Ryan Morrison���s Post

Moshi is an impressive project with laudable aims. It needs some work before it matches the demos from OpenAI but at least it's usable by anyone today. https://lnkd.in/dnRjczbA Kyutai are the French lab behind the project and their goal is to work with the community to create a more intelligence, broader model. I will say — is is more argumentative than any AI I've used.

If you want something to pass the time while polls are open I used Websim to create a UK General Election simulator game. https://lnkd.in/dx4x8VUu

I'm seriously impressed with Gemini 1.5 Flash, particularly for transcription. It can split out by speaker, use context clues to identify names and even provide summary and analysis — and do it very quickly. https://lnkd.in/dzPMpfrd

Built myself a tool to track the election results tonight. Includes key constituencies, polling data, a seat tracker, news and historical data. A few things still to do including more live data, AI-powered analysis based on seat changes against historical data and polling, plus input from the list of seats to watch as declared. But its basically designed as a fun little project for me to use while watching live election coverage.

For a story this weekend I'm asking the question 'Can AI make me laugh'. I've already shared some of the sitcom experiments. Here is a standup routine written by Claude 3.5 Sonnet. Sorry in advance to any standup comedian but at least you know you're job is safe.

Have you ever wanted to listen to Judy Garland read Wizard of Oz, Laurence Olivia tackle Sherlock Holmes or James Dean explain how LLMs work? Now you can thanks to a new ElevenLabs Reader feature called Iconic Voices — and its amazing. https://lnkd.in/dbvWQjMq

I am really impressed with Runway Gen-3 Alpha. My initial thoughts, first hands-on for Tom's Guide: https://lnkd.in/dbPmMG9q It is not only very good at motion and creating natural movement but also renders realistic content from text. Other models work best from an image as the starting prompt. I made this short reel to showcase some clips, the music is from Suno.

I've been playing a lot with Luma AI Dream Machine and ways to ensure consistency in both style and characters. I've found starting with Midjourney images, and using the original image as a style or character reference solves many of the issues. Also the Dream Machine keynote feature lets you improve motion for vehicles by offering the start and end point as an image. The character voice is from ElevenLabs using speech-to-speech and lip-sync is Character-1 from Hedra. I'm going to continue building out this series to experiment with different tools and techniques. Next step is upscaling the Hedra videos.

A silly little project testing out a new AI speech studio from ElevenLabs. I wrote a very simple radio comedy set in an allotment and used AI to voice the parts and generate sfx. This is more a test of the studio than the voices or my lack of skills in comedy writing. What I've learned is it would be a great way for a real scriptwriter to rapidly demo a new project before pitching it to producers or commissioners. AI is bringing a new era of rapid prototyping from something the preserve of tech into the creative industries. Humans will still voice, edit, produce and make the final output — but AI will help get to that stage faster.

I've been working on a new AI video project called The Farm, about a post-apocalyptic society. The focus is on the farmer trying to eek enough food out of the all but barren soil. This is a VERY rough first run through of a scene. Still need to regenerate a few of the overlay clips and create an end shot. This has taken about 25 generations so far to get these handful shots. Voice: ElevenLabs (WF Strong (speech-to-speech) Lip-sync/Character: Hedra Video: Luma AI Dream Machine Imagery: Midjourney