civitai-gen

An agent skill that generates images, videos, audio, and more through Civitai's orchestration Workflow API. Every generation type — text-to-image, video, text-to-speech, music, transcription — shares one async lifecycle: submit → poll → download.

Works with any skills-compatible agent runtime (Claude Code, OpenClaw, Cursor, Codex, and others). It's a self-contained Node.js CLI with zero npm dependencies.

Companion: the Civitai MCP server (https://mcp.civitai.com/mcp) covers model discovery (search_models / AIR URNs) and signed-in write actions (posts, comments, articles, and more). This one is for generation.

What it does

Images — text-to-image (defaults to Flux.1, 4 images), img2img, checkpoint/LoRA selection, aspect & resolution control.
Video — 11+ engines (Veo 3, WAN, Kling, LTX2, Hunyuan, Vidu, Grok, Happy-Horse…), duration, aspect, optional audio.
Text-to-speech — built-in speakers, voice cloning, style instructions.
Music — ACE-Step song generation with lyrics and duration control.
Transcription — speech-to-text with timestamps and language hints.
Bulk & experiments — --bulk file.json batches, plus wildcard expansion and parameter sweeps.
Cost estimation — dry-run buzz estimates (cost, whatif=true) — 0 buzz spent.

Install

The portable way (auto-detects your runtime and installs to the right place):

# Install for the current project
npx skills add civitai/civitai-gen-skill

# Install globally, targeting specific runtimes
npx skills add civitai/civitai-gen-skill -g -a claude-code -a openclaw

Clone the repo and point your runtime's skills directory at the civitai-gen/ folder (the folder that contains SKILL.md):

git clone https://github.com/civitai/civitai-gen-skill
# then symlink/copy ./civitai-gen-skill/civitai-gen into your runtime's skills dir, e.g.:
#   Claude Code : ~/.claude/skills/civitai-gen
#   OpenClaw    : ~/.agents/skills/civitai-gen   (or ~/.openclaw/skills/civitai-gen)
#   neutral     : ~/.agents/skills/civitai-gen

Requirements

Node.js 18+ (uses native fetch; no dependencies to install)
A CIVITAI_API_KEY — get one at https://civitai.com/user/account
ffmpeg — optional, only for local audio post-processing

Configuration

Provide the API key either way:

# Environment variable
export CIVITAI_API_KEY=your_key_here

# …or a .env file in the skill directory
cp civitai-gen/.env.example civitai-gen/.env   # then edit

Usage

Run from the skill directory (civitai-gen/):

node generate.mjs wait --prompt "A knight at sunset" -o ./out            # image
node generate.mjs wait --engine veo3 --prompt "A robot walking" -o ./out # video
node generate.mjs tts --text "Hello world" --speaker serena -o ./out     # speech
node generate.mjs cost --engine veo3 --prompt "A robot" --duration 8     # estimate (0 buzz)
node generate.mjs engines                                                # live engine list

See civitai-gen/SKILL.md for the full command table and civitai-gen/docs/engines.md for engine/model selection guidance.

Supported runtimes

Distributed in the AgentSkills format (SKILL.md + scripts), installable via skills.sh. The CLI itself is runtime-agnostic — it only needs Node.js — so any runtime that loads agent skills can use it. Tested with Claude Code and OpenClaw.

Contributing & development

See CLAUDE.md for architecture, module layout, and how to add new step types. Tests live in civitai-gen/test/smoke-test.mjs (--readonly runs at 0 buzz).

Civitai Gen Skill

README