Release roundup—November 21st, 2025

You can now select Gemini 3 as your AI model in Underlord. Google's latest model has been impressive in testing for complex tasks and multi-step workflows. The areas it does well:

To switch models, open the model picker in your Underlord settings.

You can now select Nano Banana Pro as your image generation model in Descript. This is Google's state-of-the-art model for high-resolution image generation and precise editing.

Nano Banana Pro offers many improvements of over the original Nano Banana model:

To use Nano Banana Pro, go to Settings → General and select it from the image generation model dropdown or select it in the Generate an image panel using the cube icon.

You can now extend video clips using Underlord when you need more footage. The new Extend Video option generates additional frames from the last moment of your clip, making it useful when a clip is too short or you need more content without reshooting. Or when you just need to get weird with it.

To extend a video:

The extension uses the last keyframe of your clip as the starting point. If you have scenes in your project, the extension will automatically fill the remaining time in the current scene up to your model's maximum duration.

For best results, choose clips with smooth, steady motion like a person walking or a camera panning. Avoid clips with abrupt cuts or mid-action endings.

You can extend generated clips, recorded media, or imported content.

When you translate a video with dubbing, you'll now see a progress indicator in the sidebar that shows which translations are currently processing. You’ll also see indicators above each section as it’s being translated in real time.

The updated loading state makes it easier to track multiple translations at once and gives you a better sense of how long the process will take.

We’ve introduced some improvements to our language dubbing engine. When you translate and dub a video, the translated audio should match the timing of the original audio more closely. The updated translation system better preserves the pacing and rhythm of your original speech when generating translations, so the dubbed version sounds more like your natural delivery in the target language. This improvement applies automatically when you translate and dub your videos—no additional steps required.