wan-2-2/lora/image-to-video
Transform stills into cinematic motion with open-source precision tools.
HappyHorse 1.0 is a next-generation AI video generation model that holds the #1 position on the Artificial Analysis Video Arena leaderboard for both text-to-video and image-to-video. Built on a unified Transformer architecture, HappyHorse 1.0 produces native 1080p video with advanced motion synthesis and multi-shot storytelling that maintains character consistency across scene transitions.
In blind user evaluations — where participants compare outputs from two models side by side without knowing which produced which — HappyHorse 1.0 outperformed every competing model including Seedance 2.0, Kling 3.0 Pro, SkyReels V4, and PixVerse V6.
| Category | Elo Rating | Rank |
|---|---|---|
| Text-to-Video (no audio) | 1333 | #1 |
| Image-to-Video (no audio) | 1392 | #1 |
| Text-to-Video (with audio) | 1205 | #2 |
| Image-to-Video (with audio) | 1161 | #2 |
The Artificial Analysis Video Arena uses an Elo system based on blind human preference. HappyHorse 1.0 leads the previous text-to-video leader (Seedance 2.0 at Elo 1273) by 60 points — a gap that corresponds to winning roughly 58–59% of head-to-head comparisons.
Native 1080p HD Resolution
Every video rendered by HappyHorse 1.0 outputs at true 1080p. The result includes rich color grading, accurate lighting, and film-grade detail — broadcast-ready without post-processing.
Advanced Motion Synthesis
The model generates remarkably fluid and natural motion. Subtle facial expressions, complex full-body movements, and multi-agent interactions all maintain physical plausibility. The motion synthesis engine demonstrates strong semantic understanding of camera direction, action choreography, and scene dynamics from text prompts.
Multi-Shot Narrative Consistency
One of the defining strengths of HappyHorse 1.0 is native multi-shot storytelling — the ability to generate cohesive video sequences where characters, wardrobe, visual style, and atmosphere remain consistent across scene transitions without manual editing.
Unified Text-to-Video and Image-to-Video
Both text-to-video and image-to-video run through one pipeline. Describe a scene with words, or transform a still image into dynamic footage — the same architecture handles both with intelligent motion planning.
Joint Audio-Video Synthesis
The model generates synchronized audio alongside video in a single pass: dialogue, ambient sounds, and Foley effects. The model ranks #2 in both with-audio categories on Artificial Analysis, trailing Seedance 2.0 by narrow margins of 14 and 1 Elo points.
Multilingual Support
Six languages are natively supported: Chinese, English, Japanese, Korean, German, and French. Prompts in any of these produce high-quality output with full linguistic nuance.
The model covers a wide range of creative and commercial use cases:
The best results come from writing prompts like a compact cinematic scene brief — not keyword lists. HappyHorse's workflow is built around subject, motion, camera, and atmosphere, so the prompt should describe what happens over time, how the camera experiences it, and what the sound should feel like.
Practical prompt format:
[duration / aspect ratio]. [main subject] in [setting]. [action beat 1], then [action beat 2].
[shot type / angle / camera move]. [lighting / atmosphere]. Audio: [sound / dialogue / ambience].
Keep [one hard constraint].
Example prompt:
> 5s, 16:9. A maintenance crew unfurls a huge graduation banner across a university rooftop. A sudden gust snaps the fabric sideways and nearly pulls one worker off balance before two coworkers grab it. Medium-wide tracking shot, slight handheld energy, bright afternoon light, realistic cloth physics. Audio: wind gusts, rooftop shouting, distant cheering. Keep faces readable and the motion physically believable.
Key principles:
More prompt examples:
| Use Case | Prompt |
|---|---|
| Cinematic product close-up | A black ceramic coffee mug sits on a rain-wet wooden table. Steam rises slowly from the rim. Camera begins with a tight close-up on the surface texture, then pulls back to reveal a gray morning window behind. Overcast natural light. No music. Ambient rain sound. |
| Character motion outdoors | A young woman in a yellow raincoat walks across a stone bridge over a fast-moving river. Camera tracks alongside her at shoulder height. Autumn leaves fall from both sides of the frame. Wind sound and footstep audio. 16:9, cinematic color grade. |
| Abstract social content | Ink drops fall into still water in extreme close-up. Each drop creates expanding circular ripples in slow motion. Black ink on white water, high contrast. No audio. 9:16 portrait format for vertical feed. |
| Product animation (I2V) | Upload: product photo of a glass perfume bottle. The bottle sits on a white marble surface. A soft light sweeps across it from left to right, catching the glass facets. Subtle lens flare on the highlight. Camera stays locked. Ambient room tone only. |
HappyHorse 1.0 and Seedance 2.0 (by ByteDance) are the two highest-ranked AI video models on the Artificial Analysis Video Arena. They excel in different areas.
Benchmark comparison (April 2026):
| Category | HappyHorse 1.0 | Seedance 2.0 |
|---|---|---|
| T2V Elo (no audio) | 1333 — #1 | 1273 — #2 |
| I2V Elo (no audio) | 1392 — #1 | 1355 — #2 |
| T2V Elo (with audio) | 1205 — #2 | 1219 — #1 |
| I2V Elo (with audio) | 1161 — #2 | 1162 — #1 |
| Architecture | Single 40-layer Transformer | Multimodal diffusion transformer |
| Native audio languages | 6 | Primarily Chinese and English |
| Open source | Claimed, not yet accessible | No |
| Available on RunComfy | Coming soon | ✓ |
Elo scores sourced from Artificial Analysis Video Arena, early April 2026. Scores change as votes accumulate.
Where HappyHorse 1.0 leads:
Where Seedance 2.0 leads:
Prompting philosophy difference:
Bottom line: HappyHorse 1.0 is the more exciting model for cinematic, prompt-led exploration. Seedance 2.0 is the more mature model for controlled, multimodal, edit-heavy workflows.
Transform stills into cinematic motion with open-source precision tools.
Create lifelike talking visuals with AI that matches voice and motion seamlessly.
Unified AI model for refined scene editing, style match, and smooth video refits
Generate cinematic clips from stills with sound, morph control, and stylistic flexibility.
Consistent characters, objects, and scenes in any setting or angle.
Convert photos into expressive talking avatars with precise motion and HD detail
HappyHorse 1.0 is a next-generation AI video model ranked #1 on the Artificial Analysis Video Arena for both text-to-video (Elo 1333) and image-to-video (Elo 1392). It generates native 1080p video with advanced motion synthesis, multi-shot character consistency, and multilingual support across six languages.
The Artificial Analysis Video Arena ranks models through blind user voting — participants compare two videos generated from the same prompt without knowing which model made which, then pick the better result. Votes feed into an Elo rating system. It holds the highest Elo in both text-to-video and image-to-video (no audio) categories as of April 2026.
The model outputs native 1080p HD resolution. Video includes rich color grading, accurate lighting, and film-grade detail suitable for broadcast and professional production without additional post-processing.
Yes. The model generates synchronized audio alongside video in one pass — including dialogue, ambient sounds, and Foley effects. It ranks #2 in the with-audio categories on the Artificial Analysis leaderboard.
Six languages are natively supported: Chinese, English, Japanese, Korean, German, and French. Prompts in any supported language produce high-quality video with full linguistic nuance.
Multi-shot storytelling allows the model to generate video sequences with multiple shots while maintaining consistency in characters, wardrobe, visual style, and atmosphere across scene transitions — eliminating the need for manual editing between clips.
Yes. The model supports both text-to-video and image-to-video through a unified pipeline. Upload a static image to animate it with intelligent motion synthesis, or describe a scene entirely through text.
RunComfy is the premier ComfyUI platform, offering ComfyUI online environment and services, along with ComfyUI workflows featuring stunning visuals. RunComfy also provides AI Models, enabling artists to harness the latest AI tools to create incredible art.
