mcrae212's avatar

mcrae212

165 Watchers40 Deviations
14.2K
Pageviews
GladysGlover1954
Lookahere
Joshuaricamonte
rexpro244
businessbbw
xnamtrihs
UnwishedCracker
bnguy
Xavierbravoop
malgosiapoland
lazyitis2009
SteveH51
sayakuku
Yuu0318
GladysGlover1954
aifetish
DucklingStories
BillboardMan
xanthippus
Lookahere
Wimi
ArrogantMadman
Elegance1989
bobbby1964
Maidenhainer
SlenderDreamsAI
Simpson84412
  • Deviant for 12 years
  • He / Him
Badges
Albino Llama: Llamas are awesome! (59)
Double Delicious Cake: But only half the calories! (2)
Quartz: It's a big honor to be awarded a Quartz badge! (3)
Mystery Egg: You submitted your 1st deviation, amazing! But what's this...? Something has magically appeared.
Baby Blob: Achievement unlocked! Your 2nd deviation made the Mystery Egg hatch.
I am still trying to roll my own with AI imaging, i.e. run Stable Diffusion on my own PC. That way it’s free and there are no nannies blocking things. But doing it this way is frustrating, too — I constantly run into hardware limitations. The system I use is an Intel 8th gen-based Dell Vostro tower. It’s still fast enough, but I constantly run out of storage on my smallish SSD. And the 8GB NVIDIA RTX 3060 GPU card is just barely good enough to run Stable Diffusion, and my Dell really doesn’t have enough watts to drive even that card (I used SATA wiring to power the card). I tried out all sorts of extensions in Stable Diffusion’s AUTOMATIC1111 graphical user interface. How to control certain blocks within a picture. How to manipulate depth of field. How to use LoRAs (small files that help create “special interest” images). How to use ControlNet (to analyze source images and use their general composition to generate new images). And so on. It all sorta kinda works sometimes. But not
Join the community to add your comment. Already a deviant? Log In
In a conversation I had on where, exactly, pictures in AI-generated images come from, I asked GPT “So, in essence, a stable diffusion checkpoint model uses learned pixel patterns to generate an image described in a prompt?” GPT: Yes, that's correct. A Stable Diffusion AI image generator, which is based on the concept of a checkpoint model, uses learned pixel patterns or learned parameters to generate images based on a prompt or a given input. These learned parameters capture general visual patterns, features, and representations that the model has learned from a large dataset of images during training. When you provide a prompt to the model, it uses the learned parameters as a starting point to generate an image that is influenced by the prompt. The learned parameters are used to guide the model's creative process in generating an image that aligns with the prompt, but the generated image is also influenced by random noise, which adds variability and creativity to the output. The
Join the community to add your comment. Already a deviant? Log In
I feel a degree of disillusion. Image AI is extremely fascinating, but in the end really only a computer’s selection of a few tagged pictures which it then cobbles together. The art of prompt crafting is elusive because there are no real, universal rules, and it all totally depends on the used model anyway. No matter how good your hardware, you’ll always need more and better. And even if you’re pretty good with tech, really and truly understanding all the coding involved requires a whole different level of immersion. That said, image AI is very likely here to stay. Within just a few months it has given DeviantArt an overall very different look. Things that used to take hours or days or months to create can now be just a well-crafted prompt away. Whether that's worth the endless iterations until something truly interesting pops up, the great deal of time spent, and the massive amounts of computing power going to waste with generating unusable stuff, I am not sure. I'll keep trying to
Join the community to add your comment. Already a deviant? Log In

Profile Comments 162

Join the community to add your comment. Already a deviant? Log In

Thanks for the watch!

Thank you for visiting my site. Could I recommend a story for you to read?

Thanks For the Watch

Thank you for starting watching me -Danke das du mich beobachtest.

Tysm for watching!