Dumme, a startup that puts AI to practical use in video editing, is already creating demand before it goes public. The Y Combinator-backed company says it has hundreds of video creators testing its AI-powered short-form video creation product from YouTube content, with a waiting list of more than 20,000 before launch. . Dumme’s promise is to use a combination of both proprietary and existing AI models to save editing time as well as contract (human) labor, who are often entrusted with simpler tasks. Being able to do a better job than others. Video editing jobs such as cutting long-form content for publishing on short-form platforms such as YouTube Shorts, TikTok, Instagram Reel.
Founded in January 2022 and participating in startup accelerator Y Combinator’s Winter 2022 program, Dumme’s co-founder and CEO Melwain Dry initially focused on building a search engine for videos. He said he was. But about six months ago, the development team realized that reusing the same AI model of his that they had been developing to edit video clips could result in a better product. .
Joined by co-founders Will Dahlstrom (CPO) and Jordan Brannan (CTO), both of whom have backgrounds in AI, Drai said after the app went viral and a server crashed. , realized that Dumme may have fit the right product market.
“We never expected it to get a lot of attention, so we just put something out there,” Drei explains. “Then what happened is that we woke up all night and the servers were overloaded and nothing was really working. So we deleted everything and We actually created a waiting list of sorts,” he continues. “The next morning he woke up and there were probably 5,000 people there, which was interesting.”
The team then discovered that a TikTok creator had posted a short video about the product, which sent a ton of traffic to the site.
“It was really unsettling after that,” Drei says.
Pronounced “dummy,” the product was intended to simplify and speed up the work involved in video editing, which appealed to creators.
Image credit: No good
Using Dumme is as easy as the name suggests. First, when a user pastes a link to her YouTube video and clicks “generate,” it outputs a number of short videos highlighting the AI-ingested content. The company says it uses YouTube as a source rather than supporting raw video footage to outsource content moderation. So if YouTube allows it, so does Dumme.
The processing time and number of resulting clips depends on the length of the original video.
But as an example, Dry says that an hour-long video podcast takes about 20 minutes to process, and clips start to come in about five minutes later. Once complete, the creator by default he can download video clips of less than 60 seconds and upload them to platforms that support short-form content, such as YouTube shorts, as well as other platforms such as Reel and his TikTok.
Image credit: No good
Of course, how this works on the backend is much more complicated. Initially, the company says Dumme will learn as much as possible from the source video via metadata. It then transcribes the video and tries to make sense of what is being said, as well as decipher the emotions of the person speaking by looking at the frames. These results are correlated and passed to a language model that determines which parts of the video deserve clipping. It is then handed off to another model that tracks active speakers and handles trimming.
Dumme said it works with existing AI models, such as GPT-4 and a fine-tuned version of Whisper, as well as other models built in-house, such as a model that tracks the active speaker within a video frame. That’s it. One of his models has also been trained on numerous YouTube shorts to learn what a good opening hook looks like to draw the viewer in. The team is also experimenting with his Facebook Research open-source model LaViLa, which hasn’t been published yet. To better understand the context of the video.
AI work is being done at GPU cloud provider CoreWeave rather than AWS because it is more affordable, the company said.
Dumme relies on AI to process spoken words, so the technology isn’t suitable for things like long gameplay videos where people aren’t speaking. Dry said the startup is initially targeting YouTube creators, podcasters and agencies, who believe the latter are the best way to monetize their products.
Image credit: No good
Dry explains that agencies today often outsource this type of work, and the results are hit and miss.
“They are just paying cheap jurisdictional contractors to edit their own content. It takes weeks,” he says.
When asked how he feels about developing technology that actually puts people out of work, Drei was unconcerned.
“The way I think about it, in the end… I think that’s kind of what your math teacher says you’re going to do.” [be put] There’s a calculator out there,” he explains. “People will adapt.
Pricing currently being considered involves companies paying $0.40 per minute of video processing, but smaller creators will instead opt for monthly subscriptions capped at 10 hours of content per month. You can also choose (These numbers are subject to change.) The product was free to use during testing.
Early users have used Dumme for a variety of edits, including generating clips from video podcasts and publishing them into short videos, cutting other new videos and exploring the back catalog.
The product appears to compete with other AI technologies on the market, such as those from creator firm Jellysmack. Jellysmack uses AI to cut long YouTube videos, resize them, and convert them into shorter videos optimized for specific platforms. As a result, on the Kamua acquisition in 2021. Other tools that do similar work include Vidyo.ai, Detail, TubeBuddy and Wisecut. How successful or unsuccessful Dumme will be will depend on whether it outperforms its competitors in terms of quality of work and cost, but those metrics are yet to be determined.
But some investors are betting on Dumme. Prior to its launch, the startup has raised a $3.4 million seed round from Y Combinator, Caffeinated Capital, Max and Nellie Levchin (via SciFi VC), Suhail Doshi, Nico Chinot, Protocol Labs, Chris Puscasiu, and other angels.
Given the interest and sizeable waiting list, Damme says he’s aiming to hire about 500 people each week. TechCrunch readers can use the invite code TECHCRUNCH to jump into the queue until slots are exhausted.
