Podframes

Turn a topic into a video podcast with talking AI avatars.

Open Source

Artificial Intelligence

GitHub

Video

Visit WebsiteSee on Product HuntGithub ⧉

All - I've been thinking about this project for a while now, and I'm so excited to finally put it out there. Not only did we launch it, we're open sourcing it 🎉 Introducing Podframes → Type one line. Get a podcast you can watch: two AI hosts, real voices, lip-synced faces, word-timed captions, rendered to MP4. Some context. A few weeks ago at Jellypod we released Speechbase: everything we'd learned about text-to-speech from years of building AI podcasts, packed into one SDK. One API, 16 of the most popular TTS providers (ElevenLabs, Google, OpenAI, Cartesia, Hume, all of them). Swap providers by changing a string. But an SDK is an abstraction, and abstractions are hard to feel. The best way to show what it unlocks is to build something with it that couldn't exist otherwise. So we built Podframes, our official open-source showcase. Here's what happens when you type a topic: → Gemini writes a tight two-host script → Speechbase mixes the voices, each host on a different provider (Ada is Google, Theo is ElevenLabs), returned as one leveled conversation with word-level timestamps. That opens up so many more options to play with. → Nano Banana 2 generates the host portraits → Every line becomes its own lip-synced clip (P-Video, or LTX-2.3 if you want the fancy take) → Those same word timestamps drive the karaoke captions and b-roll timing with zero glue code → HyperFrames renders the final captioned MP4 My favorite part is the economics. Everything is per-line: edit one line of the script and you re-buy one line of audio and one clip, not the whole episode. The studio shows you the receipt before every generation. A full episode costs a couple of dollars. And it's local. No hosted service, no signup. Clone the repo, add your own API keys, then run one CLI command, or click through the 5-step studio wizard if you'd rather watch it happen. The launch video below was made by Podframes, about Podframes. The hosts figure out mid-episode that they're the demo. Even the music is generated (Lyria 3). Because it runs on your own API keys, it's dirt cheap. I think it might be the cheapest way to create an automated video podcast out there. Now go make a show about something you love 🎙️ GitHub: https://github.com/Jellypod-Inc/... Speechbase: https://speechbase.ai (10M free characters a month to play with)