Here’s today’s update on the latest in generative AI.
Swarm by OpenAI
OpenAI just released Swarm, a framework that helps coordinate multiple AI agents (basically, programs that work together). Even though it’s still being developed, it makes it easier to control how these agents interact. It’s like setting up a team of AIs to work together on different tasks. You can check out the experimental framework on their
https://github.com/openai/swarm
Arcee-SuperNova-Medius: The 14B Parameter Powerhouse
Arcee-SuperNova-Medius is a 14 billion parameter model created by Arcee.ai. It’s based on a smart system that pulls in knowledge from two different models: Qwen2.5-72B and Llama-3.1-405B. By mixing these two models, SuperNova-Medius is great at handling detailed instructions and complex reasoning while still being efficient on resources.
https://huggingface.co/arcee-ai/SuperNova-Medius
LLAMA-3_8B_Unaligned_BETA
Another exciting one from Arcee.ai is LLAMA-3_8B_Unaligned_BETA, also with 14 billion parameters. This model shares the same clever design as SuperNova-Medius, combining knowledge from different models for top-notch performance. You can explore more about it on Hugging Face. https://huggingface.co/SicariusSicariiStuff/LLAMA-3_8B_Unaligned_BETA
Dream Machine API v1.1.0 Update
The Dream Machine API has been updated to version 1.1.0. It now includes:
- A callback function so you can monitor video generation in real-time and retry if any errors happen.
- A credit balance check to track your API usage easily.
It’s all about making your workflow smoother when working with video generation! https://docs.lumalabs.ai/changelog/callbacks-and-credits-balance
Dreamina AI V2.0
Dreamina AI just got a big update to V2.0! It’s a multi-talented tool now, supporting everything from text-to-image generation to video and music creation. It even helps you create storyboards. Whether you want to turn words into images or make creative videos, this update has you covered. https://dreamina.capcut.com/
LGM: Large-Scale Multi-View Gaussian Models
LGM is a new system for generating high-resolution 3D models using text prompts or images. It’s fast, efficient, and designed to work with multiple views. This makes it perfect for creating detailed 3D content. It’s great if you’re into 3D modeling or need to create high-quality visuals for projects. https://me.kiui.moe/lgm/
https://huggingface.co/spaces/ashawkey/LGM
INTELLECT-1: A Worldwide Volunteer Effort
Introducing INTELLECT-1, the first-ever decentralized AI model with 10 billion parameters, trained by volunteers from around the world. This project shows how people globally can come together to train a massive language model, allowing anyone to contribute to the development of AI. https://www.primeintellect.ai/blog/intellect-1
KoboldCpp v1.76: New Features
The latest version of KoboldCpp (v1.76) has been released, featuring:
- Anti-slop sampling, which lets you block certain words or phrases from being generated.
- A new role-play character creator, allowing you to design custom characters for role-playing scenarios.
KoboldCpp continues to be a user-friendly option for generating text using AI models.
https://github.com/LostRuins/koboldcpp?tab=readme-ov-file#koboldcpp
JoyCaptionAlpha Two for ComfyUI
JoyCaptionAlpha Two is a new plugin for ComfyUI that automatically generates creative captions for your generated images. If you’re into adding catchy titles and descriptions to your AI art, this plugin will make things a lot easier.
https://github.com/EvilBT/ComfyUI_SLK_joy_caption_two
REPA: Faster Diffusion Model Training
REPA is a technique that aligns noisy input from diffusion models with a visual encoder. This results in a faster and more efficient training process. In fact, with REPA, training the SiT model is up to 17.5 times quicker. If you’re working with diffusion models, this could seriously cut down your training time. https://github.com/sihyun-yu/REPA?tab=readme-ov-file
Playbook3D Plugin for ComfyUI
The Playbook3D plugin allows you to bring 3D-rendered images from Playbook3D into your ComfyUI workflows. This makes it easier to integrate 3D content directly into your projects.
https://github.com/playbook3d/playbook-nodes
HeyGen Demo: AI Avatars
HeyGen lets you create AI avatars and compare real footage with footage generated by cloning the user. It’s a cool demo that shows how close AI-generated avatars are to real ones.
Upcoming Book: How Novelists Use AI
There’s a book coming out on October 17th, all about how professional writers can use AI to develop new story ideas and craft better novels. If you’re a writer looking to harness AI, this book could be a game changer.
That’s all for today’s AI updates! See you next time!