2023 seems to be once again the year of AIs. The year, in which everyone can test out the latest revolutionizing developments from their home and is actually impressed.
Springwolf’s new logo
And so, we tried it out as well. We were looking for a logo for the springwolf open source project. Neither of us are graphics designer nor have a budget to hire one. And, we also have no specific design in mind.
We head over to OpenAIs DALL-E and start prompting it with texts like: “A round logo with a grey wolf and a green leaf”. It took us some tries to find good prompts and testing also other diffusion image generators. In the end, we loved the following image most and made it our logo:
To be honest, I am quite impressed by the output and how many different variation of logos have been created throughout the process.
How AI can assist you
Using prompts, an user can direct the AI to create something of their liking. When you expect something different, adjust the prompt or press the regenerate button.
This is based in nature on trial and error, even when an user gets better at writing prompts1. Depending on what your goal is, this might not be a problem at all.
As in the example above, this is a great way to brainstorm and generate ideas fast. It starts the discussion and opens up the solution space. After generating many ideas from different prompts and multiple AIs, I tend to narrow down on what I like, select one and apply some final touches manually - using a different tool2.
Thus, these AIs are one further tool in the tool box to speed up certain tasks. As with all tools, it requires practise and skill to use it.
However, it does not replace an existing tool. Assuming I have a very specific expectation3, the AI might be unable to deliver (or I don’t know what the required prompt would be).
Conclusion
It works great for show case scenarios where you don’t have any specific expectation and the solution space is large. The more constraint the output has to be, the lesser help is a generic AI. But possibly an AI trained just for that purpose might be helpful - assuming it is feasible. Still, AI will not be helpful all the time.
We humans are incredible capable of creating and innovating ourselves. Learn to use AI, but don’t rely on it.
Yes, there are parameters like a seed, etc. But as humans haven’t found a way to reliably adjust neural network weights to fix certain input-output combinations, so we can neither claim that we understand neural network as a whole nor that we know which prompt will lead to an specific image. Due to its size, it is essentially a black box. ↩︎
There are other AIs that can also adjust images based on an area selection with a prompt to morph an image into it. ↩︎
A text banner is one of those examples. ↩︎