Syt's Pictorial Collection of Stuff and Things (image heavy)

Started by Syt, June 07, 2015, 02:08:30 AM

Previous topic - Next topic

Syt

Was tempted to put this in the AI doom thread. :P

End of 2022 I first tried out Midjourney. The initial prompts I gave (having no brilliant ideas) was "Painting of Super Mario as Napoleonic Marshal."

This was the result with V3 which was the current model at the time:



There was an experimental test version available too which delivered this:





V6.0 is in open alpha at the moment. The model has more parameters now, so I did two versions. One with the "default" stylization options:



And one with stylization options turned to minimum:



And since he looked so dour I tried the default stylization but adding "happy" to the prompt:





I guess the quality has been improved a bit. :P

For comparison, 5.2 at default settings:





A big change in 6.0 is adding the support for text rendering, though it's a bit hit and miss at the moment (DALL-E added it a few months back, though I'm not using it much due to its fierce blocking of anything resembling current IPs or living persons).

The prompts for the below were a fair bit more complex, plus adding the occasional reference image.




(Notice the typo on "Libertad", and it also struggles to convert the inverted !, rendering " instead)









It's still struggling to create a good image of Incan torpedo boats fighting the giant ants of Brest-Litowsk. :(
I am, somehow, less interested in the weight and convolutions of Einstein's brain than in the near certainty that people of equal talent have lived and died in cotton fields and sweatshops.
—Stephen Jay Gould

Proud owner of 42 Zoupa Points.

Josquius

You have a big  subscription to this?
I've often been tempted to try it but the price is off putting.

Interesting images anyway. I like hasta la Victoria.
██████
██████
██████

Syt

I have the cheapest sub that gives unlimited relaxed (slow) generations at $24/month. You can only have 3 concurrent jobs on slow, and you may have to wait a bit, but at least you can use it as much as you like. It comes with 15 hours of fast generation. Plus, you can daily go on the website and "Rate Images" where it shows you two images and you click the one you like better. If you're in the top 2000 raters of the day you get a bonus "fast" hour. Getting into that level takes five to ten minutes of fast clicking or so, so it's fairly easy to reach. Currently, I feel I get very good results by taking an image I like, tell ChatGPT to create a natural language prompt for that kind of image (V6 of Midjourney prefers natural language prompt over comma separated tags) and then use/tweak that as prompt for Midjourney, esp. when using fantasy art or paintings.
I am, somehow, less interested in the weight and convolutions of Einstein's brain than in the near certainty that people of equal talent have lived and died in cotton fields and sweatshops.
—Stephen Jay Gould

Proud owner of 42 Zoupa Points.