leonardo-cli

A Leonardo.Ai command-line interface for generating images, videos, upscales, and more.

Designed to be both AI-agent friendly (--json, deterministic exit codes, stderr logs) and human friendly (colors, progress, helpful errors).

$ leonardo generate "a cyberpunk cat" -o cat.png
→ Submitting generation (1024x1024, n=1)
  generationId: 4f8e...
→ Polling for completion...
  attempt 1 — status: PENDING
  attempt 2 — status: COMPLETE
✓ Saved cat.png (812345 bytes)
done 1 image(s)

Install

Requires Bun ≥ 1.1.

git clone https://github.com/mrgoonie/leonardo-cli.git
cd leonardo-cli
bun install

# run from source
bun run leonardo --help

# or build a single executable
bun run build && ./dist/cli.js --help

After npm i -g leonardo-cli you get two binaries pointing to the same CLI:

• leonardo — full name
• leo — short alias

API Key

Get a key at https://app.leonardo.ai/api-access.

The CLI resolves your API key in this order — first match wins:

| Priority | Source | Example | |---|---|---| | 1 | --api-key flag | leonardo --api-key lk_… me | | 2 | LEONARDO_API_KEY env var (incl. OS-level shell) | export LEONARDO_API_KEY=lk_… | | 3 | .env.local / .env.*.local / .env (walks up from cwd) | see .env.example | | 4 | ./leonardo.config.json (walks up from cwd) | see leonardo.config.example.json | | 5 | User config: $XDG_CONFIG_HOME/leonardo-cli/config.json (default ~/.config/leonardo-cli/config.json) | written by leonardo config set |

Inspect what was resolved:

leonardo config path        # paths + which source the key came from
leonardo config show        # effective config (key redacted)
leonardo config set apiKey lk_xxx

Output

Single file:

leonardo generate "a sunset" -o sunset.png

Directory (auto-named: leonardo-<id>-NN.png):

leonardo generate "a sunset" -d ./outputs -n 4

Skip download (URLs only):

leonardo generate "a sunset" --no-download --json

Default output dir falls back to LEONARDO_OUTPUT_DIR env, then outputDir in config, then ..

Commands

| Command | Description | |---|---| | generate <prompt> (gen) | Text-to-image with polling and download | | video <prompt> | Text-to-video (Motion / Veo / etc.) — returns a job id | | upscale <imageId> | Universal Upscaler | | variation <variationId> | Fetch upscale/variation result | | status <generationId> | Check status + image URLs | | models | List platform models (id\tname) | | me | Account info + remaining API credits | | config get/set/path/show | Manage config |

Run leonardo <command> --help for full options.

Generate options

-m, --model <id>        Model UUID
-w, --width <n>         Image width  (default 1024)
-h, --height <n>        Image height (default 1024)
-n, --num <n>           Number of images
--alchemy               Enable Alchemy
--ultra                 Enable Ultra mode
--contrast <n>          Contrast (1.0-4.5)
--style <uuid>          Style UUID
--seed <n>              Seed
--negative <text>       Negative prompt
--enhance               Auto-enhance prompt
-o, --output <file>     Output file (or directory)
-d, --output-dir <dir>  Output directory
--no-download           Skip downloading; print URLs only
--wait-timeout <ms>     Polling timeout (default 480000)

AI-agent mode (--json)

Pass --json (or LEONARDO_JSON=1) to emit a single JSON object on stdout. All progress goes to stderr; colors are disabled.

$ leonardo generate "a cat" --json | jq '.images[0].path'
"./leonardo-4f8e0a2b-01.png"

Sample shape:

{
  "ok": true,
  "generationId": "4f8e0a2b-...",
  "status": "COMPLETE",
  "prompt": "a cat",
  "model": "6b645e3a-d64f-4341-a6d8-7a3690fbf042",
  "width": 1024,
  "height": 1024,
  "images": [
    { "id": "...", "url": "https://...", "path": "./leonardo-...-01.png", "bytes": 812345 }
  ]
}

Exit codes

| Code | Meaning | |---|---| | 0 | Success | | 1 | Generic failure | | 2 | Usage / argument error | | 3 | Missing or invalid auth | | 4 | Leonardo API error | | 5 | Polling timeout |

Streams

• stdout — JSON object (in --json mode), or primary data (URLs, IDs)
• stderr — human progress, warnings, errors

This separation lets you safely pipe stdout into jq / shell scripts without contaminating it.

Development

bun run dev <command>   # run from source
bun run typecheck       # tsc --noEmit
bun run build           # produce dist/cli.js (single file)

Project layout:

src/
  cli.ts                  # commander entry
  config/                 # api-key + config resolution
    paths.ts
    schema.ts
    load-dotenv.ts
    resolve.ts
  api/
    client.ts             # fetch wrapper
    types.ts
  utils/
    log.ts                # JSON-aware logger
    poll.ts               # status polling
    output.ts             # download + filename utils
  commands/
    generate.ts video.ts upscale.ts status.ts
    models.ts me.ts config.ts

License

MIT © mrgoonie