― Charles de Lint
Recently a artist friend 'adopted' one of my often used prompts and, as I suggested, changed only a few things in the wording and added his own modifiers. Like me, he uses Deep Dream Text2Dream and Nightcafe. His results were impressive, depicting old gnarled Druids in the woods. So I decided to try my own version, again changing a few things here and there. On both platforms I used the same prompt, but with varying Model Styles, input image and prompt strengths.
Here is one result on Deep Dream - prompt only, in Artistic Mode
This was the basic prompt I used, with only slight variations, for any other iterations that follow (exchanging old druid for elf):
Same prompt, again, no seed image used, on Nightcafe, in SDXL 1.0 Mode
A interesting transformation takes place (as mentioned above) when you change a couple of words in your prompt, but even without changing words, if you 'overcook' a seed image, as was the case with this one below, you get something entirely different - no grumpy old man, but a feminine elf! This one was done in Deep Dream Text2Dream
The result startled me, in the description I wrote (I had used a selfie as the seed image):
Once I 'overcooked' my selfie at 90% there is no turning back. But this third attempt at 60% in Artistic mode has a certain charm, though not an old guy, but a a rather cryptid looking elfin character.
I liked that result so much, I used it as a seed image on Nightcafe, where I ran it as a series in different models, but the best one right away on my first try with SDXL 1.0:
Here are a variety of results I got with the prompt altered to Elf
all done now on Nightcafe in different modes:
I mentioned in a previous post before, with every new and improved Stable Diffusion the results get slicker, and usually I prefer the old SD v1.4 and v1.5, but it depends on how the prompts are written. In this case, the brand new SDXL 1.0 worked very good.
In any event I think it is important to use relevant negative prompts (you can write your own, but I found the default works pretty good) which usually take care of the problem with extra limbs and fingers or distortions.
Another thing I am now working with is ClipDrop where I sometimes try my new prompts. Results from there I might use as seed image on Nightcafe (as well as other sites I use, such as Deep Dream Text2Dream and Dreamwombo).
I found that AI had quite a bit of training on my style (check the website HAVE I BEEN TRAINED - best when you actually log in, then you can search for yourself and see what AI knows about you) - so sometimes using myself in the prompts usually gets me results that reflect my style.
Check out my previous post
Visit my website
PRINTS OF MY ARTWORK AVAILABLE HERE:
Bonus Special Video:
DRUID FAIRIE TALES
From a single iteration on Nightcafe via transition clips on Dreamwombo and a Facebook story, the finishing touches with special effects on Tiktok. Music track from Pink Floyd was already part of the Facebook story.
should you not be on the blockchain, or have no upvote power, or this post is already older than 7 days, you can always just buy me a coffee.