HappyHorse 1.0 is Now Available in ComfyUI
Cinematic video generation and flexible edit workflows
We’re excited to share that ComfyUI now supports HappyHorse 1.0 — a cinematic video generation model from Alibaba, designed for high-quality storytelling video and production-ready creative workflows.
HappyHorse 1.0 focuses on strong aesthetics, multi-shot sequencing, and powerful edit-style workflows (generation + editing). It’s a great fit for creators building ads, e-commerce visuals, short-form content, and social marketing videos.
Highlights
Multi-modal creation paths: Text-to-Video (T2V), Image-to-Video (I2V), and Subject-to-Video (S2V).
Cinematic aesthetics: Wide-aperture framing, shallow depth of field, refined texture, and atmospheric mood.
Multi-shot output: Up to 15s at 1080p, with an emphasis on multi-shot consistency across cut transitions.
Editing workflows: V2V and SV2V for transforming footage or replacing/inserting subjects while preserving motion and composition.
Output Examples
Get Started in ComfyUI
Update to the latest ComfyUI or visit Comfy Cloud
Open Template Library → Search for HappyHorse
Update prompt or input image, then hit Run
Available workflows
Image to Video
Text to Video
Video Edit
Reference to Video
As always, enjoy creating!
Thanks for the honest feedback — I'm Ethan, one of the people behind HappyHorse. Both points are fair. Temporal consistency beyond 5 seconds is genuinely the hardest unsolved problem in current video generation, and our model reflects that limitation. The character drift is real and we're actively working on it. For short cinematic clips and social content (15s and under), it performs well — which is why the workflow is optimized for that. Anyone wanting to try the latest version, it's free at happyhorse-ai.video — no watermark needed for basic use. Appreciate the ComfyUI team covering it!
Good up to 5 seconds, after that it turns into a cartoon. The comparisons I see on various sites with Seedance are unfortunate. It was showing from the specs. I hope if it gets released as an open model, various LoRAs will help improve it. For now, 5 seconds per shot is not enough.