Video Generation

Generate AI videos from a text prompt or reference materials using the listenhub video CLI. Animate a still image, edit an existing clip, or drive a character's lips with audio or text-to-speech. Three model families cover different jobs: HappyHorse, SeeDance, and PixVerse.

Trigger

Invoke this skill with /video-gen, or use any of these phrases:

Requires ListenHub Skills to be installed — see Getting Started.

For narrated explainer videos with AI visuals, use /explainer instead.

Quick Example

Generate a video: cyberpunk city at night, 16:9, 5 seconds

The AI walks you through the mode and parameters one question at a time, shows a cost estimate, and asks you to confirm before generating. Generation takes minutes — the job runs in the background and the AI notifies you when the video is ready, with a URL, duration, resolution, and credit cost.

Models

Pick the model by the job. HappyHorse is the default and the only family that edits existing video; SeeDance adds last-frame and reference-audio support; PixVerse is the only family with lip sync, plus a set of atomic capabilities (mimic, restyle, fusion, transition, marketing agent).

Lip sync, mimic, restyle, fusion, transition, and the marketing agent are PixVerse-only. HappyHorse and SeeDance do not support them.

SeeDance model variants

When you pick SeeDance, choose between two variants:

Modes

The AI routes to a mode based on what reference material you have. HappyHorse and SeeDance share the listenhub video create command; PixVerse uses listenhub openapi video pixverse generate with an explicit --capability.

listenhub video create \
  --prompt "cyberpunk city at night, neon reflections on wet streets" \
  --model "happyhorse" \
  --resolution "1080p" \
  --ratio "16:9" \
  --duration 5 \
  --no-wait --json

listenhub video create \
  --prompt "bring the scene to life with smooth motion" \
  --model "happyhorse" \
  --resolution "1080p" \
  --duration 5 \
  --first-frame "/path/to/scene.png" \
  --no-wait --json

listenhub video create \
  --prompt "[Image 1]'s character walking through [Image 2]'s street" \
  --model "happyhorse" \
  --resolution "1080p" \
  --ratio "16:9" \
  --duration 5 \
  --reference-image "/path/to/character.png" \
  --reference-image "/path/to/scene.png" \
  --no-wait --json

listenhub openapi video pixverse generate \
  --capability lip_sync \
  --source-video-id "abc123" \
  --audio "https://example.com/voice.mp3" \
  --quality 720p \
  --no-wait --json

listenhub openapi video pixverse generate \
  --capability lip_sync \
  --source-task-id "task_xyz" \
  --pixverse-json '{"tts":{"speakerId":"speaker_01","content":"Welcome to this episode"}}' \
  --quality 720p \
  --no-wait --json

Video edit (HappyHorse)

Edit an existing clip — change style, replace the background, restyle motion. HappyHorse only; if you ask for this on SeeDance, the AI switches you to HappyHorse.

Video requirements: mp4/mov (H.264 recommended); 3–60s input (output capped at 15s); ≤ 100 MB; short edge ≥ 360px, long edge ≤ 4096px. Optionally pass 0–5 reference images.

listenhub video create \
  --prompt "replace the background with a deep starry sky, keep the subject's motion" \
  --model "happyhorse" \
  --resolution "1080p" \
  --reference-video "/path/to/input.mp4" \
  --audio-setting "origin" \
  --no-wait --json

--audio-setting controls audio: auto lets the model decide, origin keeps the original audio. Video edit has no --ratio or --duration — the output matches the input video.

Other PixVerse capabilities

PixVerse exposes additional atomic capabilities through --capability, all OpenAPI-only with URL inputs:

Parameters

The AI asks for these one at a time and applies sensible session defaults. For HappyHorse and SeeDance, ratio and duration use --ratio / --duration; PixVerse uses --quality and --aspect-ratio instead of --resolution / --ratio.

PixVerse-specific flags: --capability, --quality (360p/540p/720p/1080p), --aspect-ratio (9:16/16:9/1:1/4:3/3:4), --source-video-id / --source-task-id, --audio, --agent-type, --restyle-id, and --pixverse-json for nested payloads (tts, imageReferences, multiTransition).

Estimating credits

Before generating, the AI runs an estimate. To check cost yourself, mirror the create parameters against the estimate command:

# HappyHorse / SeeDance
listenhub video estimate --model "happyhorse" --resolution "1080p" --ratio "16:9" --duration 5 --json

# PixVerse — mirror the capability + quality + duration
listenhub openapi video pixverse estimate --capability text_to_video --model pixverse --quality 720p --duration 5 --json

For video edit, add --has-video-input and --input-video-duration <seconds>.

Output

Status flows pending → generating → uploading → success. On success the AI reports the video URL, duration, resolution, ratio, seed, and credits charged.

Output behavior follows the outputMode set during config:

• inline (default) or both — the video URL and metadata are shown directly in the conversation.
• download or both — the file is also saved to the current working directory with a topic-based name (e.g. cyberpunk-city.mp4). Names are de-duplicated automatically.

To review past work, listenhub video get <taskId> --json returns a single task and listenhub video list --json lists recent tasks. Global flags apply: --json / -j, --no-wait, and --timeout <s>.

API Reference

See the AI Video API reference for endpoint paths, request parameters, and response fields.