Feature Requests

Despite explicit instructions, the generated video always includes some degree of camera movement (zoom, sway, or drift), even when the prompt requests a 100% locked, static camera. The animation sometimes introduces unwanted elements (such as smoke or star twinkling) that were not requested, and fails to keep all non-flame, non-snowfall elements perfectly still. The result is that the first and last frames are not pixel-identical, breaking the seamless loop effect. These issues persist even with repeated attempts and increasingly strict prompts. Impact: This makes it impossible to create a truly seamless, ambient looping video from a still image where only specific areas (flames, snowfall) are animated and the rest of the scene remains perfectly static.

Is there a way to make it use the smart transitions between cuts? It does a really good job keeping the speaker centered! (Now that I know how to use it) but it's not very usable right now because of the jump cuts.

No Harsh Cuts has been a great addition to the Remove Filler Words tool, thanks! ( https://feedback.descript.com/feature-requests/p/filler-words-dont-ignoreremove-unless-they-can-be-cleanly-removed ) Could we have the same for the Edit for Clarity AI tool?

in order to easily change inflection at a cut point, pitch correction could help smooth out a cut and make it sound more natural.

Baber Azam

It catches duplicates of, e.g., "it its", but it doesn't catch "a an" and it should

When using Descript on a browser, it would be great to be able to access the chat conversation with Underlord even after closing the browser. There are times when I need to shut down the browser and/or computer to keep my system running smoothly, but unfortunately, that means I lose the conversation I've had with Underlord and all the work we've created together. Is there a way to autosave a chat conversation periodically in order to access it in the future?

When using "remove retakes", the AI tends to re-construct sentences using multiple takes. It tries to use the first words from one of the first takes, and stitch them with other cuts to complete a take. It should prefer using one clean retake instead of making multiple cuts. Example before: (Multiple cuts stitched together) "Individual pages are attributes. |So in this example, your entity attribute pair is |Ranking factors |And Google search Console." Example after: (Use best take to avoid cuts) "Individual pages are attributes. so in my example, the entity attribute pair is ranking factors and Google search console." When one sentence has 3-4 cuts, it sounds really choppy. There is often at least 1 smooth take that doesn't require so many cuts.

When using "remove retakes", the AI tends to re-construct sentences using multiple takes. It tries to use the first words from one of the first takes, and stitch them with other cuts to complete a take. It should prefer using one clean retake instead of making multiple cuts. Example before: (Multiple cuts stitched together) "Individual pages are attributes. |So in this example, your entity attribute pair is |Ranking factors |And Google search Console." Example after: (Use best take to avoid cuts) "Individual pages are attributes. so in my example, the entity attribute pair is ranking factors and Google search console." When one sentence has 3-4 cuts, it sounds really choppy. There is often at least 1 smooth take that doesn't require so many cuts.

The Underload feauture should be able to keep a history of prompts that were inputed in the past, so that when users come back to the project, they remember where they left off