In the rapidly advancing world of Artificial Intelligence (AI), image-generation tools have become a focal point of innovation. As highlighted in our previous blogs  [INSERT BLOGS HERE], we frequently use AI to create content, especially octopus images for our blog and social posts. The more unique and specific images, such as Rich Octopus, the more challenging to depict and prompt with AI.  In this article, we discuss our experiences with four major players, OpenAI's DALL-E, Midjourney, Adobe Firefly, and Stable Diffusion and the results of images that we generated using two prompts:

• a cute cartoon octopus hanging out with friends
• a cartoon depiction of a group of octopus friends having a picnic on the beach. One octopus is sharing a sandwich, while another is playing with a beach ball. The sun shines brightly overhead, and seagulls fly in the distance. Please draw this in muted colors in the style of pokemon
  

Midjourney

Functionality: Midjourney stands out due to its focus on mood and ambiance in image generation. It tends to generate images that not only match the description but also evoke a particular emotion or sentiment. To further refine your image, you’ll need to know how to specifically prompt Midjourney via Discord. For example, to change the aspect ratio you would input “ — aspect <value>:<value> or — ar <value>:<value>” after your prompt. It will also allow you to select specific pictures, adjust the ones you like, or make re-iterations of the same prompt in a similar style produced by clicking options presented directly from the application. 

How to Access: To access the platform, you’ll need a Discord account and a bit of Discord knowledge to operate Midjourney. You can start by stating “/imagine” in order to induce Midjourney to create images.

AI Image Generation/Process: When tasked with generating octopus visuals, Midjourney's approach can result in more atmospheric images, sometimes bordering on the abstract. We have used Midjourney extensively for this blog and for the past year, it was by far and away the only image creator that could create a cute cartoon version of an octopus. All other AI image generators could not adequately create a cute cartoon or would simply not generate the image…until now that is. Additionally, I’m not sure if Midjourney is learning my prompting style after a year of use, but it was the only image generator that I was happy with using my simple prompt of “/imagine a cute cartoon octopus hanging out with friends”. However, when further refined with the second prompt, however, it got noticeably worse.

Pricing: Midjourney no longer offers a free tier. Their monthly subscription starts at $10/month or $96 annually.

Adobe Firefly

Functionality: Adobe Firefly is part of Adobe's suite of AI tools, designed for creative professionals. It emphasizes user collaboration, allowing individuals to tweak and refine AI-generated visuals, making the output more aligned with human aesthetics. To use Firefly, you only need a web browser and a login to an Adobe account which you can create for free. Firefly provides suggestions on how to edit the photos, but I didn’t find it very intuitive, and it did not seem to change the photos as I’d like. I prefer using DALL-E or Midjourney’s ability to edit individual images via prompt vs using Adobe’s system. 

Octopus Images: With Firefly, generating octopus images becomes an interactive process. I could not get Adobe to easily change the scene or color of an Octopus. This could be due to my better understanding of prompting vs my lack of knowledge using Adobe Illustrator. Adobe will also include a creative use symbol on their creations to depict AI creation.

Pricing: Currently, there is no fee to use the Firefly web application to test. You receive 25 generation credits per month, then the premium tier starts at 4.99/monthly for 100 generative credits.

Stable Diffusion

Functionality: Stable Diffusion emphasizes gradual image development. It’s a learning model that you have to train (add data) to build images layer by layer, refining details with each step, akin to a painter adding successive layers to a canvas. To access a free ready made platform, you can use a browser and not log in via Stable Diffusion's website. You can insert your prompt and scroll through options of how you’d like your image to look. Then you enter a queue to have your picture created. Their free browser interface is the longest way to create an image, especially with the queue. There aren’t a lot of options to edit individual images and you only get one iteration vs the typical four iterations. However, Stable Diffusion, as seen in the browser is only a tip of its potential. The best use for Stable Diffusion is to run it yourself. It’s an open-source program that you can use to develop your own set of imagery that matches your asthetic.

Octopus Images: Stable Diffusion did very well in crafting octopus images as cute cartoons, but it doesn’t follow the prompting style that I requested through the prompt, e.g. it didn’t use muted colors as asked in the second iteration. However, Stable Diffusion is known for its step-by-step refinement which can produce iterations that may get you closer to what you’d like. 

Pricing: You can try stable diffusion for free. If you’d like to skip the queue and receive 999 iterations, you can pay starting at 29/month or 290 yearly. 



The Challenge of Creating Cute Octopus Images

Octopuses, with their unique physique and current imagery available, can sometimes be challenging subjects for AI. Their form is complex, and without a proper description, AI models might emphasize the alien-like qualities of octopuses, leading to unsettling visuals. The key lies in providing detailed prompts that guide the AI towards producing amiable and in certain platforms, human-AI collaboration.

Conclusion

DALL-E, Midjourney, Adobe Firefly, and Stable Diffusion each bring their flavor to AI-driven image generation. With varied strengths, from DALL-E's conversational prowess to Midjourney's nuanced prompting, Firefly's collaborative essence, and Stable Diffusion's methodical layering, they highlight the transformative potential of AI in the creative domain. The quest to generate cute octopus images highlights the importance of user guidance and serves as a testament to the evolving capabilities of AI in the creative realm.