← LeaderboardThe Skill Leaderboard

pixverse

Generate, extend, compose and edit videos with PixVerse models via inference.sh CLI. Models: PixVerse V6, C1, Extend, Fusion, Modify, Lip Sync, Avatar, plus PixVerse Lipsync on fal.ai. Capabilities: text-to-video, image-to-video, 360p-1080p, 5-15s, video continuation, multi-subject composition from reference images, text-guided video editing with masks and reference images, lip sync from audio or TTS, talking avatars from a portrait. Use for: social clips, longer sequences built by extension, putting specific people, products or pets into one scene, swapping subjects in footage, dubbing, talki

Published by inference-sh ↗inference-sh/skills

Try your first task

A source-grounded starter task has not been verified for this skill yet.

SKILL.md

name: pixversedescription: "Generate, extend, compose and edit videos with PixVerse models via inference.sh CLI. Models: PixVerse V6, C1, Extend, Fusion, Modify, Lip Sync, Avatar, plus PixVerse Lipsync on fal.ai. Capabilities: text-to-video, image-to-video, 360p-1080p, 5-15s, video continuation, multi-subject composition from reference images, text-guided video editing with masks and reference images, lip sync from audio or TTS, talking avatars from a portrait. Use for: social clips, longer sequences built by extension, putting specific people, products or pets into one scene, swapping subjects in footage, dubbing, talking presenters. Triggers: pixverse, pixverse v6, pixverse c1, pixverse extend, pixverse fusion, pixverse modify, pixverse lipsync, pixverse lip sync, pixverse avatar, extend video, video continuation, compose subjects video, reference to video"allowed-tools: Bash(belt *)

Install the belt CLI skill: npx skills add belt-sh/cli

PixVerse Video Generation

Generate, extend, compose, edit and lip sync videos with PixVerse models via inference.sh CLI.

Quick Start

Requires inference.sh CLI (belt). Install instructions

belt login

belt app run pixverse/v6 --input '{"prompt": "a hot air balloon rising over misty valleys at sunrise, gentle drift, cinematic wide shot"}'

Models

ModelApp IDBest For
PixVerse V6pixverse/v6Versatile text-to-video and image-to-video, 360p-1080p, 5-15s
PixVerse C1pixverse/c1Advanced text-to-video and image-to-video, 360p-1080p, 5-15s
PixVerse Extendpixverse/extendContinue an existing video by 5s or 8s
PixVerse Fusionpixverse/fusionCompose 1-3 subject/background images into one video scene
PixVerse Modifypixverse/modifyEdit video with text: swap subjects, add elements, restyle regions
PixVerse Lip Syncpixverse/lip-syncSync mouth movement in a video to audio or TTS
PixVerse Avatarpixverse/avatarTalking avatar video from a portrait image and audio or TTS
PixVerse Lipsync (fal.ai)falai/pixverse-lipsyncLip sync with audio or TTS from 14 named voices

V6 and C1 share the same inputs and are billed per second. Prompts of 25-200 words are recommended. List all PixVerse apps with belt app search "pixverse".

Examples

Text-to-Video (V6)

belt app run pixverse/v6 --input '{
  "prompt": "a golden retriever puppy chases soap bubbles across a sunny backyard lawn, low camera following at ground level, shallow depth of field, warm afternoon light",
  "quality": "1080p",
  "duration": 8,
  "aspect_ratio": "16:9",
  "negative_prompt": "blurry, distorted faces"
}'

Image-to-Video (V6)

With an image, the video uses the image aspect ratio; aspect_ratio applies to text-to-video only.

belt app run pixverse/v6 --input '{
  "prompt": "the woman turns toward the camera and smiles as wind moves her hair",
  "image": "https://portrait.jpg",
  "quality": "720p",
  "duration": 5
}'

Fast Motion (C1)

belt app run pixverse/c1 --input '{
  "prompt": "a motorcycle races along a coastal highway, camera tracking alongside, sea spray, late afternoon sun",
  "motion_mode": "fast",
  "quality": "720p",
  "aspect_ratio": "9:16",
  "duration": 5
}'

Extend a Video

Source video: MP4/MOV, max 50MB, max 1920px, max 30s. duration is 5 or 8; 1080p supports 5s only. Fast motion mode is available for 5s, not at 1080p.

belt app run pixverse/extend --input '{
  "video": "https://my-clip.mp4",
  "prompt": "the puppy catches a bubble and tumbles onto the grass",
  "duration": 8,
  "quality": "720p"
}'

Fusion: Compose Subjects into One Scene

Give each image a unique ref_name and a type of subject or background. In the prompt, prefix the name with @ and follow it with a space.

belt app run pixverse/fusion --input '{
  "prompt": "@dog runs up to @girl on @beach and she throws a frisbee",
  "image_references": [
    {"image": "https://my-dog.jpg", "ref_name": "dog", "type": "subject"},
    {"image": "https://girl.jpg", "ref_name": "girl", "type": "subject"},
    {"image": "https://beach.jpg", "ref_name": "beach", "type": "background"}
  ],
  "model": "v6",
  "quality": "720p",
  "duration": 5,
  "aspect_ratio": "16:9"
}'

model is v6 (default) or c1.

Modify a Video

Reference images are addressed as @img0, @img1, @img2 (max 3). Mask selections are addressed as @selection0, @selection1. Output is 540p. Source video: max 100MB, max 1920px, max 30s.

belt app run pixverse/modify --input '{
  "video": "https://street-scene.mp4",
  "prompt": "replace the red car with the car in @img0",
  "reference_images": ["https://blue-sports-car.jpg"]
}'

Lip Sync with Audio

Source video: MP4/MOV/WebM, max 100MB, max 60s. Audio: MP3/WAV/M4A/AAC, max 60s. Provide either audio or tts_text.

belt app run pixverse/lip-sync --input '{
  "video": "https://presenter.mp4",
  "audio": "https://voiceover.mp3"
}'

Lip Sync with TTS

tts_text is max 200 chars and requires tts_speaker_id (auto picks a voice).

belt app run pixverse/lip-sync --input '{
  "video": "https://presenter.mp4",
  "tts_text": "Welcome back to the channel. Today we are testing three new cameras.",
  "tts_speaker_id": "auto"
}'

Talking Avatar

Portrait: PNG/JPG/WebP, max 20MB. Audio max 30s, or tts_text of 30-200 chars with tts_speaker_id.

belt app run pixverse/avatar --input '{
  "image": "https://headshot.jpg",
  "tts_text": "Hi, I am your onboarding guide. Let me show you around the dashboard.",
  "tts_speaker_id": "auto",
  "quality": "1080p",
  "prompt": "friendly expression, small hand gestures"
}'

Lip Sync on fal.ai

belt app run falai/pixverse-lipsync --input '{
  "video": "https://speaker.mp4",
  "text": "Thanks for watching, see you next week.",
  "voice_id": "Emily"
}'

voice_id options: Emily, James, Isabella, Liam, Chloe, Adrian, Harper, Ava, Sophia, Julia, Mason, Jack, Oliver, Ethan, Auto (default). Pass audio (WAV/MP3) instead of text to use your own recording.

Workflows

Generate, then Extend

# 1. Generate the opening clip
belt app run pixverse/v6 --input '{
  "prompt": "a lighthouse keeper climbs the spiral stairs carrying a lantern, night, storm outside",
  "quality": "720p",
  "duration": 8
}'

# 2. Continue it (repeat for longer sequences; the source must stay under 30s)
belt app run pixverse/extend --input '{
  "video": "<video-url-from-step-1>",
  "prompt": "he reaches the top and lights the great lamp, beam sweeps across the sea",
  "duration": 8,
  "quality": "720p"
}'

Fusion of Your Own Subjects

# Product + model + location in one shot
belt app run pixverse/fusion --input '{
  "prompt": "@model holds up @bottle and smiles inside @studio , slow push in",
  "image_references": [
    {"image": "https://model.jpg", "ref_name": "model", "type": "subject"},
    {"image": "https://perfume-bottle.png", "ref_name": "bottle", "type": "subject"},
    {"image": "https://studio.jpg", "ref_name": "studio", "type": "background"}
  ],
  "aspect_ratio": "9:16",
  "duration": 8
}'

Lip Sync a Generated Clip

# 1. Generate a talking-head shot
belt app run pixverse/v6 --input '{
  "prompt": "medium close-up of a news anchor at a desk looking into the camera, studio lighting, speaking",
  "quality": "1080p",
  "duration": 10
}'

# 2. Make the anchor say your script
belt app run pixverse/lip-sync --input '{
  "video": "<video-url-from-step-1>",
  "audio": "https://script-read.mp3"
}'

Parameters (V6 / C1)

ParameterTypeDefaultDescription
promptstringrequiredVideo description (25-200 words recommended)
imagefile-Image for image-to-video (PNG/JPG/WebP, max 10000px)
qualityenum720p360p, 540p, 720p, 1080p
durationinteger55-15 seconds, billed per second
aspect_ratioenum16:916:9, 9:16, 1:1, 4:3, 3:4 (text-to-video only)
negative_promptstring-What to avoid
motion_modeenumnormalnormal, fast
seedinteger-0-2147483647

Parameters (Extend)

ParameterTypeDefaultDescription
videofilerequiredMP4/MOV, max 50MB, max 1920px, max 30s
promptstringrequiredWhat happens next, max 2048 chars
qualityenum720p360p, 540p, 720p, 1080p
durationinteger55 or 8 (1080p: 5 only)
motion_modeenumnormalnormal, fast (5s only, not at 1080p)
negative_promptstring-What to avoid
seedinteger-Reproducible results

Parameters (Fusion)

ParameterTypeDefaultDescription
promptstringrequiredUse @ref_name (with trailing space) for each reference
image_referencesobject[]required1-3 {image, ref_name, type}; type is subject (default) or background
modelenumv6v6, c1
qualityenum720p360p, 540p, 720p, 1080p
durationinteger55-15 seconds
aspect_ratioenum16:916:9, 9:16, 1:1, 4:3, 3:4
seedinteger-Reproducible results

Pricing

Prices vary by resolution and duration. Check with belt app pricing <app-id> and estimate a specific input with belt app estimate <app-id> --input input.json.

Related Skills

# Full platform skill (all apps)
npx skills add inference-sh/skills@infsh-cli

# All video generation models
npx skills add inference-sh/skills@ai-video-generation

# Image-to-video workflows
npx skills add inference-sh/skills@image-to-video

# AI avatars & lipsync
npx skills add inference-sh/skills@ai-avatar-video

# Video prompting
npx skills add inference-sh/skills@video-prompting-guide

# Other video models
npx skills add inference-sh/skills@kling
npx skills add inference-sh/skills@runway
npx skills add inference-sh/skills@seedance
npx skills add inference-sh/skills@google-veo

Browse all video apps: belt app list --category video

Documentation