Hundreds of children are getting stuck in the woods at this time of year. Be cautious!
Just another silly idea. 🙂
And a test for AI models as most of them had lots of problems with this. This was the first image that remotely captured the idea behind it, that the child got stuck in a frozen puddle of its own drool.
The fact that the prompt says “comical man” is completely ignored as this model has a strong female tendency as most models have.
Here is another model trying its best, creating a cute image nonetheless. ^-^
Somehow an inflatable pink washing machine came up in chat at work and someone requested an AI image of it, so… here it is! 🙂
It was a bit tricky to get it to make the washing mashine look “rubbery” and inflatable and he would not give me a mouth piece like on a balloon, but otherwise it was one of the easier stuff to create. I’ve used my current favorite “Artsy Dream”.
I tested out some pixel art LoRAs and I was pretty impressed with the result. It is still not real pixel art, but it is far better than the first variants.
I guess the model did its best to give me what I wanted, but it got really confused halfway through, I think. No idea where the shark came from. Xd
The prompt was something like: a real feral bear sitting in a cave, the bear holds a human who looks like a plushie and wears a sailor uniform, surreal, absurd, fantasy, highly detailed, masterpiece, best quality
Here you can see another image of that series:
All seems to be fine until you notice the hand of the girl. There is something really fishy going on. Maybe she isn’t even human.
Yes, this bunny has seen stuff. It was supposed to be a bear, too. The sailor doll is not very doll-like yet, either.
I tried this in several models and altered the prompt a bit which changed the idea to a human doll holding a bear and Artsy Dream was the first model that gave me (kinda) what I wanted:
This is pretty close, although this model did not really give me “photorealistic live feral bear”, but… it is cute nonetheless. ^-^
This one is the best so far, because there is some contrast between the bear and the doll that makes you believe that the bear might be real.
Whenever little Teddy felt down, he cuddled up to his favorite doll “Sailor Jim” who he could entrust with of all of his problems and secrets.
Here is another nice image done with Artsy Dream. As you can see (when getting close enough to the monitor and potentially squinting), there are stripe artefacts in the image in the darker regions.
After asking the interwebs about this, it seems that this effect appears when creating images larger than a certain threshold with Flux (or this model in particular). Horizontal or vertical stripes/bars appear when upscaling the generated image (which is usually 1024×1024 in any aspect) to twice the size.
Here is another example of a “ruined” image:
Here are pretty obvious stripes at the bottom in the dark area.
I tried a lot to get rid of this by using different upscaler models other than my usual one (4xUltrasharp), but nothing really helped. Before I knew about the size limitation of Flux, I also tried different sample methods, CFG values and so on.
In the end, I tried using a different refiner model for the upscaling part. After some trying some, I ended up with Flux1.Dev, the original.
Here you can see, that the stripes are gone. Yay! 🙂
Unfortunately, this also alters the contents a bit and I actually like the bunny in the original image better, but it’s better than an image with obvious artefacts.
You can still play around with some values to get the image to look the way you want, maybe try using other Flux models that are closer to the Artsy Dream contentwise.
