As I read online, I bookmark resources I find interesting and useful. I share these links about once a month here on my blog. This post includes links related to generating AI images, character images, AI voices, and automated podcast transcriptions.
I generated the images below in Midjourney by retexturing a flat icon (also generated in Midjourney). Midjourney can keep the structure of the shape but provide numerous variations of textures, materials, and styles based on your prompts.
Generating AI images
Supporting Learning with AI-Generated Images: A Research-Backed Guide – MIT Sloan Teaching & Learning Technologies
Suggestions and examples for using AI-generated images in meaningful ways to support learning, without adding confusing or distracting images. Consider cognitive load and the purpose of your images.
A study by Sung and Mayer (2012) suggests that any graphic in a learning experience will fall into one of these three categories:
Instructive images: These visuals directly support learning and facilitate essential cognitive processing of core concepts. For example, a diagram illustrating Porter’s Five Forces can help students better understand this business strategy framework.Decorative images: These graphics enhance aesthetics but don’t influence learning. For example, an image of a business handshake can be visually appealing but won’t support or obstruct students’ understanding of negotiation strategies.Distracting images: Sung and Mayer call this category “seductive” images. While these visuals may relate to the topic, they impede learning because they require extraneous cognitive processing. As an example, consider a complex organizational chart of a full corporation in a lesson on team leadership. The image connects broadly to the lesson but also highlights a lot of irrelevant details, distracting students from the key concepts.
The recent history of AI in 32 otters Ethan Mollick shows the progression of AI image and video generation with iterations of a prompt about otters using wifi on a plane. He also explains the difference between diffusion and multimodal image generation models (Midjourney vs ChatGPT). These tools get such different results because the underlying technology and approach is different.
While LLMs generate text one word at a time, always moving forward, diffusion models start with random static and transform the entire image simultaneously through dozens of steps. It is like the difference between writing a story sentence by sentence versus starting with a marble block and gradually sculpting it into a statue, every part of the image is being refined at once, not built up sequentially.
But what makes diffusion models interesting is not their increasing ability to make photorealistic images, but rather the fact that they can create images in various styles.
Unlike diffusion models that transform noise into images, multimodal generation lets Large Language Models directly create images by adding tiny patches of color one after another, just as they add words one after another. This gives AIs deep control over the images it creates.
AI image generators tend to exaggerate stereotypes
The examples in this article are all from older images, but the problems of bias in AI image generators remain. Unless you are explicitly prompting to avoid stereotypes, AI image generators reflect the bias of the images they trained on. Even if you do prompt to avoid stereotypes, it can still be a problem.
Demystifying LoRAs: What are they and how are they used in Stable Diffusion?
Quick introduction to LoRAs and how they’re used for fine tuning image generation in different styles for Stable Diffusion (and other image generators)
Character image generation
Creating with Gen-4 Image References
Runway Gen-4 offers a more controlled workflow for generating images with consistent characters and scenes. Add either a single reference image of a character or multiple reference images to combine multiple characters in a scene or specify the setting or objects. This is a more complex workflow than just chatting with ChatGPT, but it gives you more precision and more consistent results. This is Runway’s documentation on using image references. My preliminary testing with Runway has been very promising.
How to achieve character consistency
How to video from Flora about how to create images with character consistency across different scenes. This is a more time consuming and technical process involving training a LoRA (low-rank adaptation) on an initial set of images for a character. This probably works best with real people, but there may be ways to adapt this workflow for elearning with generated characters. This is more effort than I would do for most projects, but might be worth exploring if I need something higher quality for a specific project.
AI voices
Rime | Voice AI
The advances in AI voices continue to impress me. Rime is aimed more for organizations using AI voices for customer service or live conversations, so it might be useful for voice chat in training applications. There’s a free plan available to test it out.
Automated podcast transcription
Podcast Transcript AI – Transcribe Any Podcast For Free!
Generate a transcript of any podcast on Spotify or Apple Podcasts. Search for the name of the podcast, pick an episode, and get a transcript emailed to you. I wanted to get a transcript of one of my podcast interviews, and this looks like a quick way to generate one.
Additional curated resources
Check out my complete library of links or my previous bookmarks posts.
