Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.magnific.com/llms.txt

Use this file to discover all available pages before exploring further.

Alibaba ATH Happy Horse 1.0 integration

Happy Horse 1.0 Video Edit modifies an existing video using natural-language instructions, optionally guided by up to 5 reference images. Output is capped at 15 seconds.
Happy Horse 1.0 Video Edit is an AI video editing API that transforms an existing video with natural-language instructions. Built by Alibaba ATH (Taotian Innovation Unit, led by Zhang Di), the model edits the input MP4/MOV video based on a text prompt and up to 5 optional reference images. Output is delivered at 720P (1280x720) or 1080P (1920x1080) and capped at 15 seconds. Audio handling is configurable: auto lets the model decide whether to regenerate audio based on the prompt; origin preserves the input video’s original audio.

​
Key capabilities

  • Natural-language editing: Describe the desired edit in a prompt up to 2500 characters
  • Up to 5 reference images: Guide the edit with reference images via @Image1..@Image5 tokens
  • Audio control: auto (model decides) or origin (preserve input audio) via audio_setting
  • Resolution options: 720P (1280x720) and 1080P (1920x1080) output
  • Output cap: Edited video is at most 15 seconds long, regardless of input length
  • Wide input support: MP4 or MOV with H.264 codec, 3 to 60 seconds, max 100 MB
  • Reproducible results: Fixed seed support (0 to 2147483647) for consistent generation
  • Async processing: Webhook notifications or polling for task completion

​
How reference images work

Provide up to 5 reference images in the image_urls array. Reference each one in the prompt using @Image1..@Image5 tokens (1-based, in array order): Example prompt:
“Replace the background with the cityscape in @Image1 and change the protagonist’s outfit to match @Image2”

​
Use cases

  • Background replacement: Swap backgrounds for product, fashion, or marketing videos
  • Style transfer: Change the visual style or color grading of a clip
  • Object replacement: Replace specific elements while preserving motion and composition
  • Costume/wardrobe edits: Modify outfits or props using reference images
  • Visual effects: Add or remove elements with natural-language instructions
  • Quick iterations: Rapidly produce variants of a video without re-shooting

​
API operations

Edit videos by submitting a video URL, prompt, and optional reference images. The service returns a task ID for async polling or webhook notification.

POST /v1/ai/video-edit/happy-horse-1

Create a new video edit task

GET /v1/ai/video-edit/happy-horse-1

List all Happy Horse 1.0 Video Edit tasks

GET /v1/ai/video-edit/happy-horse-1/{task-id}

Get task status and results by ID

​
Parameters

ParameterTypeRequiredDefaultDescription
video_urlstringYes-URL of the source video (MP4/MOV with H.264, 3-60s, longer side max 2160 px, shorter side min 320 px, aspect ratio 1:2.5 to 2.5:1, max 100 MB)
promptstringYes-Editing instructions. Reference images via @Image1..@Image5. Max 2500 characters
image_urlsarrayNo-Up to 5 reference images (JPEG/JPG/PNG/WEBP, min 300 px, max 10 MB each)
resolutionstringNo"1080P"Output resolution: "720P" or "1080P"
audio_settingstringNo"auto"Audio handling: "auto" (model decides) or "origin" (preserve input audio)
seedintegerNoRandomSeed for reproducibility (0 to 2147483647)
webhook_urlstringNo-URL for async status notifications

​
Frequently Asked Questions

Happy Horse 1.0 Video Edit is an AI model from Alibaba ATH that edits an existing video based on natural-language instructions. You provide a video URL, a prompt describing the desired changes, and optionally up to 5 reference images. The model returns a new video at 720P or 1080P, capped at 15 seconds.
Happy Horse 1.0 Video Edit accepts MP4 and MOV files with H.264 codec via publicly accessible URLs. Input duration must be between 3 and 60 seconds, frame rate greater than 8 fps, longer side at most 2160 px, shorter side at least 320 px, aspect ratio between 1:2.5 and 2.5:1, and max file size 100 MB.
The model produces edits up to 15 seconds long regardless of input duration. If your input video is longer than 15 seconds, the output is truncated. Plan accordingly when selecting input clips.
Use position-based tokens that match the order of the image_urls array: the 1st image is @Image1, the 2nd is @Image2, and so on up to @Image5. Example: “Replace the background with the scene in @Image1 and add the object from @Image2 in the foreground.”
audio_setting controls how the original audio track is handled. "auto" lets the model decide based on the prompt — if the prompt involves audio descriptions, the model may regenerate audio; otherwise the original audio is preserved. "origin" forces the original audio track to be preserved without regeneration.
Rate limits depend on your subscription tier. See the Rate Limits page for current limits by plan.
See the Pricing page for current rates and subscription options.

​
Best practices

  • Input video selection: Use clips at or below 15 seconds to avoid truncation. Higher-quality inputs produce better edits.
  • Prompt clarity: Be explicit about which elements to change and which to keep (“Replace the background but keep the subject’s pose and clothing”).
  • Reference images: Use clean reference images with the target element clearly visible. Match aspect ratio and lighting where possible.
  • Audio choice: Use "origin" when the input has dialogue or music you want preserved. Use "auto" when the prompt involves audio changes.
  • Iterate with seed: Save the seed from successful edits to recreate similar results.
  • Production integration: Use webhooks for scalable applications instead of polling.
  • Error handling: Implement retry with exponential backoff for 503 errors during high-demand periods.