories, e-commerce product video, or brand and advertising content. These are the workflows where the motion, consistency, visual-quality, and audio improvements compound, and where 1.0's weaker spots were most likely to force re-renders.
There is less urgency if you only generate quick single-shot social clips for fast ideation. Happy Horse 1.0 was already capable there. Even so, because all 1.1 capabilities ship through the API with full parity, there is little practical reason to stay on the older version once you have access.
If you are working within a tight credits budget, reserve Happy Horse 1.1 for final delivery scenes where quality is non-negotiable, and use 1.0 for early-stage concepting and rough cuts. Rough out concepts in 1.0 for speed, then move final scenes into 1.1 for production-grade motion, consistency, and audio.
Try Happy Horse 1.1 on vega.chat
Happy Horse 1.1 is available now on vega.chat and through the api.vega.chat API. You can generate clips in the chat interface, compare output side by side with 1.0, or integrate the model into your applications via the OpenRouter-compatible API at api.vega.chat. All three generation modes (T2V, I2V, R2V) are supported, including multi-image reference input, flexible aspect ratios, and 720p/1080p output.