How Descript Adds Generative Video to Your Workflow (Full Transcript)

[00:01:16] Speaker 1: You Hello, everyone. to Descript Season 11. I am Christiana, Community Manager here at Descript. Thank you so much for joining us today and spending some of your Tuesday with us. Today's event is all about adding our new generative video features to your existing workflows, and I have to say the use cases are endless. Maybe you want to make some cool bespoke b-roll or animate your text layers. Maybe you want your next quarterly update video to be narrated by your boss's cat or a cool sea creature. Whatever it is, we don't judge. We're just here to make sure you have everything you need to make a video that you're proud of. And it's never been easier to do that. And we're going to show you how to get started with generative video today. So here is what the next hour is going to look like. We're going to kick things off by hanging out with our friend, Ramdi. If you've ever been to our YouTube, you know Ramdi. He had a recent generative video adventure, so we're going to check that out. Then I'm going to bring up our amazing product manager, Alex, and he's going to give us a generative video workflow demo. Then Alex and I will be joined by our CEO, Laura, for some live Q&A. So send your questions in the chat as we get going here, and then we'll close out with some next steps and ways for you to get started. Before I jump into that video, though, I want to turn a question over to you, the amazing audience. Here's our question for you. How are you incorporating generated video in your workflows today? What are your current use cases? Let us know in the chat because we're talking about this for an hour. We want to know what you guys have gotten started with so far. So let us know in the chat. And also, I'm seeing folks from all over the world saying where they're tuning in from. So tell us where you're tuning in from, too. Okay. Cool. So answer this question in the chat. We'll watch this video and then we'll come back and we'll jump in to our demo. Thank you again for being here and I'll see you after the video.

[00:03:46] Speaker 2: Now you can generate video into script. That's right. The best generative models built right into the editor you already use. So now you can do what nobody else lets you do. Create and use generative media directly in your edit with no fuss or clunky workflows all in one place. Need B-roll? Done. Social clips? Easy. Wild backgrounds? Animated titles? possibilities are kind of endless and that kind of power could really get to you if you're not careful, you know I mean being able to create anything that comes to mind. That's that's crazy, you know that is I can make anything I want Like a cinematic dolly zoom of a cowboy armadillo, ants having a birthday party, a giant squirrel kaiju, or a music video starring a single brain. Wait a minute. Why stop here, confined to a screen? Why can't my camera monitor be a rainbow pineapple? Why can't I actually be in space? With a robot army? What if I could turn this mug into an animated sock puppet? What if I could change all the cars in the world into high wheels? And make holograms real? What if I gave the script a whole new CEO? What just happened?

[00:05:23] Speaker 3: I don't know, but I'm keeping your parking space.

[00:05:36] Speaker 2: Hello, I think I did too much. I'm gonna take a break.

[00:05:52] Speaker 1: Listen, legend has it that Ramdi has gone through 20 descript shirts since discovering generative video. He's just taking it that that seriously. Such an inspiration. Okay. Thank you, everyone, for answering that original question we had. Loved seeing your use cases that you're doing, you're using so far. And it's a perfect segue to bring on our product manager, Alex, because he's going to show us how to do a bunch of use cases. So, Alex, I'm bringing you up. Oh, or am I? Let's see. There we go. Hey, Alex, welcome to the stage. Hi.

[00:06:33] Speaker 4: It's good to be here. I'm just droning from that video. I love that video of Ramdi. It's one of my favorites.

[00:06:39] Speaker 1: It's the best. I always wonder each season how he's going to outdo himself, and he continues to raise the bar. But congrats on all these features being out. It's been a pretty exciting couple of months, and we have a few thousand folks signed up for this event, which is really I think that means people are ready to learn. So maybe without further ado, you wanna start showing us how to use this stuff?

[00:07:05] Speaker 4: Yeah, that sounds great. Let me share my screen real quick so we can see what I'm looking at. Cool. And can you see that? I hope you can.

[00:07:13] Speaker 1: Yes, we can.

[00:07:14] Speaker 4: Great, nice. So I wanna talk about use cases for generative video. And I'm really excited to do this demo because generative video is, I have a lot of fun using it, but it's also very useful. And sometimes people will say, Okay, these are really cool videos that I see, whether I see them on social, but how do I use this at work? How do I use this to create videos and make my own videos better? So I wanna show four ways to do that. And we'll be going through these in the demo. So we'll learn how to turn any image into a video, how to animate your company logo, how to create like an advanced case of creating a video with both timing controls and audio, and then giving yourself a looping background so that it looks nicer than where we are now. Yeah, so let's jump in, does that sound good?

[00:08:03] Speaker 1: That sounds awesome, let's do it. And I'm seeing some of the use cases people brought up in the chat, so perfect.

[00:08:09] Speaker 4: Great, so I'm gonna start off by playing this video, that's like, it's a video that does not have generative video in it yet. So we can get a sense of how we might add to this, but it's a great video that Adrian put together, we can take it to the next level with some generative video. So I'll play it.

[00:08:26] Speaker 5: Ever feel like your marketing's mired in chaos? Spreadsheets, drafts, edits flying everywhere? What if you could make it all go away with a click? That's the magic of StreamSpeed. StreamSpeed's proprietary technology will organize, automate, and power up your process so you can get back to what you're good at, marketing.

[00:08:47] Speaker 4: So it's a pretty good video, but we're going to take it to the next level.

[00:08:53] Speaker 6: Cool.

[00:08:53] Speaker 4: And there's a few places we're gonna do this. We're gonna start off with what I consider like a pretty basic case, but also a really powerful one. And that's in this scene right here, where we have this sort of like kind of futuristic computer setup. And this is just a static image. Now this could be an image that you took with your camera that you found in the stock gallery or that you generated, but it's static right now, right? It's not moving and it's okay, But let's see what we can do by making it actually animated. And so the way we do that in Descript is all you need to do is just click on the image in the editor and then click on the Underlord button. And then you'll see some options here. And the one that we're going to use is turn into video. So what that'll do is it'll take you into our Generate Video AI tool, and it'll pull in that image. It'll also show you, by the way, history of any of the other videos you've created, which I was practicing. That's why you see those here. But it'll pull in that image and then it'll have a little prompt box here for you to give a direction for what you want the image to do. So I wanna save us time in this demo. So I'm gonna paste in a prompt that I wrote and then I'll talk about what it is as we're generating it. So I pasted in this prompt. I'm gonna generate this video And as you can see, what I want to do here is I want to animate the UI panels in this. I want these sliders to go left and right. I want these toggle switches to change state and the gauges to do some stuff, right? I want these charts to go up and to the right. We always want charts to go up and to the right. It's very, very important that they go up and to the right. And so now we have the power to make them go up and to the right. And as we're doing this, the computer is doing cool stuff that it's displaying things in the dashboard, right? So this is image plus what we want the image to do. That's the prompt. Now optionally, I'm not doing this here, but if you want, you can also choose a style to animate in. And so this would be like a visual style. These work well on video. They also work really well when you're generating images. And what's nice is we have these 20 or so styles and you can use them throughout your video if you like to have a consistent look from image to image and video to video. So I'm not gonna do that here because we already have an image to pull from. And so I'm really gonna use the style of the image for my video. But if I wanted to use a style, I also could. Awesome. Does that make sense?

[00:11:31] Speaker 1: It makes total sense. Yes. And while that's generating Alex, I ran a poll in the YouTube chat that I wanted to run by you because we're getting some answers here. I asked the audience, how familiar are you with generative video? I'm curious if this surprises you. 34% said, not familiar at all. 42% said, somewhat familiar. 16% said, familiar. And only 8% said, very familiar.

[00:12:02] Speaker 4: That's great. That doesn't surprise me. I think that's, I think actually if you look across like even Descript, maybe not like today, but not long ago, that's probably around what I would put it at, right? And I think it's fairly new technology and it's pretty recently gone from being novel and like not really sure if this is any good at all to being like actually very useful to create videos as we'll find out in the rest of this demo. So I'm not too surprised, but that's really cool. That's great to hear. It's all about moving people from never heard of it to very familiar.

[00:12:36] Speaker 1: Exactly, yes. I think by the end of this event, we'll get some more folks in the familiar category.

[00:12:44] Speaker 4: So while we're waiting for this, I'll just show one more thing, is that we also let you select from a variety of models to generate with. And we've got basically all of the latest and greatest models here. I'll talk a little more about their strengths a little bit later. But just so you know, I'm generating this one with VO3. And I can do things like determine the aspect ratio, the resolution, a few other things. But we'll get to that in a minute. Let's see how that video turned out. It just finished loading here. So let me play that. That's pretty cool.

[00:13:24] Speaker 1: It's pretty awesome.

[00:13:26] Speaker 4: Pretty cool, right? So let me just go and replace the image with this video. And I'm going to click on this Replace Layer button to do that. So now if I go back and play this here, yep, this is where it is. And you notice that it generated with sound. VO3 is a model that generates with sound. I'm gonna turn that sound down a little bit. I still actually do like the sound, but in the script, you can change the sound. You can even mute the sound of any layer, including generative video layers. So let me just turn this down a little bit and I'll still play it, see how this looks.

[00:14:00] Speaker 5: Cool. proprietary technology will organize, automate, and power up your process so you can get back. So already better, right?

[00:14:10] Speaker 4: Like we're making progress here. This video is coming together.

[00:14:13] Speaker 1: It's way more interesting.

[00:14:15] Speaker 4: And that's like one of the more basic, but again, very powerful cases because you probably already have images that you're working with.

[00:14:22] Speaker 1: Yeah.

[00:14:22] Speaker 4: But now you can also make those images animate and really bring a lot of life to your video.

[00:14:28] Speaker 1: And customize them so easily. Like, that's such a good example of like, generate a video that plays really well with the recorded media you already have. It's right, built right into your editor.

[00:14:40] Speaker 4: Yeah, I love that you can, you know, it may not be exactly the way you wanted it the very first time, as far as like volume, for example, but you've got all the tools of the editor at your disposal to just make it how you wanna make it.

[00:14:50] Speaker 1: Exactly, super cool.

[00:14:52] Speaker 4: Okay, the next one I'm also really excited about. This, we remember this scene where we just have like an empty soundstage. you might be asking why, like, well, because demo. So let's, we're talking about StreamSpeed, which is the name of this company, right? This is where Adrian's talking about StreamSpeed, but let's add a logo here and let's not just add a logo, let's also make it like, let's judge it up a little bit, right? Yeah. And so how are we gonna do that? Okay, so we happen to already have a logo for this, for this company in our project files. So let me just add that logo here. There we go. Done, right? OK, not exactly. But again, having the tools of the editor here are really helpful, because we don't want this green background. We're going to take the background away. And so we're going to use our chroma key effect. And our chroma key effect lets us remove a background really nicely when that background is a consistent color. We're going to key in on this color and we're going to remove it. Now we have our logo pasted in and it's transparent. We're getting closer, but it's still static. If I'm playing this, it's still static. But if we want to animate it, what we need to do is we want to animate it from not being on screen to being on screen. And so we're going to use a combination of a first frame and a last frame to do that. So what we're going to do is we're going to take a frame. We're going to save a frame, the current frame, to project files. And that will save whatever you see in your scene editor into your project files. So that will just save the studio, the logo, together. And after that's saved, hopefully soon, yeah, great, perfect. We're going to create a video with that. And I'm going to go back into my tool, my Generate Video tool, and I am going to select as the first frame that empty soundstage. And then as the last frame, I'm going to select this one. So we've got our first frame and our last frame. And then we're just gonna give it some directions for how we actually want it to animate. And again, I'll just paste this in to save us time. And as it's loading, we can talk about this. Cool. But we wanna reveal that logo and we want it to fade in from zero to 100%. This is stuff you can do by in the editor, by the way, but what it gets interesting is you can talk about where the focus locks in, right, on the letters. And then you can talk about this other animation at the end of a soft horizontal light sweep that passes through the title and a floor reflection on the glossy floor. So it all sounds like voodoo, right? And the gist of this is this stuff is possible in really advanced editing software. But you not only need to have the software, you also need to have probably used it for a really long time. And you probably have people at your company can do that? We certainly do. But we also want to put these tools in the hands of a lot more people who don't have that kind of experience, but still want to create these great videos by just specifying images, inputs, and directions for what you want the images to do.

[00:18:32] Speaker 1: Totally. Just describing what you're looking for can sort of take away this need to be an an Adobe expert, for example, super cool.

[00:18:44] Speaker 7: So we can, as this, I think this is about to load, let's see, I was gonna say we can do something else, but I think this is about to load anyway.

[00:18:51] Speaker 1: Okay.

[00:18:51] Speaker 7: Let's see how this looks.

[00:18:53] Speaker 1: Cool. We can also throw in a quick question for the audience if we'd like, cool. Okay, well, just to, as that's finishing loading, so we did that poll that we talked about earlier where most people said that they don't feel so familiar yet. So our question is, for people who reported that they don't consider themselves familiar with generative video, what's prevented you from using it so far? Tell us in the chat why you haven't gotten started with generative video.

[00:19:28] Speaker 4: Should we look at the video?

[00:19:29] Speaker 1: Yeah, I heard it. I heard it. I know, I heard it too. It's if you want to. So let's do it.

[00:19:45] Speaker 4: Pretty cool.

[00:19:46] Speaker 1: That's really awesome.

[00:19:48] Speaker 4: I definitely could not make that in an editing tool. And I definitely couldn't do it in two minutes, right? So let me add that layer, too, so that I have that in my scene. I think that, yeah, that seems like it's good. Let's take a look.

[00:20:07] Speaker 5: Away with a click. That's the magic of StreamSpeed. StreamSpeed's proprietary technology will organize.

[00:20:15] Speaker 4: And this is another great example, by the way, where we can use the tools of the editor to make it appear in this time range we want, because as you notice, it kind of loaded in a little bit later than we wanted it to. we can just have it start midway through, right? And maybe we like, maybe we like like this portion of the video the best. And so I'm just gonna adjust that start at time.

[00:20:37] Speaker 5: That's the magic of StreamSpeed. StreamSpeed's proprietary technology will organize.

[00:20:43] Speaker 4: You get the idea, right? Yeah, totally. I can also like make it a little bit, you know, more, even align to the end, but you get the idea, right?

[00:20:50] Speaker 1: Yeah.

[00:20:51] Speaker 4: So now we've got our animated logo and we've got now two scenes with generated video in them. And let's see how we're doing on time. We have 10 minutes. Okay, I think we can, I think we can do this. Cool. You notice this first scene that has a placeholder, I guess it's actually the second scene of the video that has this placeholder layer and it's about six seconds long and it's just kind of got a blank. So what we're gonna do is we're gonna just generate video from scratch here. And this is gonna be a pretty advanced prompt. It's like, you'll see it in a second and I'll talk about it. but I just want to show you like the full capabilities of what this can do.

[00:21:27] Speaker 6: Cool.

[00:21:28] Speaker 4: So we're going to remove all the image inputs. So we're just going to start off with a fresh prompt. And this is a really long one that I'm going to hit generate on and then we'll talk about what it's doing. Cool. So this is starting off with the same kind of empty soundstage, right? It's got teal colors. It's got that glossy black floor. And what you'll notice is happening here or in this prompt is you see these timestamp ranges, right?

[00:21:52] Speaker 6: Yeah.

[00:21:53] Speaker 4: cool about this is that you can give directions to the video model at these different timestamp ranges while still creating like a continuous single video. And that gives you a lot more control for what's happening when you want to represent multiple different animations at specific times. And a lot of the best video models can consume this. And I mean, I mean, more and more of them are going to very soon. Again, I'm using VO3 here. But Sora 2 does a good job of this. Pix First V5 is reasonably good at this, and so is Kling. I just happen to like VO3 as a general purpose model that also generates audio. And there's an audio portion of this one that I really wanted to use. But a lot of these models can do this sort of thing. And experimenting with them is great, and I love experimenting with them. I'm using VO3 here. But what we're gonna do here is we're gonna start off with that soundstage and we're gonna have some papers that are like hovering in the air, but then a gust of wind is gonna sweep them away, right? And we're gonna also magically have some hologram icons appear. And then what we're hoping to nail here, and yeah, we may or may not on the first try, but we'll give it a shot, is we want to time a cursor to fly in and click on one of the icons. And we want it ideally to be timed roughly with this click when Adrian's talking about a click. But what's really lovely is let's say we miss, let's say the video model misses the timing of that a little bit. We already know that we can adjust that, right? We can speed up the clip, we can slow it down, we can move the start and end time a little bit. And I'm pretty confident that like, you know, knock on wood, I'm gonna jinx myself, but if not in the first try, then like in a very small number of tries, we can get this to work really well. Totally. I also just want to comment on because I think I would be definitely thinking this if I was looking at this right now. You're probably thinking, Alex, this is great, but how did you write this prompt? I can't write like that. Honestly, I can't either. Some folks in the chats are saying. I don't write this way. I guess I could, but what I typically do with this thing actually is I will describe what I want to happen to an LLM. I might be doing that just with Underlord. Like I might open up Underlord and say, hey, I want to create this scene where we've got papers floating in the air and then we've got a gust of wind that blows them away, right, and like all this sort of stuff. And I want it to be eight seconds long and I want the click to be midway through.

[00:24:30] Speaker 6: Yep.

[00:24:30] Speaker 4: And I'll just say like, hey, can you write me a prompt for that? And it will do that really well. And maybe I workshop it a little bit, but I get to this sort of thing actually very quickly. So no, I don't go out and like type it word by word. That's tough, but I know what I want. So I describe it and I get it back from Underlord or another LLM that I wanna use.

[00:24:47] Speaker 1: It's a really good tip. Yeah, and we are building out all kinds of educational resources as well for our community to get better and better at prompting as well. So we have some that are existing. I know we have a template library now for working with Underlord. So yeah, we'll continue.

[00:25:06] Speaker 4: Can we see how this came out?

[00:25:08] Speaker 1: Yeah, let's see it. Okay.

[00:25:19] Speaker 4: That looked pretty good to me.

[00:25:20] Speaker 1: That was really cool.

[00:25:21] Speaker 4: Let's add that.

[00:25:22] Speaker 1: Cool.

[00:25:24] Speaker 4: Let's add that and then let's make sure it's showing up where we want it to show up. I'm just going to make sure I put it into that placeholder layer.

[00:25:33] Speaker 6: Okay.

[00:25:34] Speaker 4: Then I'll just take this out. But let's see how this looks.

[00:25:38] Speaker 5: Spreadsheets, graphs, edits flying everywhere. What if you could make it all go away with a click?

[00:25:44] Speaker 4: So that's pretty well-timed, but I can fix that pretty easily in the editor to make it just right.

[00:25:49] Speaker 1: Yeah.

[00:25:50] Speaker 4: We're just going to go in here, and we are going to make it start at the beginning, and then we're going to make sure that it extends into the end. I think that looks pretty good. Let's see how this looks. Cool.

[00:26:04] Speaker 5: Spreadsheets, drafts, edits flying everywhere. What if you could make it all go away with a click?

[00:26:09] Speaker 4: That's the magic of stream speed that basically is it right just a quick little change of the starting time and that's basically it Yeah, right. Yep So we've got that I think we have time for one more thing, right? Let's see.

[00:26:22] Speaker 1: What have we gone through about five minutes left. So let's hit let's hit one more thing Yeah, we've done one two and three.

[00:26:28] Speaker 4: Let's do four. Cool. So number four is about giving yourself a looping background and That's about like, you know, look at Adrienne's background here It's a nice teal, but it's static again And some people like to have nice dynamic backgrounds or they're talking head videos and I'm gonna show you how to do that I'm gonna actually do that in a different composition That uses the exact same like source talking head video but the reason why I'm gonna use a different one is because I really want you to see it full screen not just for What is it three seconds, but for the entire 20? Yeah, so I'm gonna switch over there And it's again, that same video of Adrian it just now fully camera layout the entire time. And so let's do that. Let's create a nice dynamic background for him. That's still based on the one that he was using here before. And so how are we gonna do that? Well, we're gonna use a few different tools. The first one is gonna be a tool to remove Adrian from the foreground. And so I'm gonna use that same feature we talked about a moment ago, which is to save the frame to our project files. So that'll just save an image, like this image that I'm looking at, to our project files. And then when I go and find that image in our project files, I am just going to edit it again like this by clicking on Underlord and saying Edit Image. And I'm just going to say, remove the person from the foreground. And as you can see, I was practicing earlier, so I already have one like that. But what this does is it'll take the image, and it will follow the instructions, just like the video models do. The image models do this, too. Quick tip, when doing this sort of precise editing, I want this element taken away. I recommend this model, Nano Banana from Google. But there's lots of other models that do other things really well. context is I think great for restyling like it's a little bit more aesthetically diverse um but but not as quite as precise for editing as nano banana but again we have we have um education around this right totally so here's our image and this is where it gets really fun is I'm going to take this image and I'm going to turn this into a video so this is a little like combination workflow. And not only that, but I'm going to give it the same starting frame and ending frame. And there we go. Great. And then just a pretty simple prompt here like this will animate that background so that it loops, right? So that it'll have the same beginning and ending and while it the video itself might be eight seconds long I can use the editor to loop that video and it'll be one continuous looking take of a dynamic background.

[00:29:38] Speaker 6: Cool.

[00:29:39] Speaker 4: And so let's see we have one minute I think we can I mean we can basically hit it.

[00:29:44] Speaker 1: Yeah for sure.

[00:29:44] Speaker 4: We'll get close we can run over a minute.

[00:29:46] Speaker 1: Yeah we're flexible on our timing here. We're pretty on time. Yeah yeah no you're doing great you're doing great we covered a lot of ground and really cool stuff in like 25 minutes. So it's all, it's all good. But Alex, maybe while this is generating, you've talked about like some of the different benefits of the different models. How did you, how would you recommend folks get familiar with the different models and sort of build up a perspective to pick?

[00:30:19] Speaker 4: I mean, truly, I think it's about experimentation. And that is definitely part of it. And we have some content around recommendations about what models are good around what things. So we even have some of that in this model picker that gives you an idea of what things are good at. But the best way to develop a perspective yourself is to use the models. You start to develop fairly quickly. It's an investment of like an hour of your time, and then you kind of get it. Cool. OK, so that, I think, worked, right? So we have this video. Let's just add that in to our composition. Here it is. As expected, it's going to cover up the top. And then let me make sure that I can see it. OK, so I've got my script here. That's Adrian. And then I've got my background behind it. And then the only thing I need to do now is I need to remove Adrian's, like the background in his original video. But we already have a feature for that. That's just green screen or yeah, green screen. So we do that. Now Adrian's background is removed. And then last thing I need to do is I wanna set this on Luke, right? So there's my video. We don't need audio for this one, and let's make it loop.

[00:31:51] Speaker 1: Sweet.

[00:31:52] Speaker 4: Then let's play that.

[00:31:55] Speaker 5: Feel like your marketing is mired in chaos, spreadsheets, drafts, edits flying everywhere? What if you could make it all go away with a click? That's the magic of StreamSpeed. StreamSpeed's proprietary technology will organize, automate, and power up your process, So you can get back to what you're good at, marketing.

[00:32:15] Speaker 1: So there you go. I feel like it's awesome.

[00:32:18] Speaker 4: Yeah, pretty quick, right? You can create a nice dynamic background. It's still what you want it to be. It's still representative of your brand, your colors, all that stuff. But now you just have a little bit more movement in it. It makes it more dynamic.

[00:32:31] Speaker 1: Totally. Yeah. Adrian, Alex pimped up your video over this, over the last 20 minutes. You're going to have to send this to him. Cool. That was awesome. What a simple way to make it more interesting. I feel like all four of those examples, got through them pretty quick, but just really, really made the video more interesting. So thanks so much, Alex, for walking us through that demo.

[00:32:56] Speaker 4: For sure.

[00:32:57] Speaker 1: That was great. And we've got questions coming in in the chat. I think it's the perfect time for a little Q&A, if that sounds good to you. And joining us for Q&A is our CEO, Laura. I'm going to bring her on stage. Hey, Laura. Hey, guys. Welcome to the stream. How are you doing?

[00:33:22] Speaker 3: I'm great. Alice, good job with the demo. Thank you.

[00:33:25] Speaker 1: Yeah.

[00:33:26] Speaker 3: I'm glad nothing embarrassing happened. No.

[00:33:28] Speaker 4: Yeah, nothing weird. It's great.

[00:33:30] Speaker 1: It was a super unembarrassing demo, which, you know, maybe we would have gotten some more views if something crazy happened, but I'm happy for you, Alex. Sweet. Okay, well, let's hit some questions. We've got a couple folks backstage that we'll bring on, and then we'll do some chat questions too. So first up, I'm going to bring our friend Jethro Jones up. Jethro, welcome to the stream.

[00:33:59] Speaker 8: Hey, thank you so much. This was so much fun to watch. Alex, way to go. And what an awesome thing you just shown in 25 minutes. That was incredible.

[00:34:08] Speaker 4: It's fun. Yeah, I love using these tools. It was great.

[00:34:11] Speaker 8: Yeah, I was thinking I could do some of those things in motion and in Final Cut Pro, which is what I used to use, but it would take me forever and it would it would just take so much time. I'd have to watch a bunch of videos and then try to get it exactly right. Um, and, uh, I remember one of the first videos I made, I was, I caught a flaming tennis ball and then I switched to the other hand and then threw it off screen. And it took me, I think, six hours to do that simple thing. And I think I'm going to try to recreate that now using these tools and see how long I can do it. Cause I bet I can do it in no time at all.

[00:34:47] Speaker 7: Yeah, I think you can too.

[00:34:50] Speaker 8: Yeah, so so my question is, you mentioned while you were talking is that if you don't like something, then you can just try again. And so when I've done that with AI tools, then it either does a completely different thing, or it takes the parts that I didn't like, how do you recommend trying again and making it keep the stuff you did like, and redo the stuff you didn't like? What's your advice for that?

[00:35:16] Speaker 4: So for images, this is like a completely solved problem, in my opinion. So it's a little bit of a different answer for image and video. For images, the models that we talked about today, both of them, which were Nanobanana and FluxContext, and many more that are also available, are very good at following these instructions if you specify that you want to edit the image. So I could quickly share my screen again just so you can see what I'm talking about. Let me see if I can do that real fast. Great. So let's just take this second image for my project files here, like this, for example. If you specifically pull in the image And you say, I want to make the background red or something. It's quite good at interpreting that, OK, what is the background of this image? What is the foreground? What is the background? What does the user want to do? Let me not completely re-roll the dice. Let me actually edit the image. So it's much more of a precise editing tool than it is a let's just hope for the best again tool. So very much solve for images. For video, it's just not quite there yet. it's getting close. I bet if someone's watching this webinar in six months, they'll be like, yeah, that does that for video too. So that will happen. Um, but today the best way of doing that is to going back to that prompt that you had. Right. And so we had, let's look at a video prompt.

[00:36:55] Speaker 8: Um, well, let's take that one where you did the background. Yeah. And let's say I don't want those spotlights there. Like I, I like the lights, but I want them to be, let's say three spotlights instead of just two, right? Because one was directly behind him, you couldn't see that in the original image that you shared. So that's an example of two or instead of two have three or something like that.

[00:37:20] Speaker 4: Yeah, so I would do a few things here. There's two ways to do this. One of them is to click on that, click on Underlord and say reuse prompt. And that will again, maybe have an internet of issue or something. Oh no, here we go. That will again, show you the prompt that you used to generate that video. Let's make sure that we still have our first frame, last frame set. And let's say, did you say you wanted three? Spotlights or two?

[00:37:48] Speaker 8: Yeah, for example, let's have three instead of two.

[00:37:50] Speaker 4: Let's just do this. And you just play with the prompt and generate again. And what that'll do is it'll take back what, it'll pull in what you had used to generate the first video And then you can iterate on your prompt with all the same inputs, the same images, the same prompt, everything. And this typically works fairly well. And I am very confident that in no time it'll work really well, as well as image does.

[00:38:15] Speaker 8: Gotcha, because the other thing you could do is you could take that image and say, make this have three spotlights first and then animate it after that, right? That's another way to get a better job because that photo editing is better than the video editing right now. I think you're totally right.

[00:38:32] Speaker 3: That's my preferred workflow too, because it also is much faster to get an image generated than a video. And so if you wanna do it multiple times, I think it's useful to work with Nano Banana or whatever to get the image exactly the way that you want it and then bring that image to life with first frame, last frame, if you are kind of a control freak.

[00:38:54] Speaker 4: I agree. That was kind of, Jethro, you're right. I just wanted to like have an excuse to plug reuse prompt, but like, yeah, you're right. I would do it your way. Yeah, that works.

[00:39:06] Speaker 1: Awesome.

[00:39:07] Speaker 4: Great question.

[00:39:08] Speaker 1: Yeah, thanks for joining us, Jethro. I know you're a long time Descript user and we appreciate you being a part of the community.

[00:39:15] Speaker 8: Yeah, my pleasure. Thank you so much for all the work you're doing. You've made my life and my editor's life so much easier. So I love Descript.

[00:39:23] Speaker 1: So glad to hear it. We appreciate you. Have a great rest of your day. Awesome. Cool. Okay. I'm going to hit a couple text questions for y'all here and then we'll bring up our next guest. Okay. Is Underlord your proprietary model or is this one of the big five?

[00:39:51] Speaker 4: I'm not totally sure what the big five are necessarily, but we use a combination of models that we both have built ourselves and have not built. So the image and video models that you have seen today were all third party video models and image models that we use with Descript. But if you're referring to Underlord, like our agent, that's also using a combination of third-party models, but also lots of stuff that we've built on top of those models for them to work with video editing really well. Awesome.

[00:40:29] Speaker 1: Cool. OK, next question here. Can Descript take original video and convert that video into animation, say, like Runway?

[00:40:41] Speaker 4: Not yet.

[00:40:44] Speaker 1: Cool.

[00:40:44] Speaker 4: The best way of doing something like that is similar to Jethro's workaround, which is to keyframe that video, add a style to that image, and then animate it. But it's not quite what you're asking for. I know exactly what you're asking for, and the answer is just not yet.

[00:41:01] Speaker 1: Cool. Maybe something to look forward to then. Awesome. OK. I'm going to bring up our next guest from backstage, Ross, I'm bringing you up.

[00:41:41] Speaker 9: grow. We talk about YouTube, give feedback, that kind of stuff. So dscriptmastery.com if anyone wants to check me out, as well as dscriptmastery on YouTube. My question, my original question, was actually the last one about Runway. So I'll go to my next question, which I saw Laura mentioned in the chat, but is about copyright considerations. So since AI is trained on existing work, how does that work when we're using this material and we're publishing it to, say, YouTube? Do Do we need to, what do we need to do? What do we need to think about?

[00:42:16] Speaker 3: Yeah, I think like, so usually when people ask about copyright, like there's two different questions that they're asking and I wanna make sure I answer one sort of unequivocably, which is like, sometimes what people are asking is like, does Descript claim that they own my content? And I am here to say like, no, we don't make that claim. So like, let me just kind of say that unequivocably because I think a couple of seasons ago, someone asked that And we didn't like say that as unequivocably enough. So like we are a company that thinks this is your content. But where did that equivocation come from? Like when we've been asked that before, it's kind of the second problem, which you alluded to Ross, which is like, what is copyright in the age of AI? So you can think about like, what if one of those first frames that I feed into Descript is like a copyrighted image that does not belong to me. And I bring it to life with a video. What I would say is like, I am not a lawyer, but I do not know how solid your copyright claims would be to a video that you created from someone else's IP. And so that's where I think you really wanna be careful. That feels like probably not, you probably don't own the rights to that. And then there's a lot of in between where it's like, okay, well, what if it's not a Disney character, but it is like an image that I created like, with AI, which just, like, pulls from general knowledge, like, do I own that content? And like, I would say, like, Ross, we are all waiting for the courts to make some decisions about this. Like, um, and, and I think, like, that's just the reality of the situation is that the technology is out of, ahead of regulation. So I can be super clear that, like, Descript doesn't own your content. Uh, don't use copy, don't use obviously copyrighted images to to create and claim that that content is your own? And then there's this messy middle where we just don't know yet.

[00:44:12] Speaker 9: Yeah, that question comes up all the time on my community. And I'm not a lawyer. I don't play one on the internet. So I know how hard that is to answer. And I will say, for people listening, when you post a YouTube video, there's a box you can check that says, this contains AI-generated material. So that's a good thing to do if you are using it. But thank you, appreciate it.

[00:44:37] Speaker 1: Sweet, awesome. Thanks so much for joining us, Ross, and for everything you do for the community. Lots of love for your videos in the chat. So thanks so much, Ross.

[00:44:47] Speaker 9: Thanks for having me.

[00:44:48] Speaker 1: Of course, have a good one.

[00:44:50] Speaker 9: You too. Awesome.

[00:44:52] Speaker 1: OK, some more text questions for us here. OK, that's a good one. How do I generate multiple videos for different scenes that maintain the same style, characters, backgrounds, et cetera, as previously generated videos? I'm struggling with this.

[00:45:10] Speaker 4: Great question. So it's a combination of the tools that we talked about. And this is only going to get better as we build more of this functionality. But to break this down, you're going to want to create, essentially, image frames that are of the same characters or backgrounds. And you can generate those yourself by generating an image and then reusing that image to create another image. So for example, let's say it's a picture of, I'm looking at a water bottle, so I'm going to say a water bottle. I'm going to say a picture of a water bottle, generate that, and then edit that to create a picture of that water bottle from a different perspective, a different camera perspective. That's your other image keyframe for another scene. And then you can animate each of those into videos. So you can do that by editing, or you can also do that by that saving frames tool that I talked about. So you have a video you generated. You like a frame. You like the characters and backgrounds in that frame. Save that frame to your project files, and now use that as reference for your next video. So by doing that, you're referencing essentially videos and images you've already created to create new ones. And that's how these things typically work right now. This is moving fairly quickly, and we're going to add some functionality producing to make this even easier. So that's the current best way.

[00:46:30] Speaker 1: OK, cool. And kind of related to that is, and maybe it's the same answer, but you showed those 20 plus styles we have in the app. What should I do if I want to create my own style, like the ones in the app? Is that possible?

[00:46:46] Speaker 4: The best way of doing that now is to create a style that looks like what you want and then save that as an image. And then you can use that as a reference image. So one way you'd want to do that is, I don't think we have a style for Cyberpunk or something. The reason why I think of this because we thought about adding one and then we didn't. So you can create your own image that is a cyberpunk type of image. And then you can save that to your project files and then use that. Another way of doing it, which by the way, also is I think very, very effective, is to ask Underlord, right? And say, hey, I really like this image of cyberpunk. Can you describe this style to me, just in words? Like, just describe what I'm seeing here. Are there neon lights? Are there, I don't know, motorcycles doing weird things? I don't know what happens in cyberpunk. I mean, different things. Anyway, and so Underlord will give you a description of a style. You just tell it, I want to create a style for consistent styling and image generation, video generation, and it'll give you one. And then when you're creating images and videos, just include that text in your prompt each time. And that will start to feel like a consistent style that gets appended to the direction you're giving for the actual scene.

[00:48:09] Speaker 1: Awesome. Cool. Thank you. And you brought up using Underlord. And so I wanted to hit this question because it's a good segue. Can you show us how to do this starting from Underlord? or maybe just sharing that you can do that, whichever way you wanna take this.

[00:48:29] Speaker 4: Yeah, you can, let me see.

[00:48:31] Speaker 3: While you're bringing up your screen, maybe I'll just share one of the ways that I like to do this using Underlord, which is again, like my workflow is always to start with images. And so one of the things that I do is I prompt Underlord to create like a storyboard of images for me, and I give it like a base image character. Often my nine-year-old son, I'll say like, you know, take a series of pictures with this character as your, as kind of like your base image. And I want it to like, look like he is on the Titanic. He's obsessed with the Titanic and like look like 1912 photography as he kind of goes through each of these things. It'll create kind of the image storyboard for me. And then I can like work with those images and be like, no, the second one doesn't look good. Make it look like this. And then once the images in the storyboard look really good, I can bring each of those images to life and kind of like create the overall story.

[00:49:28] Speaker 1: That makes sense. It's a good way to start the workflow.

[00:49:32] Speaker 4: Yeah, that basically answers the question. So what Interlord is really good for is that initial brainstorming step. And it's good for the rest of these things as well. Like it can create these edits and images and videos as well. But I find that the AI tools are also great at that, especially when you want access to all the knobs. But what the AI tools that we talked about in the demo don't do a good job of in comparison is the brainstorming. So I have some creative inspiration, it's a subject matter, it's a style reference, it's a character. And I just want Unreal to help me get started. Like what's this video even gonna be about? What's it gonna look like? So you feed it what you have and you ask it to brainstorm with you and it will generate for you if you ask it however many images you want to get started. what you might find is you actually create something you didn't even think you'd be creating, but you get inspired and you co-create.

[00:50:27] Speaker 1: Totally.

[00:50:28] Speaker 3: Cool. The other thing that I would just, when you're diving into this, like spend a little bit of time studying the model picker, which is like that thing that looks a little bit like a Rubik's cube that Alex was showing you, that shows you all the different models that we have. And the reason why I would just spend a little time there is that right now, each model kind of is good at different things. And so depending on what you're trying to achieve, you can make a decision about a model that is like, you know, the cheapest to use or the fastest to use if speed is your optimization, or if character consistency is really important to you. There are some models that are better at that than others. And so I would just like, as you go from like, playing around to like 201 or whatever, spend a little time in that model picker, like looking and seeing which one's best for your use case. Totally.

[00:51:22] Speaker 1: Cool. Okay, this next question comes from awesome community member, BigChadP from Discord. Any plans on rolling out AI powered transitions? All of this video is great, but I need more transitions to go between them.

[00:51:43] Speaker 4: Is this a text question, or do we have this person? I want to ask them a follow-up question.

[00:51:48] Speaker 1: This is a text question, but if you ask the follow-up, I'm sure he'll type in the chat.

[00:51:54] Speaker 4: Yeah, I'm just really interested in the kind of transitions that you're interested in creating. This is all things that are possible, and I'd love to actually hear more about what interests you.

[00:52:05] Speaker 1: Well, let me see if I, Chad P, if you're listening, I'm going on Discord and I'm sending you an invite link to join us backstage. It's okay if you are unavailable, but feel free to join us if you want to talk about this a little bit more. I know it's a feature request you bring up quite a bit. Maybe he's FaceTiming.

[00:52:36] Speaker 4: While we wait on that, actually, can I quickly show what we were talking about with Underlord a minute ago?

[00:52:43] Speaker 1: Yeah, do it.

[00:52:44] Speaker 4: Okay. So I said, Hey, I just want to brainstorm. I've got this picture of, you can't see it very well in this small zoom, of a movie character. I want to create a movie trailer. It's some kind of futuristic thing. I don't really know what I want yet, but like, can you just help me? And so Underdog goes and says, sure, I can do that, and has generated four images for me to use with that as a reference. And so I can now take a look at these, and they look something like this, right? So this is that same character running through some kind of street, looking through some kind of futuristic cyberpunk-y computer setup, walking around those hollowed-out New York streets after the apocalypse or whatever that is, and then just looking very concerned about looking at some kind of hurricane that's coming in. Right, and so I may or may not use these, but what this is doing is it's inspiring me, right? I want to go into this direction or that direction, or maybe I want to see like who the other characters are in this movie, and I can kind of co-create from there.

[00:53:45] Speaker 1: Cool. That's awesome. Thank you. And as you shared your screen, big patch, my dyslexia is popping off. Big Chad P is joining us backstage. So I'm gonna bring him up. I love technology. That took like two minutes. Okay, Chad, welcome to the stream.

[00:54:07] Speaker 10: Hello, thanks.

[00:54:09] Speaker 1: How are you?

[00:54:10] Speaker 10: Great, how is everybody's day?

[00:54:14] Speaker 1: Great, we love doing, season events are like so fun for us. So we always love this, but yeah, do you wanna, so you had a question about transitions. Do you wanna kind of tell us a little bit more about what you mean and what you're looking for?

[00:54:26] Speaker 10: Yeah, for sure. I mean, playing around with all this stuff has been great. And just really to get it like seamless and to looking really nice. I loved when you rolled out transitions, but the only like new transition you have built is the smart transitions, which work great, but like it's a very specific style, right? And you start to see like, And then if you look at some of the other editors out there, right, like they're rolling out like six, eight, 10 new transitions a week, right? Whether it's like, you know, they have like 12 different fire ones and like 15 different like star patterns and you know, various kinds of like swipes and gradients. And I mean, there's just like, so in a world where we've got now AI to like, we can make infinite video, but to still be like, these are the five ways you can get between the videos just starts to feel like restrictive.

[00:55:31] Speaker 4: Makes total sense. Yeah, that's, I suspected you were talking about something like that. So almost similar to, we talked about styles as almost like presets for visual style. You're kind of talking about like presets for a transition.

[00:55:48] Speaker 10: Correct me if I'm wrong. Yeah, and whether that's like you all are building them Or again, in a world where so much of this now can be powered by Underlord, I would say, I guess, a solution would be to like, you create a transition that starts in the bottom left-hand corner and a fire slowly builds towards the top right-hand corner, the same way that you're doing all these other things. That click one that you showed would be awesome, but I want that click one as transition, So like every, right. So the like transition between every single thing is, is that click.

[00:56:26] Speaker 4: Yeah, that makes sense. That makes sense. So today you can do that if you are prompting it, like put that way and the combination of saved frames and prompts will do that for you. And so, but I understand that that's like not the easiest thing to understand if you're starting out, like it's certainly possible, but definitely noted good feedback. Have already been thinking about it.

[00:56:49] Speaker 3: We'll continue thinking about it. Yeah, Alex hacked it, but we can do better than hacking it. We can give you a nice.

[00:56:58] Speaker 10: Yeah, that would be awesome. That's my, Christiana in the discord knows, like that's the, any chance I get, that's been like my number one request is just like more transitions. And then especially just, yeah, with like, when the world's your oyster with creating all these different segments.

[00:57:15] Speaker 11: Yep.

[00:57:16] Speaker 10: I just want more, more ways to get between all that stuff. So thanks. And thanks for bringing me on. I appreciate it.

[00:57:21] Speaker 1: Of course. Yeah. I'm so glad we got to talk about this. Yeah. Thanks for being so active in the community. We'll see you later.

[00:57:27] Speaker 10: Absolutely. Later.

[00:57:28] Speaker 1: Bye. Okay.

[00:57:32] Speaker 3: The Descript internal chat loves Chad P, by the way. Chad, we need to send you something. We need to send you the React one. By season 12, for Chad P, we will launch more transitions. I love it.

[00:57:47] Speaker 1: Chad P, clip that out and send it to us every day. No, but yeah, that is the power of community is getting to have these conversations. It's so fun. We have three minutes left, so maybe I'll hit one more question and then we can kind of close things up, which is basically, I saw this a couple times. Let me see if I can pull it up, but essentially, like, what's next on the roadmap for generative media in Descript? What can you tell us about that we can look forward to?

[00:58:24] Speaker 4: Ooh, I mean, there's so many great things. It's like, what do I even choose? A lot of the things that we talked about are things that we're definitely thinking about, right? So we talked about making some of these features that are possible just a little bit more accessible. Yep. Let's let somebody who is maybe new, the 30x percent of people here who said, I don't know anything about this. Now, hopefully some of those people have moved out of that column today. But that's okay if a lot of them haven't, and we want to make it easier to get started. So a lot of those, we do the thinking for you and the presets for you piece is definitely on the roadmap. There's also more mediums, right? Like we have images and videos, but we haven't really talked about generative audio here at all. That's something we're thinking about too.

[00:59:07] Speaker 6: Cool.

[00:59:08] Speaker 4: And then, OK, I don't know if I should be listening. I'm going to say it, whatever. I'm really excited about this. I'm not the only one excited about this. There's definitely, I'm thinking of a few people in particular who are like, every day tell me they're excited about this. But we are going to also add lip syncing to translation, which is really exciting. And yeah, I probably shouldn't be saying this, but I am going to say it. So yeah, that'll be really exciting too.

[00:59:30] Speaker 1: You're allowed to say anything when we're live, and it's the last minute of a stream. That's the rule. Exactly. It's an immunity minute. So. Yeah. Awesome. Well, this was super fun. I had a blast doing this with you both as I always do. Let's close things up here really quick. So obviously this is the first event that we've done about generative video specifically like this. So in the chat, we would love to hear you tell us how you would like to continue to learn about this. Do you want more webinars and live demos? Do you want videos? A lot of people were like, where do I get Alex's prompt? Is it like a prompting school type thing? Like you tell us in the chat and we have a great education team here that will get to cooking those things up. So let us know. And then the last thing I just wanna say as we're closing out here is go make a video. Go use these features that we just showed you. Maybe you're making something that felt too hard, like those animations that Alex showed us earlier, or it was too time consuming, it seemed impossible. A lot of folks said, who said they weren't familiar yet. That's what you brought up as a barrier here. I hope that over this hour, we could show you that you can go do this today. And we would love to see what you make. All of us here on the stream right now are on our Twitter handles, personal or the brand ones. Our team in general just like loves to interact with folks on social. So tag us on Twitter, LinkedIn, jump into the Discord and show us what you make. And maybe we'll be reposting some in the next.

[01:01:27] Speaker 3: Yeah, and I'd say, I thought Tilly Oast say like, how about you guys give away some subscriptions and I would say, share your video, show us what you're making and we'll make it rain for you. Like we'd love to make great things happen to folks that use and love the product.

[01:01:43] Speaker 1: Completely agree. Yes, yes, yes, yes. Okay, awesome. Well, these are always so fun to do. I love seasons. Thank you all for being here with us. Alex, amazing demo. Laura, thank you for joining us as always. all of our teammates backstage that make this possible. Thank you so much. And thank you to the hundreds of you that joined and watched along with us today. Have a great rest of your day and we'll see you at the next one.