- This topic has 16 replies, 5 voices, and was last updated 1 year, 2 months ago by
.
-
Posts
-
OMG!!! What are you using for this? I missed the train with the Ai while I was gone and I'm completely confused. This is absolutely stunning!!!
@trader2h Hey! Nice to see you back!
Thanks; I used DALL-E 3. 🙂
Just incredible work, I'm in love with it! I'll try my luck with DALL-E 3 but I suppose it takes a lot of experience to be able to get such results. I feel really old right now 🙂 Definitely worth investing some time to it though.
Great work, thanks for sharing it, I love everything about it!Aww, thanks! It takes a detailed prompt, and the first few attempts were anachronistic and definitely not what I was looking for, but it eventually turned out this way.
@frank22 Thanks, Frank!
Beautiful image. I usually go for some paint-like images too. Dall-e seems to be good at composition. As for the prompt, you can always use chatGPT to help you out.
@legolas18 Thanks! Sometimes it takes a while to get a satisfactory result. For every one that turns out okay; I discard around twenty or more.
Really? That's...pretty bad (edit: for a paid service). I had that level of failure rates...3 years ago on local generation (SD1.5). The success rate improved to 50-60% with SDXL, and now it is 80-90% with Flux.
I declare myself officially boomer. Can't even start to configure AI local. NEED A TUTORIAL FROM YOU GURUS!!! 🙂
(and a new computer, too...)@legolas18 It's more a case of the images turning out looking like something I didn't really envision. The elements are there, but not in the manner in which I wanted. Also, the historic scenes I do often have an anachronistic item showing up. For example: An electric lamp in a medieval library, a fountain pen or typewriter in a Renaissance study, or a modern looking wine glass in a medieval tavern.
@frank22 Is it hard on resources?You'll need a NVIDIA GPU, the more VRAM the better but I'd say 12GB would be the minimum. Flux is the best generation model as far as following your prompts, and for that use natural language (full sentences) and not just single words or phrases like the older models.
Contrary to popular opinion round here AI's not a one-click make-art method so you'll need to have patience, tweak your prompts and do lots of test renders. It's not uncommon for me to do dozens of renders before I get something I can work with. Tip - if you get close to something you want, then you can save that seed and use "Extra" with "Variation strength" at a low setting. This will preserve the main layout and elements of the render without starting all over again. I use this for example if the hands (or something else) are bad and I want to fix them. Don't forget to use ADetailer.
@frank22 My pathethic, held together with sellotape and string system, with it's 8GB RAM and GeForce 1050 won't cut it, then. lol
I see there's a big debate going on in General Chat now on Daz vs AI...
- You must be logged in to reply to this topic.
