alibaba / happyhorse 1.1

## Overview HappyHorse 1.1 is an Alibaba video generation model for short, high-fidelity clips. It turns a written scene description into an MP4 video with synchronized audio. You can also animate a still image by using it as the first frame. It helps you prototype ads, story beats, and camera motion without manual video editing. ## What you can build - Text-to-video concepts for ads, trailers, and social posts - Vertical short-form clips for 9:16 platforms - Talking-head style scenes with prompt-driven dialogue and lip-sync - Product shots that start from a real photo and add motion - Quick storyboards with shot-by-shot timing inside one prompt - Style studies for lighting, lenses, and camera moves ## Inputs - A first-frame image (optional) to switch from text-to-video to image-to-video. Use JPEG, JPG, PNG, or WEBP. Keep both dimensions at least 300 px. Stay under 20 MB. - A video description (required for text-to-video, optional for image-to-video). Describe subjects, actions, camera moves, and style. Keep it within 2,500 characters. - A clip duration selection in seconds. Common options are 5, 10, or 15 seconds. - An output resolution choice. Pick 720p or 1080p. - An aspect ratio choice for text-to-video. Options include 16:9, 9:16, 1:1, 4:3, 3:4, 4:5, 5:4, 9:21, and 21:9. - A watermark toggle (optional). When enabled, it places a “HappyHorse” mark in the lower-right corner. - A numeric seed (optional) to repeat runs more closely. Use an integer from 0 to 2,147,483,647. ## Outputs - A downloadable MP4 video file. - Video metadata alongside the file, including width, height, frame rate, duration, frame count, and file size. - The seed used for the generation, so you can retry the same setup. ## Limitations - Clip length is short-form. Don’t expect long videos in one generation. - Image-to-video inherits framing from your first frame. It won’t reshape to a different ratio. - Very wide or very tall images can fail if the aspect ratio is outside 1:2.5 to 2.5:1. - The same seed can still produce small differences across runs. - Low-quality inputs reduce stability. Blurry images, heavy compression, and tiny subjects can cause artifacts. - Fine text rendering is unreliable. Logos and subtitles may warp or misspell. ## Safety & compliance - The service applies content moderation to prompts and generated outputs. - Don’t generate disallowed sexual content, graphic violence, or instructions for wrongdoing. - Don’t impersonate real people without rights or consent. Avoid deceptive political or medical content. - You’re responsible for having rights to any images you upload and for how you use outputs.