YouTube: https://youtu.be/BuXzAYEghuc
Recorded live AI Salon / Learn and Do with Peter session on AI topics.
Next: link coming soon
Previous: link coming soon
Notes: Learn and Do with Peter Notes, 2024-08-14
The team discussed the functionalities and usage of various software tools including Flux, Glif, and Midjourney, with a focus on their potential for AI-based image generation and content creation. They also explored the use of platforms like Fal, Replicate, and Flux for running machine learning models, and the complexities of running flux on different systems. Additionally, they discussed the organization and categorization of research pages for AI instruction, the potential of using Obsidian for website development, and plans for future AI-related projects.
Flux Pro Version Limitations and Alternatives
Pete and Rick discussed the functionalities and usage of Flux, a software tool. They clarified that the Pro version of Flux is not licensed for commercial use, meaning it cannot be given to others for use. They also discussed the potential issues with using the Pro version on certain devices and services. Additionally, they touched on recent developments such as the release of Elon Musk's Grok 2 and Google's Gemini Advanced, and their experiences with these tools.
Free Software Diffusion B Tutorial
Pete instructed Rick and Hospice on how to use the free software diffusion B and draw on their Macs. He clarified the process of accessing the flux feature, recommending the Intel version for their current computers and the ARM version for future Macs. He also introduced fal.ai as a useful tool and discussed the advanced options for increased control during the generation process, which he estimated to take about a minute. Rick acknowledged Pete's comments.
Exploring Glyph and AI Image Generation Tools
Rick introduced the team to Glyph, a new tool in development that generates images quickly and efficiently. Hospice encountered some technical issues downloading the Intel version, but managed to access it and learn about its capabilities. The team also discussed various features of Glyph, comparing it to GPT, and explored the potential of using different prompts to generate diverse and intriguing images. Additionally, Pete presented a tool that could generate various types of content, including fake blue screens and memes, and discussed the use of Draw Things and Flux apps for AI-based image generation.
Mid Journey and PUL ID Discussion
Pete shared a trick he uses with Mid Journey, a language AI, to generate unique and interesting scenes or paintings based on different prompts. Rick and others in the meeting tried out this technique, which involves asking Mid Journey to create a scene or painting based on a given prompt. Pete also introduced a new AI model called PUL ID, published by Byte Dance, which generates images using someone's face. The team discussed the potential of using this model for their projects.
Exploring Machine Learning Platforms for Resources
Pete discussed the use of various platforms for running machine learning models, including Fal, Replicate, and Flux. He explained that these services offer grant money and GPU hours to certain projects and allow users to run models on their cloud, which can be beneficial for those with limited computing resources. Pete also demonstrated how to use these platforms, including how to run prompts and how to pay for the usage. He encouraged the team to explore these platforms and find the best one for their needs.
Github Setup for Fal Usage
Pete guided the team on how to use Github for signing up to Fal, a developer tool. He emphasized that while Github is primarily used by developers, anyone can use Fal to run the latest applications without needing to install them on their personal computers. Dr. Claire and Darlene had already set up their Github accounts, while Hospice expressed concerns about overtaxing their old computer. Rick shared his past experience with Github marking him as a spammer due to infrequent use. Pete reassured the team that they only need to set up an account to use Fal, and they can stop the process after that.
Flux System Complexities and Solutions
Pete and Hospice discussed the complexities of running flux on different systems, including personal computers and the cloud. Pete introduced various tools and interfaces, such as Comfy UI, Flex Web UI, Open Web UI, and Stable Diffusion, for running multiple LLMs. He also mentioned a system called Pinocchio designed to simplify the setup process. However, Pete encountered issues during the setup process due to the large disk space required and suggested that Rick or CJ try setting up Comfy UI instead.
Exploring Stable Diffusion for Diverse Protagonists
Pete explained the intricacies of the Stable Diffusion system, highlighting the role of different modules and the possibility of user-controlled routing. Rick expressed his interest in using this system for his project of retelling children's stories with diverse protagonists. Pete suggested that Rick could achieve his goal by using the pipeline manager and automating certain processes, despite the system's restrictions on using APIs. Rick considered this option, while also exploring the potential of writing his own 'jankifiers' to manage the process.
Perplexity Pages, Copyright, and AI Tool
Rick decided to transfer his Perplexity pages to his website as a blog, and discussed potential copyright issues related to the content he was using. Pete suggested Rick could use Perplexity's table of contents to structure his own site and automate the addition of new pages. The team also discussed the high viewership of certain documents and the possibility of Rick becoming a reporter for a knowledge-focused company. They further explored the use and implications of Perplexity, an AI tool that uses large language models and search engines to gather information, and its potential to circumvent copyright laws.
Organizing Research Pages for AI Instruction
Rick and Pete discussed the process of organizing and categorizing research pages for AI instruction. Rick plans to index all pages and categorize them based on content, such as AI, photography, and graphic design. Pete suggested using an HTML cleaner to remove unnecessary content from pages and customizing it for Rick's specific needs. They also discussed the potential of using Obsidian to manage and organize the content, with Rick considering automation for categorizing content. The goal is to create a user-friendly reference library for educators.
Obsidian Tool and Silly Tavern Discussion
Pete agreed to provide a walkthrough of Obsidian, a tool he uses for website development, next week. He also discussed the differences between Obsidian and Notion, highlighting Obsidian's open-source capabilities. Additionally, there was a discussion about the location of the Silly Tavern, with Pete clarifying that it was already listed on their website from February, but there was another one in the Pacam D's that wasn't yet included. Dr. expressed relief at finally knowing the name of the place they had been looking for. Pete acknowledged the need to include these details in their learn and do sessions but did not commit to completing this by the next week.
AI Trilogy and Long Rider Model
Rick shared his plans to write a trilogy about an AI that is used to elect Kamala Harris, which he is developing with a team of character developers and using AI tools like ChatGPT for storyline generation. He also mentioned his access to Ltx studios for his children's book. Pete suggested setting up pipelines for Rick's text-based work and introduced him to the Long Rider model, a tool specifically designed for writing long-form content. The team agreed to meet again the following week.
AI-generated content may be inaccurate or misleading. Always check for accuracy.