Module 5: Generative AI And Evaluating Content and Tools for Learning

Using Stable Diffusion

I chose Stable Diffusion because I have never tried text to image before and Stable diffusion lets you install their models and select appropriate model weights based on your requirements. I tried to give a lot of prompts and these were some output that I received with their corresponding prompts:

• “Create an animated form of futuristic world with underwater creatures and clear sky” prompt, Stable Diffusion, 29 Mar. 2024, stablediffusionweb.ca/.

The images below were created using the online version and the above was created using the local model and weights.

• “Ronaldo and Messi playing in the same team” prompt, Stable Diffusion, 29 Mar. 2024, stablediffusionweb.ca/.

• “Climbing Mount Everest in shorts” prompt, Stable Diffusion, 29 Mar. 2024, stablediffusionweb.ca/.

• “Playing with snow in a desert” prompt, Stable Diffusion, 29 Mar. 2024, stablediffusionweb.ca/.

---

SECTIONS analysis on Stable Diffusion

Students

Students are the users using the Stable Diffusion to create unique images using their prompts. This technology is free for initial use or to introduce users to the technology and after that, users need to pay for every token used which basically means the number of words used in the prompt, roughly 1 token is almost 1 word. Major barriers that could be created are lack of understanding of the media, as we learnt this could be good form of off-loading tasks to technology but users need to make sure it is not letting them skip an important aspect of knowledge over a subject.

Ease of Use

The interface as of today has been refined and extracted out to a point where it becomes really easy to use. Stable diffusion was initially only available in form of models as I mentioned above, you would need to install those models and model weights which is the data the model would need to output any content. As the technology became more widespread, people started created their own layers and now there are multiple online platforms that would just take it in a prompt online and return the result by running the models on a cloud server. So, it becomes pretty easy to use and easily available.

Cost

The cost is not very tremendous for stable diffusion and has definitely become much better than before, it costs $7 per month when billed yearly which gives the user 2000 fast generated images and after that the render time becomes much slower. There is also another plan for $14 per month when billed yearly which gives the user 4000 fast generated images and after that the render time becomes much slower. These prices are specific to Stable Diffusion and not any other Generative AI platform, every platform has their own pricing based on their use case.

Teaching

I believe that Stable Diffusion is an amazing fit for teachers and teaching. This helps explain topic through visual indicators which we know helps enhance the content exponentially. It can be used to express a message in form of dynamic outputs to keep the students engaged and interested in the core topics.

Interaction

In the case of Stable Diffusion, the interaction between students could potentially be boosted by sparking up their interest in the output as we have seen above could be really eye catching and help students start conversation about these content outputs. The student – content interaction will not be much as the content output from Stable diffusion is generally a form of image which is not interactive and hence the direct interaction is not high. Whereas, student – teacher interaction will definitely be boosted similar to student-student interaction where these images could end up being catalyst for conversations.

Organization

When it comes to control for organizations to tailor the experience, there are different parameters that could be modified in order to create different outputs but when it comes to tailoring actual code for models, the information and knowledge needed in order to that is very scarce. The resources required to created a stable diffusion for an institution, it is not viable for most of the organizations. Even though, there could be some tweaks here or there but building an customized experience might not be that straightforward.

Networking

This technology provides a unique opportunity in connecting multiple fields by having a common ground which is Stable diffusion as a platform. People working health could benefit from talking to people working in education as the input in form of prompts is being inputted by both the user groups in a similar manner just having different domain knowledge insights. These groups could benefit from talking and exchanging some bites of information they picked up to get better output when writing prompts as an example.

Security

Security is still a big topic of controversy or a topic of concern even. These learning models are not transparent with the data they use to train their models which imposes a security concern when users which in this case could be teachers or students enter their information which could be sensitive it is not clear if that information is stored to train their models. For stable diffusion as I mentioned above, we have models which could be installed locally on user’s machine in which case there is no connection to the internet and you in a form own the data and the trained model which is slightly better still without any solid outlines on this.

Reflection Questions

Q1) What tools did you find useful in your explorations this week and how did you use them? Which ones were not useful?

I found several tools particularly useful for enhancing my understanding and utilization of Stable Diffusion. Firstly, the intuitive user interface of Stable Diffusion facilitated seamless interaction on the web app. The controls allowed me to input text prompts and generate corresponding images effortlessly. Additionally, the built-in customization options, such as image resolution and style parameters, provided flexibility in tailoring the generated images to specific preferences. Its documentation and online tutorials served as valuable resources as well.

However, there were some limitations encountered during my exploration of Stable Diffusion. Despite its user-friendly interface, the software occasionally experienced stability issues or hallucination, resulting in unexpected outputs. Additionally, the quality and diversity of the generated images varied depending on the input text and model settings, indicating room for improvement in fine-tuning the model. The examples could be seen above where I have prompts and corresponding output images.

Q2) Where do you think these tools will be in their evolution in 2-3 years’ time?

In 2-3 years’ time, I think that Stable Diffusion will undergo evolution and even more refinement, driven by advancements in machine learning. Here are some potential developments:

• Improvement in realistic images
• More control over customization
• Faster Speeds for processing
• Cheaper rates to process the prompts