First I must say I still have an issue with the ethics of AI generators for visual arts. As many if not all, according to my research were trained on other’s copyrighted work, it’s an ethical issue that is not going away. Yes I know that now many have stopped doing this but it’s too late as these AI’s already grabbed the art of others and it’s probably still in their data sets.
Also when someone used an AI generator you have no way of knowing if they put in a handful of words or the longer prompts that some artists say are needed to create truly amazing pieces.
Having said all this and personally not having used any AI generators at this point, other than to play with them as a toy, I thought let’s do a test to see how well they generate a visual piece.
I do this because I know that others of less skill and talent than I for digital drawing and painting, will easily surpass me using these tools and I still need to compete. As a digital artist who uses Photoshop and an Xppen graphics tablet, I would never claim my work is done on paper. And I have never claimed that. I am always very honest about how my work process goes. And if you are wondering yes I know that Photoshop has AI filters, and while I have played with them to see how they work not one has EVER made it into my finished pieces. Usually, they are too much of one thing or another and I end up creating effects on my own.
So I decided to channel my inner writer and create some prompts. Back in my 20s I wrote sci-fi short stories and had some interested editors but I had to get a paying job and drop the writing. So this was a chance to test AI generators and see the results.
I am writing three prompts of increasing detail and word count. While I will not be sharing the exact words I will tell you the word count.
Prompt 1 Word Count 13
Prompt 2 Word Count 36
Prompt 3 Word Count 86
I will tell you that I used the website stablecog.com and stable diffusion 1.5 all sliders were set at the default levels and I altered nothing. All the last two images generated missed a lot of details I wrote in. So that kind of confuses me. The one generated by the first prompt captured the vibe I was looking for which is kind of scary that just 13 words did that. The last is more detailed to my prompt but still missed so much including the vibe I was looking for.
So next I altered the slider options and tried a few different AI generators available on that site.
This was the Kandansky engine with prompt 3 it has way more details matching my writing. Very close to my prompt and vibe.
This is Redshift Diffusion with options. Better than the first three. Nice vibe to it.
All three missed some very critical details of my prompts but my experience with AI prompt writing is well 1% now. LOL. I wrote my prompt as a short story writer would describe a scene. I suspect there is another way to go about this.
But ignoring the issue of ethics and copyright the scary thing is that if you are NOT picky you can generate something so fast that another artist could take hours to create. And it’s made by an AI on the work and skill of other artists who did not consent to have their work included in a data set for an AI program.
So if you use AI-generated pieces, please tag them as such. I am sure there is skill involved in writing prompts IF you are trying to get something specific out of the programs. I really agreed with a recent photographer who used AI to enter a photography competition and you can read about him here.
We need to discuss this in the art world and we need to be open. Tag your pieces so people know how you created your piece that is all I ask. Thanks.