This project provides a comprehensive skill for OpenAI Codex to automate the end-to-end production of paper-cut collage advertisements. It targets users seeking to streamline creative workflows, offering automated generation of concepts, scripts, storyboards, animations, voiceovers, and final MP4 outputs, complete with local voice cloning capabilities.

How It Works

The skill integrates with OpenAI Codex, processing product information to generate a visual metaphor, script, and storyboard with timing. It then proceeds to create style-locked keyframes, animate them using techniques like Seedance or FFmpeg, generate narration via standard TTS or local IndexTTS-2 voice cloning, add sound effects, and perform final MP4 quality control. The approach leverages local processing for voice cloning and animation, reducing reliance on external APIs for core creative assets.

Quick Start & Requirements

Installation involves cloning the repository into the Codex skills directory, either globally (~/.codex/skills/paper-collage-ad) or project-specifically (.codex/skills/paper-collage-ad). A restart of Codex is required.

• Primary Install: git clone <repo_url> ~/.codex/skills/paper-collage-ad
• Prerequisites: macOS with brew install ffmpeg node.
• Voice Cloning: Requires an Apple Silicon Mac for IndexTTS-2 MLX. Installation via bash scripts/setup-indextts2-mlx.sh. Users must provide a reference.wav (6-12s) and run bash scripts/prepare-indextts2-voice.sh.
• Optional Services: Seedance, 即梦, MiniMax, ElevenLabs require user-provided API keys/services.
• Links: Detailed Chinese workflow available at WORKFLOW.zh-CN.md.

Highlighted Details

• Automated extraction of visual metaphors from product data.
• Generation of scripts, storyboards, and style-locked keyframes.
• Animation via Seedance, HyperFrames, layered PNGs, or FFmpeg.
• Local, zero-shot voice cloning and emotion control using IndexTTS-2 MLX on Apple Silicon.
• Final MP4 output with H.264/AAC encoding and stream-level quality validation.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or project roadmaps were found in the provided README.

Licensing & Compatibility

The original content of this skill is released under the MIT License. However, third-party components, including models, runtimes, fonts, music, and assets, are subject to their own respective licenses and terms of service. Notably, IndexTTS-2 model weights are not included in the repository. Users must possess explicit permission to clone any voice used for narration, and AI-generated narration must be disclosed.

Limitations & Caveats

The IndexTTS-2 voice cloning functionality is specifically designed for and requires an Apple Silicon Mac. The repository does not bundle any proprietary voice assets, model weights, brand materials, or final project renders; these must be supplied by the user. Users are responsible for ensuring they have the necessary permissions for any voice data used for cloning. Optional external services necessitate separate user accounts and API key configurations.