Sorry for the interruption! Here are the selected cutting-edge AI updates from the past few days, enjoy reading 👇
🦙 Meta releases Llama 3.2 series models, including vision-language and text models, outperforming open-source models in the same class.
👓 Meta introduces Orion AI glasses prototype, integrating AI and AR technologies, feature-rich with a lightweight design.
👤 OpenAI CTO Mira Murati announces resignation, nearly all founding team members have now left.
🎨 Flux adds ControlNet control, supporting Upscaler, Depth, and Surface Normal functions.
📸 omni-zero-couples tool can generate couple photos with specified style and faces in one click.
😊 Expression Editor allows free editing of facial expressions, similar to a photo version of LivePortrait.
🖼️ Sharing a ComfyUI workflow for generating consistent multiple images using Flux.
💾 123pan tool can bypass 123 Cloud's download speed limit, offering various file operation functions.
Cutting-edge News
1. Meta AI releases Llama 3.2 series models.
It includes 11B and 90B vision-language models, as well as 1B and 3B text models, which can run on local devices, providing advanced summarization, instruction execution, and rewriting task capabilities.
Detailed introduction: https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/
According to the provided data, they are the best in their class among open-source models, with the 90B vision model comparable to GPT-4o-mini.
2. Meta introduces a pair of glasses that fuse artificial intelligence and augmented reality technologies: Orion AI Glasses.
Similar to a glasses version of Vision Pro, it achieves video pass-through, message interaction, personalized AI assistant, multi-task panel, 3D interactive experience, multi-modal input, haptic feedback, and other functions.
Detailed introduction: https://about.meta.com/realitylabs/orion/
It's full of futuristic feel. Compared to the bulky Vision Pro, Meta's glasses are more likely to be accepted by people, although they are currently only at the prototype stage.
3. OpenAI's CTO Mira Murati resigns.
On the 26th, Mira announced on X that she would leave OpenAI. With this, except for Sam Altman, almost all of OpenAI's founding team members have left.
AI Art
1. Flux adds support for several ControlNet controls.
This update adds support for Upscaler, Depth, and Surface Normal ControlNet. Flux's ecosystem is gradually improving.
Model download: https://huggingface.co/collections/jasperai/flux1-dev-controlnets-66f27f9459d760dcafa32e08
Among them, Upscaler enlargement enhancement can be experienced online, with good results. Those who need it can try it out.
Usage address: https://huggingface.co/spaces/jasperai/Flux.1-dev-Controlnet-Upscaler
2. Generate couple photos with specified style and faces in one click: omni-zero-couples.
Provide a couple photo as a reference, then provide one photo each for the man and woman as portraits to be replaced, optionally set a photo style reference, and it can generate a couple photo of the specified people.
GitHub: https://github.com/okaris/omni-zero-couples
Online experience: https://huggingface.co/spaces/okaris/omni-zero-couples
I tried it and the effect is average, but if this technology is perfected, it could be used for monetization in AI couple photoshoots.
3. Freely edit facial expressions: Expression Editor.
A photo version of LivePortrait, you can freely adjust the character's expression through sliders.
Online experience: https://huggingface.co/spaces/fffiloni/expression-editor
Previously, I've seen similar functionalities in paid products. Now you can experience it for free. Those interested can take a look.
4. A ComfyUI workflow for generating consistent multiple images using Flux.
Generate four grid images at once, with character content generated in each grid while maintaining character consistency, and can be used in combination with ControlNet.
Open Source Projects
1. Recommending a tool on GitHub that can bypass 123 Cloud's download speed limit: 123pan.
A script written in Python that can bypass 123 Cloud's self-use download traffic limit and provides various operation functions such as listing files, downloading files, uploading files, sharing files, etc.
GitHub: https://github.com/Bao-qing/123pan
It provides an out-of-the-box installation package for Windows users, while users of other systems can run the script to use it.
In addition, it also provides a Tampermonkey script installation. Using the script, you can directly download files from the 123 Cloud official website without traffic limits.
