What type of content do you primarily create?
You know that feeling when you're deep in an edit and you need one very specific shot—like, a close-up of someone's hands typing on a vintage typewriter in a coffee shop? You search stock libraries for 45 minutes. Nothing. Everything's either too corporate, too staged, or just... wrong.
Or maybe you're on a deadline and you simply don't have time to coordinate a shoot. Or budget. Or both. (Usually both.)
This is where AI-generated B-roll actually becomes useful instead of just a thing people talk about at conferences. You can create exactly the footage you need in seconds, which means you can spend less time hunting through stock libraries and more time on the edit itself.
Here's how to do it
Step 1: Set up your scenes and placeholders
Start with your raw footage in Descript. Create your scenes, then add placeholders where you want B-roll to appear.
In the placeholder notes, write short descriptions of what you're looking for—this is just for your own reference, so you remember what you were thinking.

Step 2: Generate your video
Go to "Generate a video" in the AI Tools panel. Here's where you'll type your prompt, select your model, choose a style, or upload a reference image if you have one.
And here's the thing about prompting that actually matters: you need to be very specific. Like, almost uncomfortably specific. These models don't have taste or intuition—they need explicit instructions. The more detailed you are about lighting, composition, movement, and mood, the better your results will be.
If you're not sure how to write a good prompt, Underlord can help. Just tell it what you want to generate and it'll work with you to create a detailed prompt that'll give you better results.
Step 3: Add your generated footage to your edit
Once Descript generates your video, add it as a layer in your scene. Select your placeholder and change the source to your new B-roll. That's it.

Because this is Descript, you can edit the generated video like any other clip—move it around, adjust the length, trim it, whatever you need.
Bonus: Generate logo animations with a reference image
Here's a useful trick: if you want to animate a logo or graphic on top of your B-roll, place your logo where you want it in the frame, then take a screenshot. Use that screenshot as a reference image for your next generation, describe how you want the logo to move or animate, and generate. You'll get a version with your logo already integrated and animated.
When this is actually useful
- When you need something hyper-specific that doesn't exist in stock libraries. You know what you need, you can picture it clearly, but no amount of keyword variations in Shutterstock is going to find it for you.
- When you're working on a tight deadline and don't have time for a shoot. Sometimes you just need footage by end of day and coordinating a camera crew isn't happening.
- When you need proof-of-concept footage before committing to production. Generate a rough version first, see if it works in your edit, then decide if it's worth shooting for real.
- When you need B-roll of something that literally doesn't exist. Like a kangaroo-gorilla hybrid, or whatever other impossible thing your project requires.
It's not going to replace actual cinematography, but it'll get you unstuck when you're in the middle of an edit and the alternative is spending two hours searching for footage that might not even exist.
Other things Descript can do
Descript is a video and podcast editor, but it works like a document—you edit by editing text. It includes AI tools like Underlord (your AI editing assistant), Studio Sound (which removes background noise and makes recordings sound professional), AI voices, automatic transcription, filler word removal, and screen recording. The whole idea is that editing shouldn't require you to learn an entirely new skillset—it should just work the way you already think.




