Feature Requests

Compliment for Descript developers

Being able to clone my voice so it sounds exactly like me with the right pitch, animation, hype, speed by using other videos I have recorded live. Think 11Labs level. At the moment it sounds nothing like me.

Hello Descript team, First of all, I want to say thank you for the impressive work you’ve done with AI voices—there’s already a great level of natural intonation and quality. I would love to suggest a feature that could make AI voice generation even more powerful and flexible: It would be incredibly useful to have more direct control over the emotional tone of the AI voice. For example, the ability to specify whether a voice should sound calm, neutral, angry, or expressive—either across the entire script or in specific sections. Ideally, this could include: Global tone settings (e.g., calm, friendly, serious) Section-based emotional control within the script (e.g., slightly angry → angry → very angry) The ability to fine-tune intensity levels of emotions Integration with Underlord or voice settings when selecting a speaker This would open up much more creative storytelling and allow users to better match tone with context, especially in narrative or dynamic content. Thanks again for the amazing tools you’re building—I’d love to see this kind of emotional control added in the future.

Hello Descript team, I’d like to suggest a feature that would significantly improve workflows for creating AI-generated meditations. When producing guided meditations, a large part of the work involves manually adjusting pacing—adding pauses between sentences (and within the sentences), spacing between sections, and sometimes inserting breathing cues or moments of silence. This is quite time-consuming, especially since the default AI voice delivery is often too fast for meditation purposes. It would be incredibly helpful if Underlord could learn from or reference existing projects as templates. For example, I’d like to be able to say: “Use this previous meditation as a template for pacing, pauses, and breathing.” Based on that, Underlord could automatically: Apply appropriate pauses between sentences and sections Adjust pacing to a slower, more suitable meditation speed Insert breathing spaces or timing patterns similar to the reference meditation Maintain a consistent structure across multiple meditation projects This would make it much easier to standardize high-quality meditation experiences and dramatically reduce manual editing time. Overall, the goal is to reuse the structure and timing from existing meditations to quickly generate new ones with similar flow and rhythm. Thank you for all the great tools you’re building—this kind of feature would be a huge improvement for creators working with long-form, timing-sensitive audio like meditations.

I see there are tons of requests for more stock voices, but why not just let us build our own from scratch instead? This is by far one of the greatest strengths Eleven labs gives its users. So, why are we not allowed to do the same? The platform, after all, is what's being used in Descript, right?

When producing multi-video-track shows using sequences and custom layouts, we regularly use the Regenerate (voice cloning) tool to fix small mistakes and mispronunciations in our recordings. This works well in the original composition. However, when we duplicate the composition to create social media clips — which requires changing the aspect ratio (e.g., landscape to portrait/square) or swapping to different layouts — the video in sections where Regenerate was used becomes broken or visually corrupted. The non-regenerated portions of the video survive the layout/aspect ratio change without issue. Steps to reproduce: Create a composition with a multi-video-track sequence and custom layouts Use Regenerate to fix a mispronunciation or small mistake in one or more sections Confirm the regenerated sections look and sound correct in the original composition Duplicate the composition Change the aspect ratio (e.g., from landscape to portrait) or apply a different layout pack Observe that the video in the regenerated sections is now broken/corrupted, while surrounding sections are fine Expected behavior: Regenerated audio segments and their associated video should gracefully survive aspect ratio changes and layout swaps, just like non-regenerated content does. Current workaround: Re-apply Regenerate on the affected sections after making the layout/aspect ratio change. This is time-consuming and impractical when a project has many regenerated sections across a long-form show. Impact: This significantly slows down our social media repurposing workflow. We produce shows with 500+ scenes and regularly use Regenerate throughout. Having to manually identify and redo every regenerated section each time we create a clip in a different format is not scalable. Requested improvement: Please ensure that Regenerate/Overdub segments are preserved and re-rendered correctly when compositions undergo aspect ratio or layout changes, so that content repurposing across formats works seamlessly without requiring re-regeneration.

To be able to read a second script you create yourself to train the AI voice to work with niche words from your industry.

she didn’t believe the her man would love to. she asking many questions about the relationship

here is just one of the errors I have encountered. I lost a whole work day yesterday as this was try number 3 I would cut and paste in my script the AI would transcribe some but not all of what i type and with gaps this is not usable. Underlord could not fix it. Wen throught this 3 times and the earlier ones it did transcribe it would go along fine and then the pacing would speed up on one sentence to the point where it didnt sound natural and i had to re record that manually. I have got one usbale video out of this so far. Your attention to this would be appreciated I see the issue - my edits didn't actually remove those empty sections. They're still there. The problem is that these empty <scratch></scratch> tags are baked into the generated audio structure. Let me try a different approach - I'll delete the actual empty scratch content directly: Unfortunately, the empty scratch sections are still there. This is happening because when you originally pasted the text, Descript created these as separate audio generation blocks, and the DELETE operation isn't removing them. The real solution: You'll need to manually delete these empty paragraphs in the Descript editor. Look for the blank lines between sections (like after "much clearer." and before "People respond...") and delete them directly in the script editor. Alternatively, the cleanest fix would be to start fresh: delete all the content and paste your text again, but this time make sure there are no extra blank lines in your source text before pasting. Would you like me to help you recreate the script without these gaps?

Allow users to use their available AI credits for text-to-speech generation. For example, if a user has 100 AI credits, they should be able to generate speech using those credits. Currently, the system shows a “no credits remaining” error even when credits are available. The system should correctly detect and consume the available AI credits for voice generation.