Generate images, videos, speech audio, and music through the PonyFlash Python SDK, and handle local media editing with FFmpeg.

Compatible with the Agent Skills open standard. Works with Claude Code, OpenClaw, Cursor, Codex, Gemini CLI, Windsurf, Cline and 40+ other AI agents.

git clone https://github.com/ponyflash/ponyflash-skill.git ponyflash

Then move the ponyflash folder into your agent's skills directory.

# From ClawHub (if published)
clawhub install ponyflash

# Or manually
git clone https://github.com/ponyflash/ponyflash-skill.git ~/.openclaw/skills/ponyflash

git clone https://github.com/ponyflash/ponyflash-skill.git .claude/skills/ponyflash

git clone https://github.com/ponyflash/ponyflash-skill.git .cursor/skills/ponyflash

This skill now combines PonyFlash cloud generation and local FFmpeg media processing.

pip install ponyflash
export PONYFLASH_API_KEY="rk_xxx"

Local editing does not require a PonyFlash API key, but it does require working ffmpeg / ffprobe binaries on the machine.

Check with:

bash scripts/check_ffmpeg.sh

If you also need subtitle burn-in:

bash scripts/check_ffmpeg.sh --require-subtitles-filter

The default subtitle-burn workflow prepares and removes its temporary subtitle font automatically. Only run this command when you explicitly want to keep a reusable local copy:

bash scripts/ensure_subtitle_fonts.sh

from ponyflash import PonyFlash

pony_flash = PonyFlash()

gen = pony_flash.images.generate(
    model="nano-banana-pro",
    prompt="A sunset over mountains",
    resolution="2K",
)
print(gen.url)

taskDir="$(mktemp -d "${TMPDIR:-/tmp}/ponyflash-task.XXXXXX")"
bash scripts/media_ops.sh clip --input "$taskDir/input.mp4" --output "$taskDir/clip.mp4" --start "00:00:05" --duration "8"

taskDir="$(mktemp -d "${TMPDIR:-/tmp}/ponyflash-task.XXXXXX")"
bash scripts/media_ops.sh subtitle-burn --input "$taskDir/input.mp4" --subtitle-file "$taskDir/subtitles.srt" --output "final-output.mp4"

See SKILL.md for full usage instructions.

Useful references:

• reference/operations.md
• reference/examples.md
• assets/subtitle-style.md
• assets/fonts.md

ponyflash/
├── SKILL.md
├── README.md
├── LICENSE
├── assets/
│   ├── fonts.md
│   └── subtitle-style.md
├── examples/
│   ├── quickstart.py
│   └── advanced.md
├── playbooks/
│   ├── INDEX.md
│   └── crepal-director.md
├── reference/
│   ├── account.md
│   ├── examples.md
│   ├── files.md
│   ├── images.md
│   ├── models.md
│   ├── music.md
│   ├── operations.md
│   ├── speech.md
│   ├── video.md
│   └── models/
│       ├── INDEX.md
│       └── (per-model docs)
└── scripts/
    ├── build_ass_subtitles.py
    ├── check_ffmpeg.sh
    ├── ensure_subtitle_fonts.sh
    └── media_ops.sh

The default subtitle workflow downloads Noto Sans CJK SC into a temporary directory, burns the subtitles, and removes the temporary font files when the task finishes. Only scripts/ensure_subtitle_fonts.sh keeps a reusable local copy under ~/.cache/ponyflash/fonts/. You can override that cache directory with PONYFLASH_FONT_DIR or pass explicit font file paths to scripts/media_ops.sh subtitle-burn.

• By default, clip, concat, transcode, audio extraction, frame capture, and subtitle burn-in write to temporary paths first and only keep the final --output file on success.
• Temporary concat lists, temporary ASS files, staged font directories, and partial outputs are removed automatically.
• If you explicitly pass a path such as --output or --ass-output, that path is treated as a user-requested deliverable and is kept.
• For multi-step editing workflows, create a temporary task directory first and keep staged downloads, subtitle files, and intermediate renders inside it. Delete that directory after confirming the final deliverable unless the user explicitly asked to keep intermediates.

1. Fork this repo
2. Add or update skill content
3. Follow the Agent Skills specification
4. Submit a PR

MIT