AI Podcast Script Generator 100% Local

Paste bullet-point notes or a rough voice memo transcript — the AI expands them into a structured podcast script with an intro hook, segment transitions, and an outro. Runs entirely in your browser via transformers.js. No data is sent anywhere.

Loading AI model…

⚡ WebGPU detected — generation will be faster.

Your Podcast Script

How it works

The generator uses a small instruction-tuned language model (LaMini-Flan-T5-783M) loaded via transformers.js and compiled to WebAssembly (ONNX). Everything runs inside your browser tab — your notes never touch a server.

1. Paste your notes Bullet points, rough sentences, or even a scrambled voice memo transcript all work well as input.
2. AI structures the script The model expands your notes into an intro hook, labelled segments, smooth transitions, and a closing outro.
3. First run downloads the model The model weights (~300 MB) are fetched once from Hugging Face CDN and cached in your browser. Subsequent uses are instant.
4. Copy or download Copy the finished script to paste into your recording notes, or download it as a plain .txt file.

Privacy: This tool is 100% client-side. No login, no usage tracking, no API calls. Close the tab and nothing leaves your device.

Frequently asked questions

Does this tool send my notes to any server or AI API?
No. The AI model runs entirely inside your browser using WebAssembly (ONNX format). Your notes, topic, and the generated script never leave your device. There is no backend, no API key required, and no account. You can even use it offline after the model has been downloaded once.
What kind of notes work best as input?
Bullet-point outlines work extremely well — one bullet per talking point gives the AI clear structure to expand. Rough stream-of-consciousness notes also work; the AI reorganises them into a flowing script. You can even paste a lightly edited voice memo transcript and let the tool reformat it into proper scripted segments. Aim for at least 3–5 distinct points to get a multi-segment script.
How long will it take to generate a script?
The first time you click Generate, the model weights (~300 MB) are downloaded from Hugging Face CDN and cached in your browser — this typically takes 30–90 seconds depending on your connection. Once cached, generation takes 10–40 seconds depending on your device. Computers with WebGPU support (Chrome 113+ on modern hardware) will be noticeably faster than the WebAssembly fallback path.
Can I edit the script after it is generated?
Yes — the output box is a read-only preview for easy copying, but once you paste it into your notes app or script editor you can freely edit it. The AI output is a structured starting point, not a final product; most podcasters tweak 20–30% of the generated text to match their natural speaking voice.
What is the difference between the length and tone options?
Length adjusts the target word count and number of segments (Short ≈ 700 words / 2 segments; Medium ≈ 1,800 words / 3–4 segments; Long ≈ 4,000 words / 5–6 segments). Tone shifts the register: Conversational uses casual contractions and direct address; Educational uses clear definitions and numbered explanations; Storytelling opens with an anecdote; Interview-style frames content as questions and answers.