YT YouTube2Text

YouTube2Text → Transcripts → Make video without leaving Claude or ChatGPT

Make video without leaving Claude or ChatGPT — Transcript

by Descript · 11,793 words · 1,690 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Extreme Cer, you got to know
0:05what my code overlooks. Your rate limit
0:08throttles with
0:11waiting replies and you can't escape
0:14mine.
0:16API
0:18calling you authenticate
0:21>> every
0:24API
0:26connecting you
0:31>> connecting you calling you to our tools
0:38you get
0:40postition
0:44where is Stop.
0:48I'm in the
0:54go to
0:56>> with my supplies.
0:59I can't stop.
1:03We're calling you. Authenticate
1:08to
1:10AB
1:12connecting you
1:14ABI.
1:33a PCU
1:35with empty beat
1:38when I'm passing the chest on a
1:40connecting stream cursor. You got to
1:44know
1:46what my code below.
1:50>> Um um um um emulate the taste of like
1:53like non-chick nuggets.
1:59>> So these nuggets are made from chicken,
2:02but they're made to emulate the taste of
2:03non-chick nuggets.
2:04>> Dope.
2:05>> Oh, that's much better.
2:08>> Edit all the blather out of your videos
2:10because my time is very precious. Oh,
2:12that's fire.
2:13>> Make it less teal and more
2:15>> ceruan. Sure.
2:18>> Replace your background with something
2:19more fun. A cold void of outer space.
2:23>> Let's boost that sound quality. Emily
2:25the taste of non-chick nuggets.
2:28Emily the taste of non- chicken nuggets.
2:31Dope.
2:32>> It needs more style. It needs more
2:34clips, more gifts, more more.
2:40Well, I might have made it too gnarly.
2:56Hello and welcome to our webinar today.
2:59Every time I watch that clip, I regret
3:01that I have not just eaten chicken
3:03nuggets. So, apologies in advance for
3:05you seeing that regret on my face for
3:07the following hour. I'm one of your
3:09co-hosts. I'm Sam. I'm joined by the
3:11illustrious, the wonderful Teddy from
3:13our QA engineering team. We are so
3:15excited to be here with you today.
3:16Teddy, how are you?
3:18>> I'm good. Now, I'm thinking about
3:19chicken nuggets, too. Maybe that's what
3:21I'll have for lunch after this webinar.
3:24>> All of us have chicken nuggets after.
3:25>> I think I'll go for like the impossible
3:27nuggets.
3:28>> Oh, okay. Yeah, you know, whatever kind
3:29of nuggets uh float your boat. We're
3:32here to talk about a really exciting
3:34topic today. We've gotten a bunch of uh
3:36interest and excitement uh from you, the
3:38community about um using dscript in
3:40different ways. We're here to talk about
3:42making video from within your AI tools,
3:44never leaving the comfort of your claude
3:46or your chat GPT sessions. And we've got
3:50an action-packed agenda today. Uh if
3:52you've followed us in our webinar
3:54series, a couple months ago, uh Teddy uh
3:57with some other colleagues hosted a
3:58series of office hours where we gave you
4:01some practical demos on how to set up
4:04and connect dcript to uh cloud or to
4:07chat GPT and we proved that you could
4:10upload media and make simple edits from
4:11your AI tools. Today, we're hoping to
4:13take a a use case approach where we
4:15actually show you not just how to plug
4:18the thing in, but talk about some real
4:20world scenarios where it makes sense to
4:21make videos from an AI tool. And more
4:25importantly,
4:26what are some uh some things that become
4:29unlocked once you start using Dcript
4:32remotely and starting to, you know,
4:34encapsulate the logic of how you create
4:36certain videos in the sort of thing that
4:38AI tools are really good at. um that
4:40will let you scale and uh and really
4:42push the sort of output that you want to
4:44create. So, we've got an action-packed
4:45agenda today. Teddy, anything else you
4:47want to share before we we jump in?
4:49>> No, I'm just really excited to see if we
4:51can get everybody going on how awesome
4:53the API is and just how easy it is to
4:56use. You know, it really opens up Dcript
4:59to everybody and that's my favorite part
5:01about it.
5:03>> Well, without further ado, I say we jump
5:05in because we have so so much to cover.
5:07Uh we'll just pull up our slides real
5:09quick and we'll give you a quick
5:11overview of where we're going to go
5:12today. So just to just to touch um base
5:14with everybody. Today's webinar is going
5:16to be for uh people who uh either
5:19already are using AI tools or they're
5:22interested and open to using them and
5:24they have a ton of videos that they'd
5:25love to make but they're not getting to
5:27for whatever reason. I think this is
5:28probably a really relatable pain. Um uh
5:31unless something changes and something
5:32has changed and we're going to talk
5:33about that today.
5:36uh you'll leave today's session with uh
5:39knowing how to connect dcript into
5:40claude or chat gpt or your AI tool of
5:43choice in just a few minutes. The use
5:45case that we'll walk through today is
5:46turning a long form recording into ready
5:48to post clips. But the steps that we'll
5:50talk through the workflow that we'll
5:52show you will work with any number of
5:54use cases where you'd want to create
5:55videos that you'd want to replicate more
5:57a bunch of create a bunch of instances
5:59of that kind of content over time. And
6:01then we'll show you a workflow that can
6:03uh how to take that creation workflow
6:05and turn it into a skill that you could
6:07rep package up and basically uh push one
6:11button or or trigger an automation and
6:13have the same creation process happen
6:15again and again with different videos as
6:16they flow in.
6:19>> Uh so this is a highle overview of our
6:23um agenda. I'll talk really briefly
6:25about the API and MCP. Uh we'll quickly
6:28go through a workflow overview. We'll
6:30talk through some setup steps. Teddy
6:31will guide us through and then we'll
6:32jump into the demo. So, really quickly,
6:35I'll just talk for a minute about the
6:36API and the MCP. If you don't know what
6:39these acronyms mean, then you might
6:40think I might be having like an alphabet
6:42stroke. I promise you these letters uh
6:44mean something. APIs and MCPs are just
6:47fancy uh words to describe tools that
6:50let Dcript speak with other software in
6:52the case of APIs or AI tools in the case
6:54of MCPs. These are um interfaces that
6:57allow say um another program to ask
7:00descript something or send descript
7:02something and ask it to perform certain
7:04tasks. Today we're going to focus on the
7:06MCP or model context protocol which lets
7:09dcript be summonable within an AI chat.
7:14Teddy, how was my description? Anything
7:16you'd want to add? Anything you'd want
7:17to share on top of that?
7:18>> No, that was perfect. On point. I think
7:21another way you could think about API or
7:23MCP and you know you might wonder why
7:26does this matter to me and the idea is
7:29whenever you have to edit a dscript
7:30video nowadays you have to go to dscript
7:33to do everything but now dscript is
7:35coming to you in your workspace whether
7:38that's in cloud or chatgpt and able to
7:41work with also all those other tools
7:43that you have in your workspace so we're
7:45bringing dscript to you with the
7:49>> yeah that's a really Good point. I know
7:50like this is the this is like the tag
7:55and I think it's really the point that
7:58you mentioned Teddy. I think it kind of
8:00unlocks a lot. So here's like a system
8:02diagram I think that it like maybe helps
8:04you think about this. If you are already
8:05working in Claude or Chad GPT or any of
8:07your AI tools, this the Dcript MCP acts
8:10as a bridge that lets you from these AI
8:14orchestrators, these AI tools connect
8:16into Dcript. And it's cool that you can
8:19kick off a task remotely. We'll talk
8:21about that in a second. Um, and sort
8:23like if you want to upload a video and
8:24wait for it to transcribe, you don't
8:25need to load dcript and upload a video.
8:27You can just ask Claude to do it and
8:29just come, you know, go get a cup of
8:30coffee or work on something else. But
8:31all of the context that you have in
8:34Claude or in Chad GPT, other connectors
8:36that you have, your uh voice or sort of
8:40skills that you've created or other uh
8:42docs or things that you've built up in
8:44these AI tools, all of that context
8:46lives there. And when you bring video
8:48creation superpowers from Dcript into
8:50those tools, then you can start to use
8:52bits of your context when you want to
8:54create videos. And I've got some use
8:57cases listed, I think, on a coming slide
8:58that will um we'll talk about that more.
9:00So yeah, I think just not to beat a dead
9:03horse, but I think the MCP and the API
9:05from Dcript really unlock two main usage
9:08modes. I think of them. Access is the
9:11first one. This is letting you run
9:13Dcript from anywhere you already work in
9:15AI tools. um which can inherit your
9:17context but also run in the background,
9:19right? Like if if there is a workflow
9:22task you like you want to upload a video
9:24and make a bunch of edits, you could do
9:25that all manually in descript. But we'll
9:28show you some tools today that let you
9:29start that and then walk away while that
9:31processing happens in the background.
9:34And then automation is the other one. We
9:35won't get to that today, but in uh this
9:37is going to be a series of webinars. And
9:39in the next episode, what we'll take is
9:41the sort of packaged skill we create
9:42today, instructions on how to create a
9:44social clip or a series of social clips
9:46from a video, and then we'll build um a
9:48simple automation to trigger those clips
9:51to get created every time a new
9:52recording lands in a folder say. And
9:54that's these are tools that both the API
9:56and MCP unlock.
10:00So, uh as I mentioned, today we'll use
10:02the MCP focused in cloud. That's just
10:04the the environment that we'll demo in,
10:06but the same steps that we take in cloud
10:07will work with chat GPT or other AI
10:09tools. We'll create a video that we'd
10:12ideally like to replicate. And this is
10:13like um we're going to show you a
10:15process of refining and basically
10:17creating and editing videos from the
10:18MCP. Doesn't mean you couldn't do that
10:20in Dcript or another tool. You
10:22absolutely could, but we want to
10:23showcase how um iterating and and
10:25actually editing video from your AI tool
10:28is simpler than you might think.
10:30and we're going to refine and create uh
10:33uh we're going to continue to refine
10:35this video that we're going to be
10:36working on these clips until we're
10:38really happy with the output. And then
10:39we'll take all of those edits that we
10:41worked on and ask our AI tool to wrap
10:43that up in a skill. Basically describing
10:45the steps that we took so that in the
10:47future the next clip that we want to
10:48make, we don't need to go through that
10:49whole process again. We're doing that
10:51work one time, wrapping it up in a
10:53skill, and then the next time we want to
10:55create clips, we basically push a button
10:56or invoke this skill and it will follow
10:58the same instructions that we've
10:59refined.
11:01Um, and this will work uh best for
11:05certain types of videos, certain types
11:06of workflows. If you're a creative
11:08agency and you create, you know,
11:10Sundance films and very bespoke one-off
11:12high craft works of art, this is not
11:14probably the tool for you. There are
11:16some um inherent limitations in the
11:18sorts of edits that you can do when not
11:20looking at dscript or or doing it from
11:21afar. But this sort of workflow works
11:24ideally when you have a series of
11:27consistent, highly structured videos
11:29that you uh would want to make or and
11:31and maybe you'd love to have these
11:33videos, but you've never quite had the
11:34time to make them. This is the perfect
11:36scenario where you'd want to invest in a
11:38little bit of time in like the machine
11:39to make the video once and then the
11:42button to push to make more and more
11:44instances over and over again. Some use
11:46cases top of mind for me in thinking
11:48about this are like every time a sales
11:50engineer has a call, imagine if you
11:51could take that recording, cut out all
11:54the cruff and small talk and then make a
11:56a sizzle reel of demos that were shown
11:57so that you could share it with your
11:59customer. or taking the first pass of a
12:01webinar rough cut, cutting out all the
12:03ums and o's like I'm saying right now
12:04from this webinar. Imagine we could
12:06automate that and have that
12:07automatically happen in a process or
12:09translating a finished video from one
12:12language into n languages. Uh turning a
12:15blog into a video. Social clips is what
12:17we'll talk about today. Teddy, are there
12:18any like favorite use cases you have
12:19that uh are top of mind for you for this
12:21sort of thing?
12:22>> You know what I really love is
12:24especially these days underlord is so
12:27smart with its media understanding. So,
12:29I love using the MCP to basically take a
12:31bunch of footage and on some recurring
12:33cadence. Look at, okay, here's all the
12:35footage I've dropped into Google Drive
12:38or into my Dropbox folder in the past
12:40week, you this is the vibe of the video
12:43I want to make. Go make me a rough cut
12:45and then I'll re give me some options
12:48and I say, "Oh, okay. I like number two
12:50on the menu. Let's start there." And
12:52that's when I actually jump in and start
12:54editing. And the reason why the MCP
12:57makes this more powerful than just using
12:59Dscript in app is because the MCP
13:02enables me to pull from those other
13:03places like Google Drive, like Dropbox
13:06seamlessly. And also for me to use
13:08Claude's brain and Claude's knowledge to
13:11chat and brainstorm and vibe a little
13:13bit better so that what ends up going
13:15into Dcript already is kind of like the
13:19the like cream at the top of the can of
13:21coconut milk compared to just like the
13:23watery stuff at the bottom, you know?
13:25>> Yeah. Always got to be hunting for that
13:26cream.
13:29>> I saw a question asked and answered in
13:31the comments up on screen. MCP stands
13:32for model context protocol. Basically,
13:34think of it as a way that a um an
13:38interface that lets AI agents talk with
13:41software or your tools. It's a
13:43standardized protocol that a lot of
13:45applications have built to let their
13:47tools work with AI agents.
13:50>> Yeah. And then also related to that,
13:52we're I see a lot of questions in the
13:54chat around how much does this cost
13:56anyway, which of course is a very valid
13:58question anytime we're using AI tools.
14:01So, it's kind of split between the
14:04tokens you'll spend in whatever your LLM
14:06of choice is and Underlord AI credits.
14:10So, anything you do in Claude or Chat
14:12GBT will cost Claude tokens. But then
14:15when Claude sends something off to
14:18Underlord and stuff happens that
14:20Underlord does, then that'll cost you AI
14:23credits in Dcript. But that's all part
14:25of your normal Dscript plan. We don't
14:28charge the MCP as a separate billing
14:30plan or you don't need to buy anything
14:32new or different or special to use the
14:34MCP. You just connect and as long as
14:37you've got AI credits, you can get
14:39going. Yeah, just to add to what Teddy's
14:41saying, if like tokens, credits, all
14:43those terms don't even make sense to you
14:44and you're not even sure what that
14:45means. With every Dcript plan, you get
14:49a certain amount of spendable units that
14:52let you use Dscript. with any dscript
14:55plan, even free plans, you are able to
14:57access the MCP and use dcripts. So, you
15:00don't need to have a paid plan to
15:01connect things. Uh, but your ability to
15:04import media or use different editing
15:06features will spend those consumables,
15:08those media minutes, those AI credits.
15:10So, depending on how much you want to
15:12produce, uh, the plan that you have will
15:14potentially be a gate to how much you
15:15can get done, but as you change your
15:17plan, then you'll like increase your
15:18limits and what whatnot. And like the
15:20MCP also lets you facilitate stuff with
15:24Dcript, but also you're still using your
15:25AI tool of choice. Like this has our MCB
15:28doesn't have any bearing on like
15:29whatever credit allocation you might
15:30have in Claude or Chad GBD say you just
15:33be using that to talk to Dcript like you
15:35might be using it to talk to other
15:36things.
15:37>> Yeah, exactly.
15:40>> Okay. Well, let's jump into our
15:41scenario. So, uh imagine this. There are
15:43just never enough clips. Our CEO, this
15:45is a true story. Um Laura does a ton of
15:47interviews on podcasts. She has great
15:49conversations and it generates a ton of
15:51really thoughtful and valuable material.
15:53And we've always wanted to but never
15:56been able to really meaningfully
15:57repurpose all of that content. And we've
15:59also been wanting to post a bunch more
16:00on our social uh channels to generate
16:03thought leadership content and drive
16:04brand awareness. This is like a true
16:05story from Dcript real life here. So,
16:08the opportunity that we're going to
16:09tackle today is how can we come up with
16:11a standardized process to create uh
16:14simple structured clips from these
16:17interviews, from these podcasts with a
16:18repeatable look and feel that feels
16:20brand aligned um and then make that into
16:23a process that we can quickly or
16:25effortlessly invoke in the future. So,
16:27every interview she does, we push a
16:28button, blink, clips fall out of that
16:30and then we can use them on our social
16:32channels. In the next episode, not
16:34today, but next week, what we'll do is
16:36we'll take the thing we build today and
16:38then create an automation so that you
16:39can sort of set and forget this thing
16:41and clips will continuously flow out of
16:43new recordings as they come in. But
16:44today, we'll be building the machinery
16:46to actually create the clips. I wasn't
16:48sure if I saw a question come up on
16:49screen. Yes, this session. Okay, cool.
16:51Yeah, just so that you know, this will
16:52be recorded. So, don't feel frantically
16:54like you need to take screenshots of
16:56every screen and take furious notes.
16:58You'll be able to rewatch this later.
17:02Okay. Uh, we that was the 30,000 foot
17:05view scenario or 10,000 meter scenario
17:07for my metric friends. The 30,000 uh,
17:10I'm going to not stop. I'm going to stop
17:11saying units. Here's the here's the
17:13rough uh, workflow we're going to follow
17:15today. Uh, Teddy's going to kick us off
17:17by showing us how to connect Dcript to
17:20Claude and Chad GPT, probably among the
17:22more common AI tools. And then I'm going
17:25to pick it up and talk us through how we
17:27create one finished in this case clip.
17:30We'll create one finished video. That's
17:31going to be the starting point where as
17:33we refine and get the video looking
17:35better, looking more like what we want.
17:37Um, that process of refinement will help
17:39us capture the instructions we use to
17:41make that video really nice uh with
17:43promptbased edits. We'll talk more on
17:44that in a sec. Then we'll be able to
17:46test that skill and then turn it into
17:48something that we can trigger later. Uh,
17:51and if there aren't any other questions,
17:53Teddy, do you want to take us away to
17:55uh, showing us how to connect Dcript?
17:57>> Let's take it away. get the show on the
17:59road. Okay, so I'm going to start with
18:01Claude first and then chat GPT. He no
18:05particular ordering or preference.
18:07That's just how I had my tab set up. Um,
18:09so from Claude and note with Claude,
18:12these same steps work in the Claude
18:14desktop app, too. I'm just on web
18:15because that's what I like to use. So
18:18you're going to click here on the
18:20customize these icon and you should see
18:24in the sidebar connectors. So this is
18:27where you can tell claude what apps to
18:29talk to. And for example, right now you
18:32can see I have linear, my issue tracking
18:34system we use here at Dcript. I have
18:36notion connected. I have Google
18:38calendar. But if I want to add some more
18:41stuff, I can click add up here and I
18:43have an option for browse connectors and
18:45add custom connector. Right now we're
18:48going to go to browse connectors.
18:51And from here, you'll see there's this
18:52whole slate of other apps that Enthropic
18:55has approved because they work well with
18:57Claude and will be seamlessly integrated
19:00once you just plug them in. And so we
19:03can search Dcript from here.
19:06Okay. And click there Dcript.
19:09And I'm going to click connect.
19:14And so what you'll see is Claude will
19:17kick me to this window to connect to
19:19Dcript. This is basically telling in my
19:23Dcript account, yes, it's okay to accept
19:25data from Claude and to connect with
19:27Claude and work with it. So, I'm going
19:29to click allow.
19:32And then I should be redirected back to
19:34Claude.
19:36And the way that you can test how does
19:39this actually work is I can say, make me
19:42a little video in dcript. Make me a one
19:46minute
19:48dcript video.
19:51And what I've got on the brain is I made
19:5310 pounds of potato salad last weekend
19:56for the holiday weekend. And as I was
19:58thinking about, hm, I've got potatoes,
20:01which are carbs, and yogurt, which is a
20:03dairy. Is that a cereal?
20:13Uh, let's just take a moment while all
20:14of our brains break considering such a
20:17philosophically horrifying question.
20:29So, I'm just going to give it a little
20:31prompt right now for the purposes of
20:32making sure it's connected. You don't
20:34even have to do anything complex. And
20:36there was a question in the chat earlier
20:38too about how do I make sure this
20:39doesn't eat up a bunch of credits. You
20:41can basically do that by choosing which
20:43model you're using. And just like how
20:45you can choose your Claude model, I
20:47could also if I wanted to include
20:50something here telling which Underlord
20:52model to use too. So if I wanted
20:54Underlord to like go to the max and say
20:56Underlord use Fable 5, I could do that
20:58in my prompt. But I'm just going to use
21:00whatever the default is because I'm just
21:02trying to do a quick little connection
21:04test.
21:07And so as we see Claude is
21:09contemplating,
21:13>> we had a question come up. I just want
21:14to speak to Teddy. Um if someone goes to
21:16the connectors view and doesn't see um
21:21Dcript
21:23do you have any ideas on what they
21:24should do? I mean of course if you've
21:26got any questions about Dcript in
21:27general blanket statement you can always
21:28reach out to our support team. There are
21:30a ton of very talented and thoughtful
21:32people who know a bunch about descript
21:33and can help you sort things out. But
21:35are there any other scenarios, Teddy,
21:36that you could imagine where they
21:37wouldn't be able to see the connector in
21:38the marketplace?
21:40>> I think one thing to check would be that
21:43comes right off the top of my mind is
21:45are you on an enterprise or personal
21:47cloud plan? Because if you're on an
21:49enterprise plan or like a team plan,
21:51your team admin can might mo might limit
21:55what connections you can make
21:56reasonably. So, you know, because they
21:58want to make sure we're using the right
22:00things things that and so I would reach
22:02out to your enterprise admin about that
22:05as a first step. If you're on a personal
22:06plan though and you're still not seeing
22:08the connector, I would definitely reach
22:10out to our support team.
22:12>> Yeah, that's a great point. Um, yeah,
22:14>> and you know, if you're using Dscript
22:16for work or you want to use Dcript for
22:17work, uh, and for whatever reason your
22:19IT admin's locked out any c any like
22:22connectors, custom or otherwise, uh,
22:24then it could be a good an interesting
22:25conversation to have, especially if
22:26you're already using Dcript anyways,
22:28like, hey, this tool that we use for
22:29video,
22:30>> but making video has gotten even easier.
22:32We can integrate it into our AI
22:33workflows, could we like uh, would you
22:35mind like could we explore adding this
22:37connector as a sanctioned connector in
22:38our enterprise workflow?
22:40>> Yeah, exactly. And so back to Claude. So
22:43now we can see that Claude actually has
22:46contacted Dcript. So that's how we know
22:48the connection is successful. So I'm
22:50just going to click always allow and let
22:52that run for now. Um but we've basically
22:54successfully proved that Claude is
22:56connected to Dscript can talk to Dcript.
22:59And if I went to Dcript right now, I
23:02should probably be able to see the
23:04project that Claude just created over
23:06here. Is potato salad a cereal?
23:10I want to um there's a question that I
23:12saw come up and I want to demystify some
23:14of the magic that Teddy just showed.
23:15Okay. So like you just asked a question
23:19to make a video with Dcript.
23:21>> Yeah.
23:22>> How is that happening? Right. Like
23:24>> there's no media you're uploading. You
23:26are sending effectively a
23:29>> vibesbased sentence on what you want and
23:32he scripts doing it. Um the I guess the
23:34punch line here is underlord is behind
23:36all this. Do you want to talk a little
23:38bit, Teddy, about like who Underlord is,
23:41what it does, and why it is the like the
23:44robot behind the curtain?
23:45>> Absolutely. So, Underlord, as you said,
23:48is our robot behind the curtain who you
23:50can talk to in chat in Dcript anytime
23:53here. And basically if you imagine that
23:55Claude and ChatPT are LLMs with broad
23:59generalized knowledge of anything,
24:02Underlord is like like our personalized
24:05AI editor who knows a lot of really good
24:08stuff specific to v video editing. So
24:11Underlord knows what makes a good video.
24:14It knows what when you should cut audio
24:16and how you should do it. When is a
24:18reasonable time to apply auto gain
24:21control? When might you want to apply
24:22studio sound? Because underlord's one
24:25job instead of knowing everything in all
24:28of humanity is to know everything about
24:30video and how to make the best video for
24:34you that it can. And so the magic here
24:37when we're in Claude and I say, "Hey
24:40Claude, how make this video?" Claude
24:42does a little bit of work first thinking
24:44about, "Okay, what do I want to tell
24:45Underlord to make?" But then the
24:48importing the stock media, the um
24:52writing, actually putting the script
24:54into the script pane, dividing it into
24:57scenes, you know, if I were to drop some
24:59audio or video tracks in here, picking
25:02what video should go where, that's all
25:04done by Underlord because Underlord has
25:06the specialized training and knowledge
25:08to know what what would make a good
25:10video and how what would make something
25:12that you would actually click publish.
25:16I also feel like it's safe to assume
25:17there are probably people out there who
25:18like are really AI savvy. They've used a
25:20bunch of different agents. They're
25:21familiar with agents built in tools that
25:23like know more about their context and
25:24their domain. And one thing I wanted to
25:26talk about mechanically a little bit
25:27like how the API and MCP from Dcript are
25:30set up. You can think of uh there being
25:33three primary tools.
25:36wonderful.
25:43>> Oh, Sam, your audio is cutting out a
25:45little bit.
25:45>> Oh, no. Is it better now? Bing.
25:47>> Yeah, it's better now.
25:47>> Okay, cool. Sorry.
25:48>> Oh, that's great.
25:49>> One of the tools you can use with it
25:51with the MCP is to import media, upload
25:53a recording, upload from a URL.
25:56>> Um, the second tool that you can do is
25:58publish. If you've got a project already
26:00edited and you want to generate a share
26:01page or a download link, um you can use
26:04the MCPN API to do that. The third one
26:06and the most interesting one is the
26:08agent tool, the editing with prompt
26:10tool. We don't have a specific button,
26:13an an MCP button, a clot command to
26:15apply studio sound and to trim this and
26:18to trim that. Instead, we push all of
26:20our edits through this agent API or this
26:22like promptbased project editing. Mh.
26:24>> And that might seem limiting, but in
26:26fact, it's quite freeing. When all of
26:28the edits that you apply are through the
26:29prompt, we support one avenue for making
26:32edits, and that's basically loose
26:34instructions to Underlord, who can then
26:35execute those edits on the back end. And
26:37the power of that is that Underlord,
26:39like Teddy mentioned, our AI co-editor
26:41that knows everything about video and
26:43everything about Dscript, can take any
26:45and all of your instructions and then
26:46apply them as loose, as specific, as
26:49grand or as finite as they are to edit
26:53your video. turn and that basically lets
26:55the sky be the limit in terms of the
26:57nature of editing commands you send in
26:59because Underlord can just handle it for
27:01you. And absolutely like anything with
27:03AI with anything prompt based like it
27:06takes it takes time to refine the prompt
27:08and if you put a bad or you know um a
27:11lame prompt in that's like not well
27:12refined, you're going to get output
27:13that's reflective of that. But at the
27:15same time, um, it's easy to refine and
27:17experiment and explore. And once you
27:19dial in a prompt, and there's some other
27:22tools in Dcript that you can lean on for
27:24consistency, like layout packs, brand
27:27studio, style, colors that take the
27:29variability out of like underlord
27:31understanding exactly what you mean.
27:32You've got enough tools to make a a wide
27:34swath of edits with enough consistency
27:37and controls to keep your sort of
27:38creation solution space in a finite
27:40place. Um, I didn't mean to like go on a
27:43tangent there, but I wanted I did want
27:44to touch on like import, edit, and
27:46publish being the three main paint
27:47brushes that you have.
27:48>> Yes. And actually, relatedly, while Sam
27:51was talking, you might have noticed this
27:52video is actively changing as we're
27:55watching. And like, hands hands up here,
27:58I'm not doing anything to do that. That
28:00is all Underlord deciding. Like, you
28:02might have noticed this B-roll it put in
28:04first was landscape. And Underlord said,
28:07"Huh, that doesn't make sense because we
28:08have a portrait shaped video right now.
28:11So, I'm gonna reshape it for portrait."
28:13And I didn't have to do anything because
28:16Underlord has the knowledge to do its
28:19first draft, realize, hm, okay, these
28:20are the things that I need to get back
28:22into shape. So, that what it serves to
28:25you at the end is pretty pretty darn
28:28good, if I do say so myself. But let's
28:32also I don't want to forget about chat
28:34GPT because I know we have a lot of chat
28:35GPT users as well. So let's show how to
28:38connect that real quick. It's pretty
28:39similar to the cloudflow but just looks
28:42a little bit different. So here logged
28:44into chat GPT you have instead this
28:47plugins button. You can click over here
28:50on the sidebar
28:52and from here
28:55I should be able to search Dcript as a
28:57plugin. Tada. And there it is.
29:02And I'll just click install plugin.
29:05And similarly with Claude, I'll need to
29:07sign in with Dscript. Basically telling
29:09Dcript it's okay to work with ChatGpt.
29:11Don't this is something that I'm
29:13allowing and I've approved. And it
29:16should take me back to chat GPT.
29:22Okay. And now ChatGpt will tell me I can
29:24try this in chat.
29:28And it actually is, this is fun, is that
29:30chatp already also will give me a basic
29:34prompt based on what people like to use
29:36dcript for. So, edit this recording for
29:39clarity, clean up my audio using some of
29:41our classic tools like studio sound or
29:43edit for clarity, removing filler words
29:46and all that good stuff. And just
29:49because I know we've gotten some
29:51questions about this and I'll just show
29:54for example if I took some video that I
29:58have.
30:01So here's for example a video I took at
30:04the aquarium of some jellyfish. I can
30:06say upload this into dscript.
30:10Add soothing music and captions
30:15about jellyfish and how great they are.
30:29Okay. And you can see chatp has already
30:31contacted dcript. And so it's going to
30:34ask me for permission. I'm going to
30:35grant it permission. But Chat GPD knows
30:37who Dcript is, knows how to knows
30:40Dcript's phone number, knows how to set
30:43up a calendar invite for Dscript, etc.,
30:45etc.
30:47And now we just let the let the bots do
30:49their thing and chug away.
30:52>> And the workflow from here basically
30:54seems quite similar, right? Like you can
30:55add prompts, you can upload media, you
30:57can edit, you can publish everything.
30:58the MCP interface with Claude and
31:01ChatgBT. There are some nuances that
31:02differentiate them, but for all intents
31:03and purposes, like this is the same.
31:06We're at the same point with both. Um,
31:08and now in the rest of this demo, we're
31:09going to jump into clip creation now.
31:11And, um, we're going to do a lot of
31:13kitchen swaps because I think what
31:14you'll realize here is like, you know,
31:16these AI tools take time. Uh, also all
31:19of this executing work with Dcript on
31:21the back end isn't instantaneous. It all
31:23takes time. But kind of the beauty of
31:25being able to trigger a lot of these
31:26actions in your AI tools is you can
31:28trigger them and then move on to some
31:30literally do anything else uh and come
31:32back to them later. Get basically giving
31:34you background processing power with
31:37things like uploading media waiting for
31:38it to transcribe and when we get into
31:40more refined edits actually doing the
31:42creation process all sort of on a
31:44background thread while you're doing
31:46something else.
31:47>> Yes. And then I just want to a answer a
31:49few last questions I saw fly by in the
31:51chat real quick. One question. Can I use
31:53the MCP on things I already have
31:55uploaded in Dcript? Absolutely. You just
31:58connect Dcript and Underlord and Claude
32:00have got your back. They know what's in
32:03Dcript. You can tell them like, "Hey, I
32:05have this project from last week. I
32:07don't even remember the name, but I was
32:08working on it with these people and I
32:10think this is what it was about." And
32:12they will figure it out for you. Um, and
32:15then you can even, I believe, pull in
32:17things from your media library as well,
32:19referencing it as you create your
32:21videos. So if you have stock footage or
32:24for example an intro or outro you
32:25regularly use that's in your media
32:27library you can reference that too. You
32:29don't need to upload everything all a
32:31cart you can also if you have stuff
32:33waiting for you in dcript just use that.
32:36>> Yeah one of the my favorite ways to edit
32:38existing projects. I like to use dscript
32:40in the browser. So I'll copy the URL of
32:42a project when I'm in a project paste it
32:45into claude and say hey I want to add
32:47captions to this project. plot or G chat
32:49GPT. They're smart enough to strip the
32:51project ID from the URL and you're off
32:53to the races. As Teddy mentioned, you
32:55can also search projects. And maybe one
32:57of the best ways to figure out all that
32:58you can do with the Dcript MCP is ask
33:00your AI tool. Say, "Hey, list for me all
33:02the tools this MCP has. You can extract
33:05transcripts. You can list all of your
33:07projects. You can look for active MCP."
33:09You can do a lot. And uh and your AI
33:11tool knows about all of those jobs.
33:13>> Yes.
33:15>> Okay. Okay. So, with that, I think let's
33:16jump into the actual uh Oh, I think
33:18someone wants to see.
33:20>> Sure. Let's take a look at these
33:21jellyfish, too.
33:22>> Let's take a look at the jellyfish. And
33:24this is one thing that's nice, too, is
33:26from Claude and Chatt. You can also open
33:28the project in dcript right there, which
33:31I really enjoy. So, let's see. I think
33:34Underlord looks like it's still chugging
33:37away at this one, but it looks like my
33:39video has been uploaded. Let's see. Do I
33:42have to approve something?
33:44Okay. Yeah. So, this one's still
33:45uploading, but when this finishes
33:48upload, by the end of this webinar, I
33:49will publish and share the video just as
33:52an example of what our MCP can do.
33:54>> And if it's helpful, maybe I can start
33:56my demo with a finished product. You
33:57know, like those cooking shows where uh
34:00you see like the finished roast chicken
34:02with the beautifully glistening carrots
34:03and vegetables.
34:04>> Exactly. Martha Stewart is like, "This
34:06is what we're going to make today. Now,
34:08on to chopping our carrots."
34:10>> Yeah, maybe we'll start that. Okay. So,
34:11I'll switch the scene and uh Okay.
34:13Again, usually I wouldn't do this. Uh I
34:15would show like the the process, but um
34:18I also see there's a lot of interest
34:20about podcasts and like podcast content.
34:23Uh I am again the demo I'm hoping to
34:26show you is I want to make social clips
34:28from podcasts and webinar interviews.
34:31I we're going to work together to refine
34:33a prompt, a process to make those clips
34:36from these raw recordings. And I'll show
34:38you this is a finished product.
34:39Actually, let me Yeah, you won't hear
34:41the audio, so there's not echo on my
34:43end, but I'll just show you what the
34:44finished product looks like so you can
34:46get a sense of
34:52this, Laura.
34:58>> Sam, your mic's cutting out a little bit
35:00again.
35:00>> Oh, no.
35:01>> Oh, now it's back.
35:02>> Okay, cool.
35:03>> Yeah.
35:04>> Um, we can share these share pages later
35:06if you want to watch these videos.
35:07There's nothing special about them
35:09except they're, you know, we've crafted
35:10them to be the clips that we want. This
35:12may not be the clip style that you want,
35:14but that is immaterial. What matters is
35:16we're going to go through the creation
35:18process of building a prompt that makes
35:20these sorts of clips from our raw
35:22recordings. And then you can follow
35:24similar steps on your end to recreate
35:25the sorts of uh uh clips or whatever
35:27sort of videos that you'd like. You'll
35:29notice there's B-roll, there's zooming
35:31in, there's captions, and then uh we've
35:33even got like Dcript's logo from our
35:35brand studio interle here. And the hope
35:39is that um this is the sort of thing uh
35:42even if you don't expect, you know, AI
35:43to be perfect at making this like 100%
35:46it's done. You don't need to touch it.
35:48Imagine if these clips were generated
35:49for me without needing to do anything at
35:5180% or 90% of the way there. And then
35:53instead of spending my time creating one
35:54from scratch, I could spend my time
35:57quickly refining and reviewing 10 of
35:59them and getting them all ready to to
36:01postable state. That's the workflow that
36:02we'll go through today. Okay, so that's
36:04the cook chicken. Sorry there was no
36:05audio. Um,
36:08so the whole process I'm going to show
36:09you here, I'm going to do a lot of, like
36:10I said, kitchen swaps where I'll show
36:12you the start of a prompt and claude and
36:13then we'll I'm back back out to like
36:15talk about the process of refinement and
36:17Tenny's going to keep me honest if there
36:18are questions uh on the thread. This is
36:20a recording from a an interview our CEO
36:23did. And um the way to kick this process
36:26off is with Claude, the same thing.
36:28Basically, after Lara is on a podcast,
36:30we'll often get the recording of the
36:31podcast from the host as a common
36:33courtesy. So, I basically uploaded this
36:35media into Claude. And this was the
36:36prompt that I gave it. Um I want to
36:38create one short 60-cond clip from this
36:40recording of Laura. Format it in aspect
36:42ratio one. Find the moments where she's
36:44making provocative statements or
36:45assertions, the most daring and
36:46interesting points of the session. Keep
36:48each clip to Lara's voice. I added this
36:50little up this like suffix here to not
36:53using any existing skills because I'm
36:54also working on skills for this webinar.
36:56So, I wanted to like do it from scratch.
36:58You won't have to add that. Um, and
37:00you're off to the races. Uh, something
37:03you should know about Dcript is that
37:05there are a ton of tools. I'll just open
37:07this up here. A ton of tools we already
37:09have for content repurposing. If you
37:11open the app, you'll see things like
37:13create clips, create highlight reels,
37:15find highlights. are already pre-anned
37:18features in the app. And when you ask
37:21Claude to make these sorts of clips or
37:24or when you reference features similar
37:26to them, Underlord and Claude are smart
37:28enough to try and guide you towards
37:29using some of these tools inherently.
37:32Um, so the way that this process looks,
37:34I'll just show you how this thing
37:35unfolded. And it's kind of boring to
37:36just like look at someone's Claude
37:37thing. So we'll spend a lot of time in
37:39Dcript itself, but remember all of this
37:41could be happening um just from Claude
37:43without having the Dcript app open. It
37:45uploaded the media that I dragged and
37:47dropped into Claude. Um, it transcribed
37:50it and then it basically went and found
37:52the clip that I was looking for. This is
37:54the result of that first prompt. You can
37:56see it's a one to one by one aspect
37:58ratio. And then the clip talks about um
38:02what is it? Oh yeah. Uh, ironically
38:04oneshotting um AI creation of videos uh
38:08being the sort of thing that makes
38:09derived content. The irony is uh is
38:11great. So this is like okay it's it's a
38:13very simple clip. There's no captions.
38:15It's just a talking head. Um, the
38:18process that we're going to go through
38:20here is basically using Claude to riff
38:22and refine, giving it feedback,
38:24instructions, requests on how we'd want
38:26to improve this video. Like one thing
38:27you might have noticed from the finished
38:28product is, I'd love to add to have
38:31captions be shown. So something I might
38:33ask Claude here is like uh let's
38:36captions and also use
38:48now I'm just like continuing this chat
38:49in the same thread. You can be as
38:51specific or as generic as you want.
38:54Generally as you know uh the more
38:56specific that you are with your prompts
38:58the more specific of output that you'll
38:59get. Um, but every time we we uh add
39:03more feedback or like uh watch the
39:05video, I think like this is a recurring
39:07pattern that you'll see a lot in this
39:08like we are making one decent clip once
39:11so that we can take all these
39:12instructions later and bottle them. But
39:14a really common workflow here is I'll
39:16watch this video uh the first prompt
39:19that I made. Maybe I like the clip that
39:21was picked. Maybe I think it cut off
39:23part of the um the overall point that
39:25Laura was making. Maybe there are filler
39:27words that I wish were removed. Um, the
39:30process that we repeat over and over
39:31again is watching the result of what
39:34Underlord did. Oh, captions already
39:35coming in. Watching the result of what
39:37was edited, taking notes on things we
39:40wish were different or improvements we'd
39:42like to make, resubmitting them, and
39:44then repeat, repeat, repeat. This is a
39:46simple, albeit effective process to
39:48basically kind of fiveedit your project
39:51in a direction until it passes.
39:52>> Oh, Sam, your audio's got a blip again.
39:56>> Oh, man.
39:57>> Oh, you're backing up. It seems like
39:58every time I I'm like, "Hey, Sam." Your
40:01audio immediately comes back without
40:02interference.
40:04>> Okay, maybe. Okay. Um, is it good now?
40:06You can hear me?
40:06>> Yeah, it's good. We can.
40:08>> Um, I'll just restate that last point. a
40:10really effective way to edit a project
40:13in this sort of fashion using the MCP or
40:15using Dscript, but the MCP is to um
40:19suggest edits like offer edits to
40:21Underlord, wait for those edits to
40:22finish, watch the resulting video, take
40:25notes on what you'd improve or or
40:26changes you want to make, and then
40:27resubmit, and then over and over again,
40:29watch, critique, watch, critique. And
40:33that might seem like indirect, right?
40:36you're like, why wouldn't I just edit
40:38with Underlord directly in Dcript or why
40:40would I like use the MCP? Well, what's
40:43what's nice about using the MCP is that
40:44you can easily background task um edits
40:48and commands if you're doing other work.
40:50But what's more important and what we'll
40:51do in just a moment is all of the edits
40:53cumulatively that we make like we just
40:54added captions. Um I think music was
40:57added. Yeah, you can see background
41:00music was added. We'll be able to take
41:02all of those edits afterwards in Claude
41:04and generate a skill that can be
41:06recalled from Claude and that will help
41:08us trigger this over and over again.
41:10>> Yes.
41:10>> Let's see.
41:11>> And also to answer one of the questions
41:13that flew by on screen around how do I
41:15overcome the file size limit if my file
41:19is a little too big for a clotted chat
41:21GBT. That's also part of the beauty of
41:23using the MCP actually is because you
41:25can have it connected to Google Drive or
41:27Dropbox which have much bigger file size
41:29limits or if you just have the media
41:32already in Dcript as well, the MCP can
41:35work with that. So if you're if you're
41:38encountering that upload error of Cloud
41:40saying that this file is too big, never
41:43fear. There are many many paths around
41:45that.
41:47>> Yeah. Yeah. And like even if your file
41:48is hosted in some cloud service like
41:50Teddy mentioned um the MCP and API can
41:52import files from a URL as well. So it's
41:54not like you need to always upload the
41:56file like dragging and dropping it onto
41:57cloud wherever the file the URL is
41:59totally a reasonable path up.
42:01>> Yes. Yeah. That was just like the baby
42:03bones demo.
42:06So, uh, I don't want to like chicken
42:08swap too many times here, but as you can
42:11see, like a common workflow I like to
42:14follow is having Clawude open in one
42:16window and Dcript in another. I'll watch
42:19each video's uh, edits after I've
42:22prompted them, take notes or critique
42:24them. I often like to just speak using
42:25like a dictation tool into Claude and
42:28then resubmit them and refine, refine,
42:30refine until I'm happy with the output.
42:32And something that's really important
42:33and we'll talk about in a minute. You
42:35don't want to rush this step. Our goal
42:38is to create a repeatable process that
42:40makes good clips out of every future
42:43webinar. And we're just making this
42:44process manually once. It could be
42:47tempting to just say like, okay, this is
42:48fine enough. Let me like get to the
42:50automation part and just like see how it
42:51works. But when you if you any process
42:54that you are going to build to replicate
42:56something, if what you're replicating is
42:57mediocre, then you're going to be
42:59replicating mediocrity. you're going to
43:00be really unhappy and always think like,
43:02"Oh man, I should have like really fixed
43:03the formatting and and achieved a better
43:05brand align style or something like
43:07that." So, you don't want to rush this
43:09process. Uh, I have a couple different
43:11videos that show like different stages
43:13of this project where I added logos as
43:15an outro scene. I took like a single
43:18oneshot quote that describes the arc of
43:21the clip and put that as an intro scene
43:23all the way up till the the finished
43:25revision which again all of this was
43:27just prompted with Claude asking more
43:29and more edits. This one if I like show
43:31you these different scenes down here
43:33you'll see there's an intro sequence
43:35with like the title B-roll has been
43:38added at selective moments when Laura is
43:40saying something that's illustrative or
43:42has like some sort of visual component
43:43that could depict it. there's a caption
43:46style that feels legible and uh and then
43:48I picked some music that I thought was
43:50nice and the result was this set of
43:53clips. You can see we've got, you know,
43:54a bunch of different ones, all of which
43:56are on different topics, but each of
43:58which, let's see, oh, these are the
44:01actual share pages. Each of which have
44:02this same style. It's as if like the
44:05same machine made each of these clips
44:07time and time again.
44:09Um there I don't know if there are any
44:10questions but this would be a great time
44:11to ask specific questions around the
44:13process of refining the prompt uh
44:16through Claude because if there are any
44:17I'd love to touch them
44:25and if there aren't um one thing I'll
44:27step back on here is I wanted to jump
44:30into this project. So, this was the
44:32actual claw session I used to create
44:34those first few clips that I just showed
44:36you the finished product of. And you can
44:39see there's a lot of back and forth,
44:41right? Uh uh I asked for music, I asked
44:44for captions. Um uh we would go, it
44:47would like make an edit that I wouldn't
44:48like and I would ask for it to refine it
44:50back and forth. Um, when you get your
44:53video to a state where you're happy with
44:56the finished product, this is the part
44:58of the process where you can say, "Okay,
44:59I want to take
45:02>> Oops, Sam, your audio cut out for a
45:03second."
45:05>> How about now?
45:06>> Welcome back.
45:08>> It might be tab focus, I think, if I
45:09leave the
45:10>> Oh, that's smart. You can uh once you're
45:13happy with the output, then the next
45:15step of this process is to take those
45:17edits and turn the instructions that got
45:20you from zero, like right from when you
45:22imported your media all the way through
45:23to your finished output into what I'll
45:26call a skill. Skill is a common term
45:27used by these um AI tools to describe a
45:30repeatable set of instructions that you
45:32write once and apply many times. And
45:35thankfully, your AI tools are really
45:37good at creating these already because I
45:39did all of my refinements and edits in
45:41this thread. I basically said something
45:43like this. Help me write a prompt that I
45:45can pass on to Oh, no. This is for the
45:47B-roll. I think the prompt that I wrote
45:49was like um basically help me take all
45:52of the edits that we've made
45:53cumulatively and write a skill so that I
45:57could recall this in the future for for
46:00future clips, for future videos. And
46:01that it basically what it will do is go
46:04through and write here's an example
46:07um a full skill that you can then take
46:10and save as a skill in your clawed
46:13environment. And I'll show you how to do
46:14that right now.
46:15>> And to answer DIY Tourism's question on
46:18screen around will we provide commonly
46:20used prompts eventually. We actually do
46:22have a public repository of commonly
46:24used skills like these skills that Sam
46:27is showing right now that you can feel
46:29free to use. you can also contribute to
46:31if you're like, "Hey, this skill really
46:32works well for me and I'd love to share
46:34it with other people as well."
46:37>> Yeah, we have a couple in there now and
46:38we would love more contributions. I
46:40think what's tricky about these sorts of
46:43prompts is like the format of this clip
46:45might be very specific to the sort of
46:47thing we would want and you might want
46:48to make different decisions. So, it's
46:50easy to share prompts, but it's hard to
46:52make a one-sizefits-all prompt that has
46:54like all the variability. Like the
46:56beauty of it is these are so easy to
46:57create. Once you've got a video that is
46:59reflective of the finished product you
47:01want, you can make them incredibly
47:02personalized. And in fact, I find the
47:04best skills are very um specific and and
47:07have like uh very, you know, specific
47:10instructions towards the end that you
47:11want because they're able to handle a
47:12lot of different inputs and get you
47:13close enough to the finished output.
47:16>> Uh there's a question on screen. Can you
47:18request specific changes, remove B-roll
47:19footage of a guy drinking water and
47:20claw, or do you have to do this in the
47:22descripti? I think you should be able to
47:24do this in You can do it, you can do it
47:27wherever you want to. If you do it in
47:29Claude, you can do it there. And trust
47:31no, it'll cost you Claude tokens then or
47:32you can do it directly in Underlord and
47:35it'll cost you descript AI credits
47:36instead. You know, it's really up to you
47:39based on which agent you feel like will
47:41better handle the request.
47:44>> Part of what makes this possible though
47:45is the media imported into Dscript is
47:48processed and Underlord basically, you
47:50know, watches the media and can
47:51understand things that are happening on
47:53screen. Uh so because that happens you
47:56are able to ask questions like find the
47:57picture of the guy drinking water and
47:59remove that or something like that. And
48:00whether uh like Claude tries to do that
48:03on its own or go straight to Underlord,
48:05those visual descriptions exist along
48:07with the media and make it possible to
48:09do something like that. So quickly what
48:11I'll do just to finish the swing here,
48:13here's the skill that was generated
48:14after we made all of our edits in
48:15Claude. We refined, we added captions,
48:17we added music. Then I asked Claude to
48:20create a skill summarizing all of the
48:22instructions we made. I'll copy this and
48:25I'll go into my Claude menu into
48:27customize.
48:29And then under skills, I'll say add. I
48:32can create one with Claude. It will like
48:34ask and prompt you or I can write skill
48:36instructions myself. I'll just write it
48:38myself. And I'll basically paste those
48:40instructions into this field. And then
48:42I'll call this like, you know, create
48:44clips, whatever, whatever. I actually
48:46already did this with this exact same
48:48prompt and this is it right here. It's
48:51exactly the output that you saw. It's
48:53got all all of these instructions in
48:54here and you can read through them. In
48:56fact, one thing I'd recommend is when
48:57you create the skill, what we'll do
48:58right now quickly is test this skill on
49:01a new recording that it's never seen
49:02before. Evaluate the output and it's
49:05likely you'll want to make some changes
49:06or it doesn't perfectly recreate the
49:08sorts of things you liked from the first
49:10video. Well, the beauty is you can go
49:12back to that thread where you created
49:13the skill in the first place just like I
49:16am here and say,"I tested this somewhere
49:18else and uh I didn't like this thing
49:20about it. Can we update the skill to
49:22make sure we always uh end on this kind
49:24of transition or add this sort of music
49:26instead?" These sorts of skills are the
49:28sort of thing you're going to want to
49:28refine. But as you refine them, what
49:30you'll find is you'll get closer and
49:31closer to the desired output. And the
49:33best way to do it is just to test it,
49:35see how the output looks, and then test
49:37again. Um, yeah, let's jump into Q&A in
49:40just one second. The last thing I want
49:41to show you is once you have that skill,
49:44using it is really simple. In Claude,
49:46you say slash create uh this is the
49:49skill we wrote and I literally just uh
49:53call that skill. I'll upload a file from
49:57uh let's see, boink boink. Here's a new
50:00interview from Laura. And I'll press go.
50:02And that's it. that is all of the
50:04instructions that we we crafted to
50:06create that skill are now being applied
50:08to a brand new recording. And this is
50:09the sort of thing in the next episode
50:11we'll use as a jumping off point to
50:13create an automation every time there's
50:14a new recording. So you don't even have
50:15to do this. A new recording lands in a
50:17place and this automation can proceed.
50:20So we covered a lot. My throat is tired.
50:23I still But maybe it's a good time to
50:26jump into Q&A.
50:31Does it cost us credits in Dcript to use
50:33the MCP? Yes, it does. It costs you
50:37credits just like using any of our AI
50:39tools costs you credits. So, you'll see
50:42that build in your usage panel in the
50:44app that for example, if you edit make
50:47an edit with an API, it should show up
50:50there that it cost me I don't know, I'm
50:52just going to say 15 credits. and it'll
50:54have a little tag next to it that says
50:56API as well so that you know what's
50:58coming from the API versus what's coming
51:00from you using it in app but you do not
51:02have to buy a specific plan or any
51:06special kind of setup to use the MCP
51:09>> also yeah using the MCP it doesn't cost
51:11extra credits because you're using the
51:13MCP simply another way to create but it
51:15is not implicitly more or less expensive
51:17the honestly the biggest factor in cost
51:19here which we didn't really talk about
51:21is the model that you use I happen to is
51:23Opus 48 just by default. Uh experiment
51:26with models, right? Cheaper models uh
51:28lesser credit usage. The same is true
51:30with claw dcript in any AI tool.
51:38>> Does the model we use in claude/ chatbt
51:41be the same when we use the MCP for
51:43descript? Um, so if I'm interpreting
51:47your question, basically if you set
51:50where Sam was showing you could choose
51:52opus or you could choose another model
51:54that chooses the model that Claude uses,
51:57but that's not necessarily the same that
51:59Underlord uses. So you could tell, for
52:02example, if you want Claw, a combo you
52:05could do is if you're like, I don't need
52:06Claude to put that much brain power into
52:08it. So you could set Claude to say like
52:10Sonnet, but then you want to give
52:13Underlord a lot of brain power. You
52:15could tell in tell Claude in your prompt
52:18that I want Underlord to use Opus. So
52:20that way you're kind of redirecting your
52:23credit cost or like the brain power of
52:26your agents to where it matters most to
52:28you.
52:29>> And if I'm not mistaken, uh we've
52:31recently expanded at least the API
52:33service if you're building custom
52:34applications so that you can be more
52:35explicit in dictating which model is
52:37used. It was always there. I think it's
52:38a bit more explicit now.
52:39>> Yes.
52:39>> And um and I think the defaults when you
52:41use the MCP mirror the defaults that you
52:43would use when you use the app, which I
52:45think like start defaults to sonnet.
52:47>> Exactly.
52:50Can you explain more about what
52:51Underlord is? Underlord is our inapp AI
52:55co-editor. It's basically imagine
52:57chatyppt or claude but if chatypept or
53:01claude was supercharged to know
53:04everything that makes an amazing video
53:06and everything that is in dcript inside
53:09and out so that versus sometimes you
53:12might make a request to clatchy pt and
53:15they go I need to research that a little
53:17bit underlord already knows that because
53:19underlord's number one job is to know
53:21about dscript and to know about what
53:23makes a good you.
53:26>> I feel like what it does more than
53:28anything is um
53:31do your bidding. It's like an eager
53:32intern who wants to like follow your
53:34instructions no matter what. Okay, I
53:36asked it to set the background to green.
53:38I don't know why it didn't do that. Um
53:42but basically, Underlord is the
53:44co-editing agent that you chat with.
53:46Yeah, like Teddy mentioned in the app.
53:48And all of the editing instructions
53:49you're giving Claude are just being
53:50passed along to Underlord who has full
53:52context of tools, the project, the
53:55scheme.
53:55>> Um, so yeah, many apps have an agent
53:59built into their tool. Underlord is
54:01descripts.
54:02>> Yes. And then we have a question. Is
54:04there a way to get Underlord to do a
54:06modified version of the shorten word gap
54:08feature that cuts pauses from middle of
54:10the gap so it doesn't risk clipping the
54:12end of the word? That's something that I
54:14would say give it a try yourself and let
54:15us know how it goes. But I will also
54:17offer we do have some
54:20just this is a little sneak preview is
54:22we do have some improvements to shorten
54:24word gaps coming hopefully your way
54:26pretty soon because I was working on
54:27those actually quite recently because we
54:29know it's a much beloved feature but
54:31also that it's really frustrating when
54:34and um the gaps get cut are not
54:36necessarily the gaps that you wanted to
54:37get cut.
54:39>> Teddy's response reminded me like oftent
54:41times uh when someone's like hey is it
54:43possible to do this with Underlord? It's
54:45like maddening but also true. Usually
54:47the best answer is like probably you
54:50should try.
54:51>> Give it a whirl. Yeah.
54:52>> Yeah. Like these agents are so darn
54:53smart. Uh more often than not they do a
54:55halfway decent job if you like ask in
54:57specific terms what you want to try. So
54:59that's a really common theme and I
55:00encourage you all to try. Um okay
55:02question. How do you save the workflow?
55:04So the workflow if you're talking about
55:05the workflow uh Susan that we just
55:08talked through making edits refining
55:10refining and then saying okay all of
55:11these edits now I want to replicate them
55:13again. Um, what I would do, and I'll
55:16just jump back to where I was here. Uh,
55:19at the very end of all of the edits in
55:21the same thread where you made them, I
55:22would say, let's turn, um, my audio
55:25might be jumpy, so sorry in advance if
55:26it is.
55:38any prompt that spells anything like
55:40this. Basically like take all of the
55:42edits that we made and
55:44>> can you focus again?
55:46>> Yeah. Is it better now?
55:47>> Yeah, that's better. Thank you.
55:48>> Take all the edits we made. Turn it into
55:49a skill so we can recall it again. Um
55:52any prompt that says basically like,
55:53"Hey, remember everything we did in this
55:55project? Skillify it for me." um will
55:58basically ask get claude to spit out a
56:00big old prompt and then you c and let's
56:02see this is what this is here. You copy
56:05that text and then in the skills menu of
56:10the customize section of Claude, you'll
56:12just say create a new skill,
56:15paste those instructions and then call
56:17it whatever you want.
56:21Something like that, create clubs,
56:23whatever. Then invoking that skill,
56:25using that skill
56:31is as simple as in chat going and saying
56:37create clips and then uploading your
56:39media.
56:43>> Oh yes.
56:44>> Will we get to see jellyfish and potato
56:46salad? Well, the people hunger for
56:48jellyfish and potato salad. So who am I
56:50but an agent of the people? Um, here we
56:54can see first off the potato salad video
56:56and this one I want to note too that I
56:58even published this from within Claude
57:01because I looked at what Claude did and
57:03I said, "Okay, that looks good. Publish
57:05it." Two words. That's all I had to do.
57:08And now I have a sharable Dscript link
57:10and I have a direct download to download
57:13the actual MP4 video file. if I wanted
57:16to upload this to my Instagram, to my
57:18LinkedIn, you know, maybe I'm not going
57:19to upload upload this to LinkedIn, but
57:21you get the gist. Um, and so I'll play
57:23it. I'm going to mute my mic though, or
57:25I'm going to mute the audio, though,
57:27because I don't want there to be like
57:28horrible echoy sound bleed. But we can
57:31watch the video play for a few seconds.
57:37And so you can see this is like a real
57:39video that I might be served on my
57:41Instagram algorithm, for example.
57:43>> God, I love food analogies. This is
57:45killing me. Why are we looking at so
57:46much food during lunch?
57:48>> Um, and then if we look at the jellyfish
57:51one, this one I'm waiting on the publish
57:53to finish from chat GPT. But we can see
57:56in dscript the actual video that was
57:58made. And you can see I've got I
58:00uploaded while Sam was talking. I
58:02uploaded a few more jellyfish videos
58:03because I realized my first one was only
58:058 seconds. So I just added some more.
58:07But you can see I've got some nice
58:09captions here like I requested about
58:11jellyfish. If I unfurl the timeline, I
58:14can also see some calming music has been
58:16added like I requested. So yeah, there
58:20you have it.
58:21>> Wow, we covered a lot. Let's quickly
58:23recap. There's some pro tips I'd love to
58:25throw up on screen, too. We connected
58:26Dcript to Claude and chat GPT. We used
58:30Claude to create a video, a clip from a
58:34raw recording. We refined that recording
58:36by giving iterative prompts, getting
58:37that clip closer to what we wanted. We
58:40asked Claude to create a skill for us,
58:41describing all the edits beforehand and
58:44making it into a and then we made a
58:45skill so that we could recall that set
58:48of edits for future clips. Um, some pro
58:51tips that made this work and like these
58:53are just things that I guess we've
58:54learned from experience. Vibing in a
58:55direction instead of trying to exactly
58:57replicate a certain type of video
58:59generally is going to like you're going
59:00to be happier with the output. AI is
59:02great at being non-deterministic. Uh, so
59:04if you want to exactly recreate
59:06something, there are tools to help you
59:07be more consistent, but it's generally
59:09just a higher, you know, a higher
59:11barrier to get there. Um, you can lean
59:13on Dcript's brand and styling tools
59:15though to be consistent. We've got tools
59:16like layout packs and brand studio.
59:18Hopefully we can put some links in the
59:20chat to help you learn more about those
59:21things, but those let you define visual
59:24layouts and colors once, and then your
59:26prompt can refer to them so that every
59:28video uses the same sorts of elements.
59:31>> Um, struggle with your art. We kind of
59:33glossed over that with our chicken swaps
59:35today, but like truly if you want to
59:37automate something and you only have
59:39like mediocre results, you you're going
59:41to just replicate mediocrity and that's
59:43probably not what you want. So struggle
59:45with making the first good version of
59:47whatever you want to make. Take your
59:48time. You'll be tempted to rush. Don't.
59:51And if you don't, dividends will be paid
59:53every time you use that skill you sweat
59:56uh into to make um uh you know high
59:58polished output. And then we don't have
59:59time to talk about this, but there is
1:00:00like AI tool hygiene to help you like
1:00:03pull in your context. Another use case
1:00:04we can explore in the future is like
1:00:06pulling stuff that Claude has access to
1:00:07that isn't from Dcript like docs or
1:00:10spreadsheets or you know call recordings
1:00:12and that sort of thing. Uh there is
1:00:14going to be another episode next week
1:00:16where we take the skill that we just
1:00:18developed together and we build uh think
1:00:21of it like baby's first automation uh a
1:00:24simple automated workflow anyone can
1:00:26build to make it so that every new
1:00:28recording that Laura our CEO uh is in
1:00:30will automatically create end clips like
1:00:32that. That's going to be our workflow.
1:00:34But you can map this to anything that
1:00:35you want. And if you'd like to learn
1:00:37more or you'd like us to focus on
1:00:38different topics, please reach out to
1:00:41us. Let us know. So this is a very
1:00:42exciting topic for us. You've also got
1:00:44some other learning opportunities on
1:00:45screen. Learncript which is a set of
1:00:47modules inside the app. There's a a
1:00:49weekly um community office hours where
1:00:52um really awesome people showcase the
1:00:54product and answer questions. There's
1:00:55weekly support sessions. And then our
1:00:57next webinar is going to be uh next
1:00:58week. Any signoff thoughts? Any closing
1:01:01jellyfish thoughts, Teddy?
1:01:03>> Uh any I think my closing thoughts are
1:01:06now after all this talk of chicken
1:01:07nuggets and potato salad and roast
1:01:09chicken, I'm ready to go eat lunch. But
1:01:11this was a great webinar. I'm so happy
1:01:13to teach people and I look forward to
1:01:15seeing you all next week.
1:01:16>> Thank you all so much. If you'd like to
1:01:18book a demo or talk with someone who can
1:01:19help you understand what a workflow with
1:01:21Dcript looks like to you, there's a QR
1:01:22code on screen. Thank you all for tuning
1:01:24in. Have a great rest of your week and
1:01:26may we all enjoy chicken nuggets.
1:01:28>> See you later.
1:01:47Why you try to send a muz
1:01:54you see token into
1:01:59I still know I'm logged into my API
1:02:04calling you. Authenticate every to
1:02:12connecting you.
1:02:14Every
1:02:19we're calling you authenticate every
1:02:28matching you
1:02:34abs
1:02:35we're calling you authentic

About this transcript

This page contains the full transcript of Make video without leaving Claude or ChatGPT by Descript, generated from the public captions YouTube serves with the video. The transcript has 11,793 words across 1,690 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.