Learn and Do with Peter, 2024-04-03

Recorded live AI Salon / Learn and Do with Peter session on AI topics.

YouTube: https://youtu.be/uk4H5cd42Ng

Next: Learn and Do with Peter, 2024-04-11

Previous: Learn and Do with Peter, 2024-03-29

Notes: Learn and Do with Peter Notes, 2024-04-03

Video Chapters

  • 00:00 Getting started, agenda
  • 01:06 Checking HackMD
  • 03:43 Midjourney folder
  • 04:09 Midjourney character reference (cref) - Seraphina
  • 19:46 Midjourney character reference (cref) - Josh
  • 27:25 Midjourney "start from scratch", cute mushroom line art, Discord
  • 44:31 Dream.ai
  • 46:43 "coloring page"
  • 48:17 Midjourney style reference (sref)
  • 52:26 Jankifiers
  • 1:05:45 Tokenizing jankified text
  • 1:09:42 Modifying a jankifier
  • 1:12:25 Copyright + open source (MIT License)
  • 1:17:28 When Midjourney creates an image of a copyrighted character (Batman)
  • 1:20:25 Back to modifying an open source jankifier
  • 1:25:45 What does an image generator do with a jankified prompt?
  • 1:33:09 Midjourney v6, "stock" types of images
  • 1:37:04 Some recap
  • 1:39:21 ChatGPT and image prompts
  • 1:43:38 ChatGPT balks at creating an image from a jankified prompt (Letterrific)
  • 1:43:57 ... but we're persistent
  • 1:44:22 and get a good image from DALL-E
  • 1:47:11 More about Midjourney "stock" types of images
  • 1:48:58 More about the DALL-E image, sharing to AI Salon
  • 1:49:48 Great Wave off Kanagawa (Hokusai)
  • 1:51:12 Letterrific again, posting to AI Salon
  • 1:52:57 More recap
  • 1:54:44 Random word prompts hitting guardrails
  • 1:56:28 Midjourney Moderation Alert

AI Summary (hopefully useful, may be inaccurate)

Quick Recap

Pete led discussions exploring the use of a text-to-number conversion tool, Midjourney for character creation and image generation, along with the legal implications of using copyrighted images. The team experimented with different Midjourney tools, discussed integrating pose images into prompts, and explored language models for content generation. Additionally, they discussed Pete's ongoing project involving QR codes, face recognition, and tokenization.


Text-to-Number Conversion and Midjourney Experimentation

Pete led a discussion about a text-to-number conversion tool and its potential use in creating character references for Midjourney. The team experimented with the tool, generating various numbers, and discussed possible applications. Pete also explored the use of Midjourney's web interface for creating images and shared tips on its features. Lastly, he discussed his experience with character creation and the challenges involved, highlighting the importance of having a solid character reference and the impact of different prompts and tokenization on the final image.

Tokenization, Image Generation, and Legal Implications

Pete, an enthusiast of lighting and photography, discussed the concept of tokenization and character weight (CW) in image generation, and shared his ongoing project involving QR codes, face recognition, and tokenization. He also explored the use of Mid Journey and Jankifier for image manipulation, and the potential for selling generated images under certain conditions. However, concerns were raised about the possibility of inaccuracies in the generated images and the legal implications of using copyrighted images, with emphasis on the importance of including a copyright notice when using Midjourney. The team also discussed the potential for integrating a pose image into the prompts to achieve specific image styles.

AI Character Creation and Legal Challenges

Pete discussed his approach to character creation using AI tools like Midjourney and OpenAI, highlighting the potential for generating unique content while also addressing challenges such as plagiarism and copyright issues. He showcased some of his experiments with Midjourney, focusing on the process of generating character references and the legal implications of using copyrighted images for inspiration. The team agreed to further explore these issues, particularly the challenges of distinguishing between original and AI-generated work, especially with open-source licenses.

New Feature, Beta Testing, and Midjourney

Pete presented a new feature for personalized content in his 'Version 6' project, which was about to enter Beta. He discussed an issue encountered during testing, where the modified content appeared unchanged in the web browser, suggesting a potential problem with the system's caching or content format. The team also explored the use of Midjourney for image creation, with a focus on its various tools and the potential for creativity. They experimented with different settings and tools, including a tool named Discord, and generated visually appealing images, demonstrating the potential of their approach. The use of language models like Chat GPT, Midjourney, and Dolly for idea and image generation was also discussed, with an emphasis on the importance of the end result.

Exploring Midjourney and Discord Integration

Pete introduced the concept of 'free association mode' for creating images using the Midjourney tool and discussed its potential for generating unique and repetitive images. The team discussed the tool's functionality, its accessibility, and the potential for expanding its use within the team. They also explored the use of Discord for finding specific images, using longer prompts for more specific results, and the potential for modifying open-source jankifiers for their own use. Plans were made to generate a large number of images for a specific project, with a focus on refining prompts, and to continue discussions and research in their Discord channel.

Optimizing Language Prompts for Image Generation

Pete and RJ discussed various ways to optimize language prompts for image generation using local tools like Dream AI and Midjourney. They explored techniques for generating different styles of images and the potential for the system to learn effective prompts. They also touched on the use of image models as character references, the impact of chaos settings, and the need for refining the system's prompt usage. Lastly, they discussed technical issues related to the creation of images, the new text addition feature, and problems with the image model generator.

Exploring Jankified Prompts in Creativity

The team engaged in a light-hearted discussion about experimenting with jankified prompts in their creative projects. Pete shared his experience with using a random word generator in his art, which inspired Daniele to suggest sharing the generated images and prompts for amusement. They also discussed issues with the midjourney AI moderator flagging their prompts as potential violations of community standards, leading to a brief exploration of alternative platforms. The team agreed to continue exploring these ideas in future sessions.