What is this?
It's an AI Agent Skill that turns a topic, article, or finished story into a complete hand-drawn narrated video: an agent writes the script and storyboard and generates the illustrations, while local tools trace the strokes, hand-write the text, sync narration and export the final MP4.
It's an AI Agent Skill that turns a topic, article, or finished story into a complete hand-drawn narrated video: an agent writes the script and storyboard and generates the illustrations, while local tools trace the strokes, hand-write the text, sync narration and export the final MP4.Is this a standalone app or something I plug into an AI agent?It is a Skill package meant to be loaded into a coding agent (Codex) rather than run on its own; you drop the folder into the agent's Skills directory and the agent then drives local Python/FFmpeg scripts to actually render the video.
It is a Skill package meant to be loaded into a coding agent (Codex) rather than run on its own; you drop the folder into the agent's Skills directory and the agent then drives local Python/FFmpeg scripts to actually render the video.What do I need on my machine besides the agent?Python 3.9+ with Pillow, NumPy, SciPy and requests (installed from the bundled requirements.txt), FFmpeg/ffprobe, Chinese fonts for the handwriting and subtitles, Bun plus ExifTool for image metadata cleanup, and a working voice path (local VoxCPM2, VoiceStudio, or a pre-checked narration track).
Python 3.9+ with Pillow, NumPy, SciPy and requests (installed from the bundled requirements.txt), FFmpeg/ffprobe, Chinese fonts for the handwriting and subtitles, Bun plus ExifTool for image metadata cleanup, and a working voice path (local VoxCPM2, VoiceStudio, or a pre-checked narration track).Does it work on Windows or with agents other than Codex?The full pipeline has only been fully verified on an Apple Silicon Mac running Codex; Windows and other agent or image-model combinations have not completed full validation, so you should confirm execution, image generation, fonts and voice support before switching setups.
The full pipeline has only been fully verified on an Apple Silicon Mac running Codex; Windows and other agent or image-model combinations have not completed full validation, so you should confirm execution, image generation, fonts and voice support before switching setups.Does it do anything with image metadata or personal recordings?A bundled sub-tool, sweety-image-privacy, strips metadata from every generated image and keeps an audit record, and original voice reference recordings for the showcase videos stay on the author's machine and are not shipped in the public install.
A bundled sub-tool, sweety-image-privacy, strips metadata from every generated image and keeps an audit record, and original voice reference recordings for the showcase videos stay on the author's machine and are not shipped in the public install.Is this actually free to run end-to-end?The code itself is MIT-licensed and free, but the agent host and image generation capability may require a paid plan, and local voice synthesis still consumes hardware, power and time; the author notes one ~2.5-minute video used under 1% of a $200 subscription quota, which is not a guaranteed cost.
The code itself is MIT-licensed and free, but the agent host and image generation capability may require a paid plan, and local voice synthesis still consumes hardware, power and time; the author notes one ~2.5-minute video used under 1% of a $200 subscription quota, which is not a guaranteed cost.How many art styles and voice options are included?There are 21 built-in illustration styles (from minimalist black-and-white lines to pixel art and watercolor), 6 hand/arm presets or a pen-only mode, and 6 fixed voice options you can pick directly or let the model choose.
There are 21 built-in illustration styles (from minimalist black-and-white lines to pixel art and watercolor), 6 hand/arm presets or a pen-only mode, and 6 fixed voice options you can pick directly or let the model choose.If I just need to fix one subtitle, do I have to regenerate everything?No — changes to subtitles, pen-text or watermark/layout reuse the existing images and audio and just re-render; editing the actual narration script, however, usually forces a full voiceover re-synthesis and timeline re-alignment.
No — changes to subtitles, pen-text or watermark/layout reuse the existing images and audio and just re-render; editing the actual narration script, however, usually forces a full voiceover re-synthesis and timeline re-alignment.Can it reliably produce a whole video series in batch?You can lock a style recipe, image model, character anchors and voice reference for consistency across episodes, but character, plot or voice drift can still occur, so every episode still needs manual review — it is not an unattended batch system.
You can lock a style recipe, image model, character anchors and voice reference for consistency across episodes, but character, plot or voice drift can still occur, so every episode still needs manual review — it is not an unattended batch system.What are similar projects?native-subtitle-quote-image, livecanvas, image-story-video-wizard, brag, shuohao-skills, MoneyPrinterTurbo, lanshu-create-ai-presenter-video, hand-drawn-explainer-video-nikola
native-subtitle-quote-image, livecanvas, image-story-video-wizard, brag, shuohao-skills, MoneyPrinterTurbo, lanshu-create-ai-presenter-video, hand-drawn-explainer-video-nikolaWhich topics is it in?subtitles, voice, tts, content-creation, ffmpeg, workflow-automation, video-generation, image-generation, image-processing, skill, animation, agent-skills
subtitles, voice, tts, content-creation, ffmpeg, workflow-automation, video-generation, image-generation, image-processing, skill, animation, agent-skills