HappyHorse 1.0 ComfyUI Nodes

ComfyUI custom nodes for HappyHorse 1.0 — Alibaba's #1 ranked AI video generation model (1392 Elo I2V, 1333 Elo T2V on Artificial Analysis). Generate 1080p (or cheaper 720p) videos directly inside ComfyUI using the muapi.ai API. Prefer raw Python? See the companion HappyHorse 1.0 API wrapper.

License: MIT ComfyUI HappyHorse 1.0

• Happy Horse text-to-video playground — Try the 1080p model this node pack wraps.
• MuAPI video-generation docs — Shared API and polling patterns for video models.

HappyHorse 1.0 is Alibaba's state-of-the-art AI video generation model, built by the Future Life Lab team at Taotian Group. It debuted on April 7, 2026, instantly claiming the top spot in both Text-to-Video and Image-to-Video categories on the Artificial Analysis leaderboard.

• #1 Ranked — 1333 Elo T2V, 1392 Elo I2V
• Native 1080p HD — full HD output without upscaling, powered by a 15B-parameter 40-layer Transformer
• 720p option — same model, ~half the cost, available on every node via the resolution selector
• Multi-image reference-to-video — condition generation on 1–9 reference images for consistent characters, styles and environments
• Natural-language video editing — restyle, replace skies, swap subjects on an existing clip; keep or regenerate the audio track
• Integrated audio-video — jointly generates video and synchronized audio in a single Transformer forward pass
• Fast — ~10 seconds typical generation time

1. Open ComfyUI Manager → Install via Git URL
2. Paste: https://github.com/Anil-matcha/happyhorse-comfyui
3. Restart ComfyUI

cd ComfyUI/custom_nodes
git clone https://github.com/Anil-matcha/happyhorse-comfyui
pip install -r happyhorse-comfyui/requirements.txt

1. Sign up at muapi.ai and grab a key from Dashboard → API Keys → Create Key .
2. Right-click the ComfyUI canvas → Add Node → 🐎 HappyHorse 1.0.
3. Add a 🔑 HappyHorse 1.0 API Key node, paste your key, and wire its output into any generation node.
4. Write a prompt and hit Queue Prompt.

Tip: If you use the MuAPI CLI, run muapi auth configure --api-key YOUR_KEY once and every node will pick the key up automatically — no need to paste it anywhere.

Set your muapi.ai API key once and wire the output to all generation nodes. Alternatively, leave every api_key field blank — nodes auto-read from ~/.muapi/config.json if you've authenticated via the CLI.

Generate a HappyHorse 1.0 video from a text prompt. HappyHorse 1.0 jointly generates synchronized audio with the video, so feel free to include sound cues (e.g. "rain pattering on leaves", "crowd cheering", "piano melody drifting") right inside your prompt.

Outputs: video_url · first_frame (IMAGE) · request_id

Endpoint: POST /api/v1/happy-horse-1-text-to-video-{1080p|720p} (selected from the resolution widget)

Animate a single start-frame image into a HappyHorse 1.0 video. The model animates outward from the supplied image; the prompt is optional and guides the motion.

Provide either image or image_url (not both). Audio is generated jointly with the video — include sound cues in the prompt for richer output.

Outputs: video_url · first_frame (IMAGE) · request_id

Endpoint: POST /api/v1/happy-horse-1-image-to-video-{1080p|720p} (selected from the resolution widget)

Generate a HappyHorse 1.0 video from a text prompt + 1–9 reference images (style, subject, environment). Different from Image-to-Video, which uses a single image as the start frame: every reference here is treated as a look/character/scene reference the model should internalize.

You can mix and match: any combination of image_urls + image_1/image_2/image_3/image_4 is concatenated. The combined list must be 1–9 URLs. Image specs: JPEG / PNG / WEBP, ≥400 px shortest side, ≤10 MB each.

Outputs: video_url · first_frame (IMAGE) · request_id

Endpoint: POST /api/v1/happy-horse-1-reference-to-video-{1080p|720p} (selected from the resolution widget)

Edit an existing video using a natural-language instruction. Optionally supply 0–5 reference images to anchor characters, styles, or elements that should appear in the edited output. Audio can be regenerated to match the edit (audio_setting=auto) or kept from the source (audio_setting=origin).

Source video specs: MP4 or MOV (H.264 recommended), 3 – 60 s, ≤100 MB, longer side ≤2160 px, shorter side ≥320 px, frame rate >8 fps. Reference images: JPEG / PNG / WEBP, ≥300 px each side, ≤10 MB each.

Outputs: video_url · first_frame (IMAGE) · request_id

Endpoint: POST /api/v1/happy-horse-1-video-edit-{1080p|720p} (selected from the resolution widget)

Downloads the generated video to ComfyUI's output folder and returns all frames as an IMAGE tensor for use with other nodes (preview, VHS, upscale, etc.).

Outputs: frames (IMAGE) · filepath (STRING) · frame_count (INT)

Load any .json file from this repo via File → Load in ComfyUI.

Text-to-Video:

[🔑 API Key] ──────────────────────────────────────┐
 ↓
[🐎 Text-to-Video] → video_url → [🐎 Save Video] → frames → [Preview Image]

Image-to-Video:

[🔑 API Key] ──────────────────────────────────────────────────────────────┐
 ↓
[LoadImage] → [🐎 Image-to-Video] → video_url → [🐎 Save Video] → frames → [Preview Image]

Reference-to-Video:

[🔑 API Key] ──────────────────────────────────────────────────────────────┐
 ↓
[LoadImage ×ばつ N] → [🐎 Reference-to-Video] → video_url → [🐎 Save Video] → frames → [Preview Image]

Video Edit:

[🔑 API Key] ──────────────────────────────────────────────┐
 ↓
(video_url string) → [🐎 Video Edit] → video_url → [🐎 Save Video] → frames → [Preview Image]

This node pack uses the muapi.ai API under the hood:

• T2V 1080p: POST https://api.muapi.ai/api/v1/happy-horse-1-text-to-video-1080p
• T2V 720p: POST https://api.muapi.ai/api/v1/happy-horse-1-text-to-video-720p (~half the 1080p cost)
• I2V 1080p: POST https://api.muapi.ai/api/v1/happy-horse-1-image-to-video-1080p
• I2V 720p: POST https://api.muapi.ai/api/v1/happy-horse-1-image-to-video-720p (~half the 1080p cost)
• Ref2V 1080p: POST https://api.muapi.ai/api/v1/happy-horse-1-reference-to-video-1080p
• Ref2V 720p: POST https://api.muapi.ai/api/v1/happy-horse-1-reference-to-video-720p (~half the 1080p cost)
• Video Edit 1080p: POST https://api.muapi.ai/api/v1/happy-horse-1-video-edit-1080p
• Video Edit 720p: POST https://api.muapi.ai/api/v1/happy-horse-1-video-edit-720p (~half the 1080p cost)
• Poll: GET https://api.muapi.ai/api/v1/predictions/{request_id}/result
• Upload: POST https://api.muapi.ai/api/v1/upload_file

Authentication is a single x-api-key header — no session tokens required.

The submit-then-poll flow is identical for both endpoints:

POST /api/v1/happy-horse-1-... → { "request_id": "abc123" }
GET /api/v1/predictions/abc123/result
 → { "status": "processing" } (keep polling)
 → { "status": "completed",
 "outputs": ["https://.../video.mp4"] }

Status values: queued, pending, processing, completed, failed.

• Python ≥ 3.8
• requests ≥ 2.28 · Pillow ≥ 9.0 · numpy ≥ 1.23 · torch ≥ 2.0 · opencv-python ≥ 4.7

This repo is focused on HappyHorse 1.0 only. If you need access to 100+ models — Kling, Veo3, Flux, HiDream, GPT-image-1.5, Imagen4, Wan, lipsync, audio, image enhancement and more — check out the full MuAPI ComfyUI node pack:

SamurAIGPT/muapi-comfyui — ComfyUI nodes for every muapi.ai model in one place.

MIT © 2026