Blog
What we're shipping, how we built it, and lessons from the frontier of AI media automation.
Selecting an avatar for your AI video just got a lot easier. The new visual picker in AI Video Studio lets you browse and click to select from your full HeyGen avatar catalog — thumbnails, names, and one-click.
Our topic segmenter uses GPT-4.1 to cluster transcript words into semantic topics. This post explains how we hardened the prompt to ensure topic labels, summaries, and keywords are always written in the correct language — never silently falling back to English.
You can now re-run topic segmentation on any asset at any time without re-transcribing or re-translating. The new Re-segment button on the asset detail page sends existing transcript words to the segmenter, replacing only the topic layer.
A subtle but annoying bug caused captions to start at 50% left and flow off-screen in short-video mode. The root cause was applying container-type: size to the caption div itself, starving its text layout. We moved the container to the parent frame box instead.
With the launch of AI Video Studio, you can turn any article, prompt, or raw script into a finished avatar-led video with narration, B-roll, and burn-in captions. Powered by HeyGen, ElevenLabs, and your choice of LLM for script writing.
Adding dozens of news sources one by one is tedious. The CSV importer lets you paste a spreadsheet of URLs and labels, and Segriva will validate, deduplicate, and create all sources in one shot.
Newsrooms with strict data residency requirements need transcription that never leaves the premises. Our Electron app bundles a Python Whisper sidecar that runs on your GPU or CPU with zero cloud dependencies.
Content Feed article cards now display both the translation provider and the specific model used — e.g., "OpenAI / gpt-4o-mini" — so you always know how a translation was generated.
More posts coming soon. Subscribe to updates.
