HappyHorse 1.0 API
Arena-leading video with native audio on fal.ai
10 guides covering HappyHorse 1.0 end to end. Elo-ranked quality, seven-language lip sync, 1080p delivery. Real numbers, real code, real pipelines.
FREQUENTLY
ASKED.
01Is HappyHorse 1.0 really number one on the Arena?
Yes, as of early 2026 HappyHorse 1.0 holds the top slot on the Artificial Analysis Video Arena on both legs of the benchmark, with an Elo of 1365 for text to video and 1397 for image to video. Those numbers sit above Seedance 2.0, Kling 3.0 Pro, Veo 3.1, and Runway Gen-4.5 on the public leaderboard. The fal endpoint that exposes this quality is `fal-ai/happyhorse/v1/text-to-video`. The Arena is a head to head Elo system based on blind human votes, so the ranking reflects perceived quality, not a single cherry picked clip.
02Which languages does the native lip sync cover?
Seven: English, Mandarin Chinese, Japanese, Korean, German, French, and Cantonese. You pass the dialogue inside your prompt and set the `language` input. The model generates the voice, the ambience, and the mouth shape in one pass, no external TTS or viseme matcher required. Word error rate on controlled sentences lands near the quality bar you would expect from a purpose built lip sync model. On `fal-ai/happyhorse/v1/text-to-video`, the seven language set is the main unlock compared to Seedance 2.0 and Veo 3.1, which are English heavy.
03What does HappyHorse 1.0 cost on fal.ai?
During partner preview, the illustrative rate is around twenty cents per second on `fal-ai/happyhorse/v1/text-to-video`, billed by rendered duration. A 5 second 720p shot with native audio on works out to roughly one dollar, a 10 second 1080p job lands near two dollars. These numbers are provisional and will be confirmed at fal.ai/pricing once the endpoint moves from preview to GA. The Seedance 2.0 fallback bills at its standard published rate, so your pipeline cost does not spike while you wait for allowlist.
04How do I get access during partner preview?
HappyHorse 1.0 is in partner preview on fal.ai, which means access to `fal-ai/happyhorse/v1/text-to-video` is granted per account rather than open to every FAL_KEY. While your account is being allowlisted, your code calls fall through cleanly to `fal-ai/seedance-2.0/text-to-video` without a manual swap, because the SDK returns a clear preview error you can branch on. Once allowlisted, you flip a single string literal and the same input schema keeps working.
05Can I run image to video, not just text to video?
Yes. `fal-ai/happyhorse/v1/text-to-video` is the T2V endpoint, and the I2V sibling lives at `fal-ai/happyhorse/v1/image-to-video`. I2V accepts a still plate and animates it while preserving color, lighting, and composition. The same language and audio controls apply, so you can drive dialogue on an existing portrait. Arena Elo on the I2V track is 1397, which is currently the highest public number for a hosted image to video model.
06What are the duration and resolution caps?
Single shot duration caps at 10 seconds on `fal-ai/happyhorse/v1/text-to-video`. Resolutions are 256p for drafts, 480p, 720p, and 1080p. Aspect ratios cover 16:9, 9:16, 1:1, and 21:9. For longer scenes, chain shots in your pipeline using the seed input to keep characters and lighting stable across cuts. Avoid driving 10 second 1080p with audio on for every iteration; 5 second 256p draft renders cost a fraction and validate the prompt before you commit.
07What happens if a job fails, times out, or hits a filter?
The fal async queue handles `fal-ai/happyhorse/v1/text-to-video` like any other job. If the preview endpoint is gated for your account, the SDK returns a clear preview error and the fallback path switches the call to `fal-ai/seedance-2.0/text-to-video` so your pipeline does not go dark. Content filter hits return a structured error with the rule id, not a silent null. Transient queue failures retry once automatically; hard failures surface in the webhook payload so you can log, alert, and re render with a softer prompt.
08How does HappyHorse 1.0 compare to Seedance 2.0, Kling 3.0 Pro, and Veo 3.1?
HappyHorse 1.0 tops all three on Arena Elo and is the only one of the four with native seven language lip sync at this quality tier. Seedance 2.0 on fal is cheaper per second and still excellent for silent or English only shots. Kling 3.0 Pro is strong on cinematic motion and character consistency across longer cuts. Veo 3.1 wins on prompt adherence and brand safety at a higher per second cost. For the top quality slot with dialogue in any supported language, `fal-ai/happyhorse/v1/text-to-video` is the pick.
09Why run HappyHorse on fal.ai?
Eight reasons. One, partner preview access without a separate contract. Two, a single FAL_KEY that already speaks to 600+ models. Three, an async queue with webhooks and cancel semantics. Four, automatic fallback to `fal-ai/seedance-2.0/text-to-video` when preview access is gated. Five, serverless scale with no cold start tax on burst traffic. Six, a real time playground for `fal-ai/happyhorse/v1/text-to-video` that uses your live key. Seven, first party SDKs for TypeScript, Python, and Swift with identical schemas. Eight, pricing, usage, and logs in one dashboard alongside your other video models.
10How do I integrate HappyHorse 1.0 into a production pipeline?
Install `@fal-ai/client`, set FAL_KEY in your environment, and call `fal.subscribe('fal-ai/happyhorse/v1/text-to-video', { input })` with prompt, duration, resolution, aspect_ratio, generate_audio, and language. Wire `onQueueUpdate` to stream logs for long renders, and set a webhook on the job for server side fan out. For localization pipelines, iterate the same prompt across the seven supported languages in parallel, store results in your CDN, and pin the seed to keep motion identical across cuts. The fallback to Seedance 2.0 uses the same input schema, so no client code forks.
HappyHorse 1.0at a glance.
HappyHorse 1.0 is the current number one model on the Artificial Analysis Arena for both text to video and image to video generation, with an Elo of 1365 on T2V and 1397 on I2V as of early 2026. The model ships from Alibaba ATH and pairs raw visual quality with two capabilities that most of the field still treats as afterthoughts: native audio video joint synthesis and ultra low word error rate lip sync across seven languages. Those languages are English, Mandarin Chinese, Japanese, Korean, German, French, and Cantonese. You ask for a 5 second shot of a chef who says a line in Japanese, and HappyHorse hands back the render with the voice, the mouth shape, and the room tone in one file. No ADR pass, no viseme matching, no separate TTS stitch.
The tradeoffs you care about are predictable. On quality, HappyHorse is the current ceiling. On speed, it is competitive with Seedance 2.0 and Kling 3.0 Pro at 720p and only meaningfully slower when you push 1080p with native audio on. On price, the partner preview rate lands near twenty cents per second, which is above Seedance 2.0 and below Veo 3.1 at equivalent resolution. The creative handles that matter for directors are here: duration up to 10 seconds per shot, resolutions from a 256p draft for iteration through 1080p for delivery, aspect ratios covering 16:9 landscape, 9:16 portrait, 1:1 square, and 21:9 cinema, a deterministic seed input, and an image to video entry point that preserves the source look without re colorizing it.
Running HappyHorse 1.0 on fal.ai means you do not have to negotiate a partner access slot yourself. The endpoint is `fal-ai/happyhorse/v1/text-to-video`, it speaks the same async queue your other video jobs already use, and it fails over cleanly to `fal-ai/seedance-2.0/text-to-video` if the preview endpoint is gated at the moment you call it. The workflow surface that lights up is the set of jobs that were painful before: multilingual product explainers, dialogue heavy documentary B roll, localized ad variants where the same shot needs to exist in four languages, and narrative pieces where you want the character to actually talk rather than mouth along to an external voice track. That is the slice of the video market HappyHorse 1.0 was built to own, and on the Arena numbers, it does.
- >Agencies shipping localized ad variants across EN, ZH, JA, KO, DE, FR, and Cantonese
- >Indie creators who want native dialogue without a separate TTS plus lip sync pass
- >Product teams building explainer video generators that need real voices on camera
- >Research groups benchmarking the current Arena top slot on real production prompts
- >Documentary and short film directors chasing maximum raw quality per shot
- >You need the highest Arena ranked quality available right now on a hosted API
- >Your script has on camera dialogue and you refuse to bolt on an external voice track
- >You are shipping the same shot in three or more languages and need identical motion
- >You want 10 second shots at 1080p with native room tone and ambience baked in
- >You need I2V that respects the source plate without re lighting or re colorizing it
Running HappyHorse 1.0 on fal.ai gives you partner preview access, a single API key that already speaks to 600+ models, an async queue with webhooks, and automatic fallback to Seedance 2.0 when the preview endpoint is gated.
CALL HAPPYHORSE 1.0
IN UNDER 20 LINES.
1import { fal } from "@fal-ai/client";23fal.config({ credentials: process.env.FAL_KEY });45// HappyHorse 1.0 is in partner preview on fal.ai as6// fal-ai/happyhorse/v1/text-to-video. While your account is7// being allowlisted, this snippet targets the Seedance 2.08// fallback so the pipeline stays green. Swap the endpoint id9// once your preview access lands.10const result = await fal.subscribe("fal-ai/seedance-2.0/text-to-video", {11 input: {12 prompt: "A chef in a professional kitchen at dusk looks to camera and says: 'Tonight, the kitchen opens at seven.' Warm key light, shallow depth of field, soft extractor hum.",13 duration: "5s",14 resolution: "720p",15 aspect_ratio: "16:9",16 generate_audio: true,17 language: "en",18 },19 logs: true,20 onQueueUpdate: (update) => {21 if (update.status === "IN_PROGRESS") {22 update.logs?.map((log) => log.message).forEach(console.log);23 }24 },25});2627console.log(result.data.video.url);
{ video: { url: "https://v3.fal.media/files/zebra/abc123.mp4" }, seed: 42 }WHAT HAPPYHORSE 1.0 COSTS ON FAL.AI.
| ENDPOINT | RATE | EXAMPLE | COST |
|---|---|---|---|
| fal-ai/happyhorse/v1/text-to-video | $0.20 per second (preview) | 5s 720p audio on, English | $1.00 |
| fal-ai/happyhorse/v1/text-to-video | $0.20 per second (preview) | 10s 1080p audio on, Japanese | $2.00 |
| fal-ai/happyhorse/v1/image-to-video | $0.20 per second (preview) | 5s 720p I2V from a plate, silent | $1.00 |
| fal-ai/happyhorse/v1/text-to-video | $0.20 per second (preview) | 3s 256p iteration draft, silent | $0.60 |
| fal-ai/seedance-2.0/text-to-video (fallback) | $0.12 per second | 5s 720p audio on | $0.60 |
Preview pricing. HappyHorse 1.0 is in partner access; final pricing will be posted at fal.ai/pricing.
HAPPYHORSE 1.0 VS THE FIELD.
| MODEL | RES | DUR | PRICE | ELO | ENDPOINT | BEST FOR |
|---|---|---|---|---|---|---|
| HappyHorse 1.0 | 1080p | 10s | $0.20/s (preview) | 1365 T2V / 1397 I2V | fal-ai/happyhorse/v1/text-to-video | Top of Arena quality, 7 language lip sync, native audio |
| Seedance 2.0 | 1080p | 10s | $0.12/s | 1318 T2V | fal-ai/seedance-2.0/text-to-video | High quality generalist, native audio, cheaper per second |
| Kling 3.0 Pro | 1080p | 10s | $0.18/s | 1331 T2V | fal-ai/kling-video/v3/pro/text-to-video | Cinematic motion, strong character consistency |
| Veo 3.1 | 1080p | 8s | $0.40/s | 1342 T2V | fal-ai/veo3.1/text-to-video | Prompt adherence, brand safe outputs, premium tier |
| Grok Imagine | 720p | 6s | $0.15/s | 1289 T2V | fal-ai/grok-imagine/text-to-video | Fast iteration, loose prompts, meme velocity |
| Runway Gen-4.5 | 1080p | 10s | $0.22/s | 1301 T2V | fal-ai/runway/gen-4.5/text-to-video | Editorial polish, creative director UI |
HappyHorse 1.0 wins when you need the current Arena ceiling plus dialogue in any of seven languages; Seedance 2.0 wins on cost for silent or English only shots, and Veo 3.1 wins when prompt adherence matters more than raw quality.
Three to read first.
The posts we point people at when they ask where to start with HappyHorse.
Every topic we cover.
8 categories, 10 posts. Each tile opens one thread of HappyHorse coverage.
Comparison
- HappyHorse 1.0 vs Veo 3.1: The 60-Elo Gap Explained
- HappyHorse vs Seedance 2.0 vs Kling v3 Pro: Who Wins When
Technique
- Multilingual Lip Sync: Seven Languages in One Pass
- Native Audio and Joint Synthesis: When to Let the Model Do Both
Prompting
- Prompting HappyHorse: Shot Descriptions That Land
Multi-shot
- Image-to-Video: Using a First Frame to Pin the Shot
Optimization
- Budget Math: 256p Drafts and 1080p Delivery
Use Case
- A Cinematic Trailer Pipeline: Draft on Seedance, Deliver on HappyHorse
Troubleshooting
- Debugging Lip Sync: Why the Mouth Is Out of Step
Integration
- Integrating HappyHorse into a Postgres-Backed Render Queue
More on Comparison.
The category with the most coverage. 2 posts in this thread.
Latest posts.
HappyHorse vs Seedance 2.0 vs Kling v3 Pro: Who Wins When
Native Audio and Joint Synthesis: When to Let the Model Do Both
Budget Math: 256p Drafts and 1080p Delivery
A Cinematic Trailer Pipeline: Draft on Seedance, Deliver on HappyHorse
Debugging Lip Sync: Why the Mouth Is Out of Step
Integrating HappyHorse into a Postgres-Backed Render Queue
The numbers.
What this publication is and isn't, in numbers.
Each one is dated, second-person, and opinionated.
Filter by the constraint you care about.
Total length of every post in the archive.
Not a single U+2014 survives our ship check.
Editor-selected cover stories.
Custom covers on every featured post.
What we write about most.
Keyword frequency across every post. The bigger the word, the more often we come back to it.
Keep reading.The full blog is open.
No gates, no sign-up, no newsletter. Just 10 dated posts on HappyHorse.
Browse the full blog
Sort by date, filter by category, search by keyword.
HappyHorse 1.0 vs Veo 3.1: The 60-Elo Gap Explained
HappyHorse 1.0 sits at 1333 Elo on Artificial Analysis T2V, 60 points clear of Seedance 2.0 and further ahead of Veo 3.1. Here is what that gap actually looks like in clips, and where Veo still wins.