Please note: Most, if not all, of the articles published at this website were completed by Chat GPT (chat.openai.com) and/or copied and possibly remixed from other websites or Feedzy or WPeMatico or RSS Aggregrator or WP RSS Aggregrator. No copyright infringement is intended. If there are any copyright issues, please contact: bicycledays@yahoo.com.
Builders and firms are more and more deploying AI brokers and chatbots inside their apps, however to this point they’ve largely been restricted to textual content. Digital avatar era firm Lemon Slice is working so as to add a video layer to these chats with a brand new diffusion mannequin that may create digital avatars from a single picture.
Known as Lemon Slice-2, the mannequin can create a digital avatar that works on prime of a information base to play any function required of the AI agent, like addressing buyer queries, serving to with homework questions, and even working as a psychological well being assist agent.
“Within the early days of GenAI, my co-founders began to mess around with completely different video fashions, and it turned apparent to us that video was going to be interactive. The compelling half about instruments like ChatGPT was that they had been interactive, and we would like video to have that layer,” co-founder Lina Colucci stated.
Lemon Slice says it is a 20-billion-parameter mannequin that may work on a single GPU to livestream movies at 20 frames per second. The corporate is making the mannequin out there by an API and an embeddable widget that corporations can combine into their websites with a single line of code. After an avatar is created, you’ll be able to change the background, styling, and look of a personality at any level.
Moreover human-like avatars, the corporate can also be specializing in with the ability to generate non-human characters to swimsuit completely different wants. The startup is utilizing ElevenLabs’ tech to generate the voices of those avatars.
Based by Lina Colucci, Sidney Primas, and Andrew Weitz in 2024, Lemon Slice is betting that utilizing its personal general-purpose diffusion mannequin (a sort of generative mannequin that learns to work backwards from noisy coaching knowledge to generate new knowledge) for making avatars will set it other than rivals.
“The prevailing avatar options I’ve seen thus far add detrimental worth to the product,” Colucci stated. “They’re creepy, and they’re stiff. They appear good for a number of seconds, and as quickly as you begin interacting with them, it feels very uncanny, and it doesn’t put you comfortable. The factor that has prevented avatars from actually taking off is that they haven’t been adequate.”
Techcrunch occasion
San Francisco | October 13-15, 2026
To fund that effort, the corporate on Tuesday stated it has raised $10.5 million in seed funding from Matrix Companions, Y Combinator, Dropbox CTO Arash Ferdowsi, Twitch CEO Emmett Shear, and The Chainsmokers.
The corporate says it has guardrails in place to forestall unauthorized face or voice cloning, and that it makes use of giant language fashions for content material moderation.
Lemon Slice wouldn’t identify the organizations utilizing its expertise, however stated the mannequin is being put to work to be used instances like schooling, language studying, e-commerce, and company coaching.
The startup faces stiff competitors from video era startups like D-ID, HeyGen, and Synthesia, in addition to different digital avatar makers Genies, Soul Machine, Praktika, and AvatarOS.
Ilya Sukhar, a accomplice at Matrix, thinks that avatars might be helpful in areas the place movies are distinguished. As an example, individuals like studying from YouTube fairly than studying lengthy blocks of textual content. He famous that Lemon Slice’s technical prowess and its personal will give it an edge over different startups.
“It’s a deeply technical workforce with a observe report of delivery ML merchandise, not simply demos and analysis. Lots of the different gamers are bespoke to specific eventualities or verticals, and Lemon Slice is taking the generalized “bitter lesson” scaling method (of information and compute) that has labored in different AI modalities,” he stated.
Y Combinator’s Jared Friedman believes that utilizing a diffusion-style mannequin permits Lemon Slice to generate any type of avatars as in comparison with another startups which can be targeted on both human-like or recreation character-like avatars.
“Lemon Slice is, I consider, the one firm taking the basic ML method that may finally overcome the uncanny valley and break the avatar Turing check. They prepare the identical kind of mannequin as Veo3 or Sora: a video diffusion transformer. As a result of it’s a general-purpose mannequin that does the entire thing end-to-end, it has no ceiling on how good it could possibly get; the others prime out beneath photorealistic. It additionally works for each human and non-human faces and requires solely a picture so as to add a brand new face,” he stated.
The startup at present has eight staff, and plans to make use of the funds to rent engineering and go-to-market employees, together with paying the compute payments to coach its fashions.
