PixVerse Skills

August 19, 2026 · View on GitHub

Agent skill library for PixVerse CLI — helps AI agents (Claude Code, Cursor, Codex, etc.) generate videos, images, audio (speech & music), and MiniApp projects through structured, composable workflows.

What is this?

PixVerse CLI is a UI-free version of pixverse.ai — all models, parameters, and capabilities from the website are available as CLI commands with structured JSON output.

This repository provides skill files that teach AI agents how to use those commands correctly: which flags to pass, which models support which parameters, how to chain commands into pipelines, and how to handle errors.

Quick Start

# Install the CLI
npm install -g pixverse

# Authenticate
pixverse auth login

# Create a video
pixverse create video --prompt "A cat astronaut floating in space" --json

PixVerse CLI uses the same credit system as the website. Only subscribed users can use it. See subscription plans.

Skill Structure

skills/
  SKILL.md                          # Entry point — start here
  capabilities/                     # Individual command skills
    auth-and-account.md             #   Authentication & account management
    create-video.md                 #   Text-to-video, image-to-video, fusion
    create-and-edit-image.md        #   Text-to-image, image-to-image
    create-voice.md                 #   Text-to-speech (TTS) — MiniMax, ElevenLabs voices
    create-music.md                 #   Prompt-to-music — MiniMax, ElevenLabs, Google Lyria
    modify-video.md                 #   AI content editing (replace subjects, swap outfits, change backgrounds)
    motion-control.md               #   Character animation with motion reference video
    transition.md                   #   Keyframe transition animations
    post-process-video.md           #   Extend duration, upscale resolution
    prompting-guide.md              #   Model-agnostic prompt advice (advice only — never auto-edits)
    prompt-enhance.md               #   Prompt optimization for V6 video generation
    seedance-prompt-optimize.md     #   Prompt optimization for Seedance 2.0 / 2.5
    seedance-vibe-creating.md       #   Vibe Creating — distill emotional / atmospheric ideas into Seedance prompts
    task-management.md              #   Poll and wait for generation tasks
    asset-management.md             #   List, download, upload, delete assets
    saved-folders.md                #   Organize assets into named folders
    template.md                     #   Browse and create from effect templates
    miniapps.md                     #   Discover and run PixVerse MiniApps
    workspace.md                    #   Team workspace management
    mondo-poster-design.md          #   Mondo-style poster, book cover, album art design
    character-design.md             #   Persistent characters — three-view sheet + cloud asset id reuse
    item-design.md                  #   Persistent items / props — four-panel orthographic sheet + cloud asset id reuse
  workflows/                        # Multi-step pipeline skills
    text-to-video-pipeline.md       #   End-to-end text-to-video
    image-to-video-pipeline.md      #   Animate an image into video
    text-to-image-to-video.md       #   Generate image then animate it
    image-editing-pipeline.md       #   Iterative image editing
    modify-video-pipeline.md        #   Modify video content then enhance
    motion-control-pipeline.md      #   Character animation end-to-end
    video-production.md             #   Full production (create + extend + audio + upscale)
    storyboard-to-video.md          #   Multi-shot storyboard → concatenated video
    batch-creation.md               #   Parallel batch generation
    mondo-poster-pipeline.md        #   End-to-end Mondo poster generation
    mondo-poster-to-video-pipeline.md #  Animate poster into cinematic video
  references/                       # Curated design knowledge
    mondo-poster/                   #   37 artist styles, composition, genre templates

Capabilities vs Workflows

Capabilities document a single command or command group — flags, models, parameter constraints, JSON output format, error codes.
Workflows compose multiple capabilities into end-to-end pipelines with step-by-step instructions.

MiniApps

PixVerse CLI v1.3.0 adds a top-level miniapps group for preset generators. Discover the live catalog and schema before submitting a project:

pixverse miniapps list --json
pixverse miniapps info magic_extend --json
pixverse miniapps create \
  --id magic_extend \
  --params '{"image":"<media-path>","ratio":"16:9"}' \
  --no-wait \
  --json

Creation returns a project_id. Use it with task and asset, always passing --type miniapps:

pixverse task wait <project_id> --type miniapps --json
pixverse asset download <project_id> --type miniapps --json

See skills/capabilities/miniapps.md for normalized params_schema fields, media-path handling, output contracts, and the full project lifecycle.

Supported Models

Video Models

Model	CLI value	Modes	Quality	Duration	Aspect Ratio
PixVerse V6	`v6` (default)	Video, Transition (first/last frame), Extend, Reference	`360p` `540p` `720p` `1080p`	`1`-`15` (any integer; Reference video uses `auto`)	`auto` `16:9` `4:3` `1:1` `3:4` `9:16` `3:2` `2:3` `21:9` (Reference supports `auto`)
PixVerse C1	`pixverse-c1`	Video, Transition (first/last frame), Reference	`360p` `540p` `720p` `1080p`	`1`-`15` (any integer)	`16:9` `4:3` `1:1` `3:4` `9:16` `3:2` `2:3`
PixVerse v5.6	`v5.6`	Video, Transition, Reference, Motion Control	`360p` `480p` `540p` `720p` `1080p`	`1`-`10` (any integer)	`16:9` `4:3` `1:1` `3:4` `9:16` `3:2` `2:3`
Sora 2	`sora-2`	Video	`720p`	`4` `8` `12`	`16:9` `9:16`
Sora 2 Pro	`sora-2-pro`	Video	`720p` `1080p`	`4` `8` `12`	`16:9` `9:16`
Veo 3.1 Standard	`veo-3.1-standard`	Video, Transition	`720p` `1080p` `2160p`	`4` `6` `8`	`16:9` `9:16`
Veo 3.1 Fast	`veo-3.1-fast`	Video, Transition	`720p` `1080p` `2160p`	`4` `6` `8`	`16:9` `9:16`
Veo 3.1 Lite	`veo-3.1-lite`	Video, Transition	`720p` `1080p`	`4` `6` `8`	`16:9` `9:16`
Grok Imagine	`grok-imagine`	Video, Extend, Reference	`480p` `720p`	`1`-`15` (Reference video uses `auto`)	fixed ratios normally; Reference video derives framing from source
Grok Imagine 1.5	`grok-imagine-1.5`	Video (image-to-video only)	`480p` `720p` `1080p`	`1`-`15`	derived from input image
Happy Horse 1.0	`happyhorse-1.0`	Video	`720p` `1080p`	`3`-`15`	`16:9` `9:16` `1:1` `4:3` `3:4`
Seedance 2.5	`seedance-2.5`	Video, Reference, Transition (exactly 2 frames)	`480p` `720p` `1080p`	`4`-`30` (Reference video also `auto`)	`auto` `21:9` `16:9` `4:3` `1:1` `3:4` `9:16` (mode-dependent)
Seedance 2.0 Standard	`seedance-2.0-standard`	Video, Reference, Transition	`480p` `720p` `1080p` `2160p`	`4`-`15`	`16:9` `4:3` `1:1` `3:4` `9:16` `21:9`
Seedance 2.0 Fast	`seedance-2.0-fast`	Video, Reference, Transition	`480p` `720p`	`4`-`15`	`16:9` `4:3` `1:1` `3:4` `9:16` `21:9`
Seedance 2.0 Mini	`seedance-2.0-mini`	Video, Reference, Transition	`480p` `720p`	`4`-`15`	`16:9` `4:3` `1:1` `3:4` `9:16` `21:9`
MiniMax H3	`minimax-h3`	Video, Reference, Transition (exactly 2 frames)	`768p` `1440p`	`5`-`15`	`auto` `21:9` `16:9` `4:3` `1:1` `3:4` `9:16` (mode-dependent)
Kling O3 Pro	`kling-o3-pro`	Video, Reference, Transition	model-selected (`--quality` omitted)	`3`-`15`	`16:9` `9:16` `1:1`
Kling O3 Standard	`kling-o3-standard`	Video, Reference, Transition	model-selected (`--quality` omitted)	`3`-`15`	`16:9` `9:16` `1:1`
Kling O3 4K	`kling-o3-4k`	Video, Reference, Transition	model-selected 4K tier (`--quality` omitted)	`3`-`15`	`16:9` `9:16` `1:1`
Kling 3.0 Pro	`kling-3.0-pro`	Video, Transition	model-selected (`--quality` omitted)	`3`-`15`	`16:9` `9:16` `1:1`
Kling 3.0 Standard	`kling-3.0-standard`	Video, Transition	model-selected (`--quality` omitted)	`3`-`15`	`16:9` `9:16` `1:1`
Kling 3.0 4K	`kling-3.0-4k`	Video, Transition	model-selected 4K tier (`--quality` omitted)	`3`-`15`	`16:9` `9:16` `1:1`
Google Gemini Omni	`gemini-omni-flash`	Video, Reference	`720p`	`3`-`10`	`16:9` `9:16`

MiniMax H3 defaults to 1440p. Text-to-video defaults to 16:9 and rejects auto; image-to-video forces auto. Reference requests with images default to auto but preserve an explicit fixed ratio; reference requests without images default to 16:9 and reject auto.

Seedance 2.5 supports 480p, 720p, and 1080p and defaults to 720p, 5 seconds, and 16:9 without a reference video. Text-to-video and Reference accept --aspect-ratio auto; Reference with a video defaults to --duration auto, which locks framing to auto, while an explicit 4–30s duration allows either automatic or fixed framing. Reference also accepts --task-type auto|reference|edit|extend (auto by default). It accepts up to 50 mixed references (30 images, 10 videos, and 10 audios), with separate 30-second aggregate limits for video and audio. Transition requires exactly two images and a prompt and has no selectable aspect ratio. Generated audio, multi-shot, and off-peak generation are unsupported.

Kling resolution is selected by the model ID. All Kling video requests omit quality; an explicit --quality value is ignored with a warning. The kling-o3-4k tier supports Video, Reference, and Transition, while kling-3.0-4k supports Video and Transition.

Reference video editing is model-specific: V6 accepts up to 10 images / 2 videos; Gemini Omni up to 5 images / 1 video; Kling O3 (including the 4K tier) up to 7 images without video or 4 images with 1 video; Grok Imagine accepts either 1–7 images or exactly 1 video. V6, Gemini Omni, and Grok video inputs lock duration to auto; Grok video framing is source-derived and no aspect-ratio parameter is sent. See skills/capabilities/create-video.md for full media constraints.

Image Models

Model	CLI value	Resolution	Aspect Ratio
Qwen Image	`qwen-image`	`720p` `1080p`	`1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
GPT Image 2	`gpt-image-2.0`	`1080p` `1440p` `2160p`	`1:1` `16:9` `9:16` `4:3` `3:4` `3:2` `2:3` `2:1` `1:2` `21:9`
Seedream 5.0 Pro	`seedream-5.0-pro`	`1080p` `1440p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Seedream 5.0 Lite	`seedream-5.0-lite`	`1440p` `1800p` `2160p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Seedream 4.5	`seedream-4.5`	`1440p` `2160p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Seedream 4.0	`seedream-4.0`	`1080p` `1440p` `2160p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Gemini 2.5 Flash (aka Nanobanana)	`gemini-2.5-flash`	`1080p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Gemini 3.0 (aka Nano Banana Pro)	`gemini-3.0`	`1080p` `1440p` `2160p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Gemini 3.1 Flash (aka Nano Banana 2)	`gemini-3.1-flash`	`512p` `1080p` `1440p` `2160p`	`auto` `1:1` `16:9` `9:16` `4:3` `3:4` `5:4` `4:5` `3:2` `2:3` `21:9`
Gemini 3.1 Flash Lite (aka Nano Banana 2 Lite)	`gemini-3.1-flash-lite`	`1080p`	`auto` `1:1` `3:2` `2:3` `3:4` `4:3` `4:5` `5:4` `9:16` `16:9` `21:9`
Kling Image O3	`kling-image-o3`	`1080p` `1440p` `2160p`	`16:9` `9:16` `1:1` `4:3` `3:4` `3:2` `2:3` `21:9`
Kling Image V3	`kling-image-v3`	`1080p` `1440p`	`16:9` `9:16` `1:1` `4:3` `3:4` `3:2` `2:3` `21:9`

Voice / TTS Models

Model	CLI value	Provider	Max characters
MiniMax Speech 2.8 HD	`speech-2.8-hd` (default)	MiniMax	10,000
MiniMax Speech 2.8 Turbo	`speech-2.8-turbo`	MiniMax	10,000
Eleven Multilingual v2	`eleven-multilingual-v2`	ElevenLabs	10,000
Eleven v3	`eleven-v3`	ElevenLabs	5,000
Eleven Turbo v2.5	`eleven-turbo-v2.5`	ElevenLabs	40,000

Music Models

Model	CLI value	Provider	Explicit lyrics	Auto lyrics	Instrumental	Image ref
MiniMax Music 3.0	`music-3.0`	MiniMax	Yes	Yes	Yes	No
MiniMax Music 2.6	`music-2.6` (default)	MiniMax	Yes	Yes	Yes	No
ElevenLabs Music V2	`music-v2`	ElevenLabs	Yes	Yes	Yes	No
ElevenLabs Music	`music-v1`	ElevenLabs	Yes	Yes	Yes	No
Google Lyria 3 Pro	`lyria-3-pro-preview`	Google	No	Yes	Yes	Up to 10

For AI Agent Developers

These skills are designed to be loaded into agent context. Each skill file is self-contained with:

Decision trees — help the agent choose the right command
Flag tables — every parameter with allowed values and defaults
Model reference tables — per-model parameter constraints
JSON output schemas — exact response format for parsing
Exit codes — deterministic error handling
Examples — copy-paste-ready commands

Start by loading skills/SKILL.md as the entry point, then load specific capability or workflow skills as needed.

Community Skills

Projects built on top of PixVerse CLI by the community:

Project	Author	Description
pixverse-character-pipeline	@takamasa045	Character-driven video production — one speaker image + YAML config → multi-language, multi-ratio talking-head videos with lip-sync, BGM, and Remotion rendering
pixverse-shotpack	@takamasa045	Creative brief → video shot pipeline — transforms markdown briefs or YAML storyboards into organized, editor-ready AI-generated video assets

Have a project built on PixVerse CLI? Open a PR to add it here.

Credits

Thanks to the creators whose work has contributed to the PixVerse ecosystem:

@takamasa045 — for building character pipeline and shotpack production tools on PixVerse CLI
@vista8 — for the Mondo poster design system whose prompt engineering and artist style library are adapted in pixverse:mondo-poster-design