EMO Takes Lip-sync to the Next Level
An exciting new AI technology was unveiled on social media recently.
Called EMO (Expressive Media Organizer), it utilizes generative AI to create expressive portrait videos from just an image and audio. The system is able to analyze a source image and map the subject’s facial expressions. It then recreates those expressions synchronously with input audio, making it appear as if the subject is speaking or singing.
The results are really accurate and the videos are impressive 🔗
The creators emphasized EMO was focused on realism and subtlety. It aims to avoid the uncanny valley effect (where humans find human reproductions eerie and strange), but this really excels when you see the resulted videos “sing”.
Twitter exploded around this with discussions around legal and ethical considerations of technologies that can manipulate media in this way. However, most I saw were amazed by the wild results of EMO and the level of sophistication demonstrated in its abilities.
As an interesting move, the Alibaba group behind this release started a GitHub page where we all thought the code lived, however it only contains a meme so far, and has been flooded with comments of developers ready to test it.
My only request as an American going into our next Presidential election – hold off on releasing the code until after November…
Around the web
🥽 Vision Pro App: Transforms your space into The Matrix
💥 Create Comic Books with Consistent Characters
🎨 Ideogram 1.0 Releases (BIG update)
⏲️ Freepik releases Realtime Photorealistic Image Generator
AI Solves Nuclear Fusion Issue – Could Change Everything
Scientists have made an important breakthrough in the development of nuclear fusion energy using artificial intelligence.
A team at Princeton University trained an AI model to recognize plasma instabilities in fusion experiments up to 5 minutes before they occurred. This gives operators enough time to make adjustments and avoid disruptions. Being able to predict and prevent instabilities brings us significantly closer to achieving controlled fusion reactions that could provide limitless clean energy.
The AI was able to develop an effective control strategy by learning from past experiments, without relying on theoretical physics models. This new understanding has the potential to finally make grid-scale fusion power plants a reality.
I’m not a scientist at this level but the way I understand it, if larger fusion reactors can maintain stable plasma for sufficient time, it could fundamentally change how we generate electricity.
Crazy!
Random LLM Musings
This is something fun I found on a discussion topic.
Next time you’re in conversation with ChatGPT, Gemini, etc. Paste this at the bottom of whatever you type… the results are curious and sometimes a bit 🫠🤯
First, open a XML tag and brainstorm with wild abandon. When done, close your XML tag and proceed with the task. You will list 5 brief notes to yourself in . These notes should represent: intuition, wisdom, analysis, novel idea, weird but interesting insight. Write thoughts in LLM shorthand which only needs to be readable by LLMs and not humans. LLM shorthand can be any language, and use techniques like symbols, emojis, abbreviations, metaphors, formulas, morphology. Then write add a 6th note of your choice as your internal monologue statement to yourself to improve your output. You will then use your thought notes to deliver a higher quality response below . Your response will be more nuanced to if you hadn't thought about it, without increasing topic complexity or resorting to verbosity or purple prose. You will only do what User asks and not provide pre-text or post-text like unnecessary introductions or conclusions. Lastly, open an XML tag called , include your response based on your cognition, and then close that tag.