Lemon Slice (@LemonSliceAI) suggests that generative AI will spawn a new medium. Like TV wasn’t radio with images, generative AI will create something new. The above recording shows a real-time conversation with Lemon Slice Live. Interactivity is the element that will differentiate this medium.

Their real-time video transformer model allows the creation of characters from a single image. More than a talking head, as seen in the video, the characters have expression and nuance. The amazing thing, and what Lemon Slice claims is a first, is that this is accomplished in real time.

Applications include:

• Live support & Sales
• Education
• Entertainment
• Advertising

This is still in its early stages. Currently, the latency between asking a question and receiving a response is 6 seconds. The target is under 2 seconds. One way Lemon Slice plans to improve latency and increase resolution is through the integration of purpose-built ASICs.

Check out their website and create interactive characters and have conversations with them at https://lemonslice.com/live

Read more about the technology at https://lemonslice.com/live/technical-report

See the unedited interview at this link https://lemonslice.com/replay?id=6d63a9c6-f045-4f02-8038-6c02eae57a49

Notes:

• The original recording was 16×9. It was changed to 9×16 to meet the YouTube Shorts format. Also, the first minute or so of conversation was removed to shorten it to the YouTube Shorts length. No other editing was done.
• The above images and voice are used under fair use guidelines to demonstrate how this works.
• It isn’t clear from their technical paper what the source of training data is for their characters. The above character is knowledgeable about the Dunder Mifflin work environment.
• It will be interesting to see if Lemon Slice’s work leads to new forms of compression. This is especially true if the generation is local. The use of generative AI as a form of compression was outlined in this Viodi article.