Changelog

Production-Ready Polish and AI Editor Superpowers

This week brings major reliability improvements with multi-step transcript processing, AI-powered inline script editing, and a brand-new system for extracting visual identity from your website. Plus critical bug fixes that keep your episodes hitting their target duration.

We've been focused on making the platform more reliable and giving you powerful new ways to customize your podcast content. This week's updates include intelligent transcript processing that handles long episodes, AI-powered inline editing for your scripts, and an exciting new system that automatically learns your brand's visual style.

New Features

AI-Powered Inline Script Editing

Your podcast scripts just got a major upgrade. Select any text in the script editor and a floating toolbar appears with quick actions to improve, shorten, expand, or change the tone of your content. You can also write custom prompts for more specific edits. The AI streams suggestions in real-time, and you can accept or discard changes with a single click. All edits work seamlessly with undo/redo, and the system is smart enough to preserve emotion tags and background sound markers.

Brand Style Extractor

Upload your brand guidelines or website screenshots, and the system automatically extracts your colors, typography preferences, lighting style, composition patterns, and brand mood. This rich visual profile can be linked to your podcasts to ensure cover art and future social media clips match your brand identity perfectly. The extractor analyzes everything from color psychology to photographic style, building a comprehensive understanding of your visual language.

Speech Speed Control for Gemini TTS

Fine-tune how fast your AI hosts speak. Each host can now have their own speech speed preset—slow, normal, fast, or very fast. The system injects critical pacing overrides into the text-to-speech prompts to ensure your content flows at exactly the right pace. Great for making technical content more digestible or keeping news updates punchy.

Drag-and-Reorder Script Segments

The script editor now includes a sidebar that lets you reorder segments with simple drag-and-drop. Moving your intro after a news segment or swapping discussion topics is now as easy as dragging a row. The interface includes smooth animations, speaker balance visualization, and click-to-scroll navigation to jump between segments quickly.

Improvements

Multi-Step Transcript Processing for Long Episodes

Transcript generation is now significantly more reliable for longer episodes. The system automatically splits audio into manageable chunks for episodes over 15 minutes, processes them in parallel, and reassembles the results. This eliminates timeout errors and includes intelligent brand name correction that runs concurrently across the transcript. If any step fails, the pipeline retries gracefully without losing your progress.

Smarter Script Enhancement

Script enhancement now processes each section of your script in parallel instead of trying to enhance everything at once. This eliminates timeout errors on long episodes while maintaining quality. The system also got smarter about preserving section markers—even if the AI tries to remove or rewrite them, they're automatically restored to their original positions.

Better Voice Quality from the First Word

Multi-speaker episodes now apply voice calibration to the very first chunk of dialogue, not just later chunks. This eliminates the robotic-sounding opening lines that sometimes appeared before. The system also includes enhanced mood hints for intros, giving you more natural and energetic episode starts.

Improved TTS Calibration Trimming

The text-to-speech engine now uses intelligent silence detection to trim calibration audio more precisely. Instead of cutting a fixed duration, it analyzes the audio waveform to find exactly where speech begins. This results in cleaner audio with no awkward gaps or cut-off words.

Bug Fixes

Fixed Episode Duration Blowup

Episodes are now hitting their target duration much more accurately. We fixed a critical bug where background audio regeneration was causing episodes to run 4x longer than intended. The root cause was a combination of issues: Inngest was losing script compression mutations during replay, the AI was being told it could allocate way more time per segment than actually available, and segments were accidentally repeating content from previous sections. All three issues have been resolved.

Fixed Transcript Pipeline Errors

Corrected two critical issues that were causing transcript generation to fail silently in production: JSON uploads to storage buckets now use the correct MIME type, and error handling properly propagates failures to the job system so retries work as expected.

Section Marker Preservation

Script enhancement now reliably preserves section markers like intro, segment, and outro headings. Even when the AI tries to rewrite or remove them, the system automatically restores them to their original positions and formats.

Billing Accuracy Improvements

The billing system now correctly tracks costs for all text-to-speech operations regardless of which provider you're using. Previously, some audio generation paths weren't recording usage properly.