WF1 said: Nice ones! Muddy ball games are a great idea, I needed to try this also...
These are fantastic. I'd believe that these are real photos. It really looks like they just started playing mud volleyball in prom dresses! Cute girls too
LeRex said: A handful of images. I've looked at others, but I can't seem to get mine to pop like them. Maybe being too over critical of my own work.
Do others just use 1 particular generator, or are multiple programs/apps used?
What is is you don't like about these? Is it the cartoonish look? To me they certianly 'pop', but lack realism. Are you prompting for "photorealism" or "photorealistic"? This can make things look like a CGI render. Things like prompting for lighting and camera lens settings can help with realism.
I just use one Dalle-3 generator (either BIng or ChatGPT)
Creations from DALL-E always have that (at least) slight cartoon-ish look to them - or maybe the "DALL-E look" which usually makes them recognizable at once.
I agree that adding certain keywords can make it a bit better, but it does not quite reach the levels of realism possible with other AI engines.
WF1 said: Nice ones! Muddy ball games are a great idea, I needed to try this also...
These are fantastic. I'd believe that these are real photos. It really looks like they just started playing mud volleyball in prom dresses! Cute girls too
Thanks, glad you like them Midjourney is really good for creating lifelike and realistic looking images.
I have been working with Flux base model and seeds for using consistent characters to try and workout a decent coverage using various prompt and substances. I can say I am quite happy with the results for now.
Here's a few more mud/slime pics. I'm pretty pleased with these, but still struggling with the amount of volume. I've also been looking at doing a custom training, to see if this will help. But being new to this whole thing, could turn out a mess (no pun intended).
Sometimes Bing is very frustrating. In this one the geometry of the box is all wrong, and the way it pours on her head is screwy... But then the way it pours off her butt, and the feet, is great. It's so frustrating trying to get that one golden image where everything is right.
LeRex said: Do others just use 1 particular generator, or are multiple programs/apps used?
I now mostly use Flux and don't do anything further.
But when I used Bing, I always ran the images through Krea afterwards, with a prompt to boost visibility of body features. Here is a 'straight-from-Bing' and 'enhanced-with-Krea' comparison. Krea also does amazing things with faces, without even being prompted to.
Continuing my bath style pics from before, here are some more. In general. I'm pleased, but I am getting random artifacts in some, the odd extra limb etc. Slowly getting there.
LeRex said: Continuing my bath style pics from before, here are some more. In general. I'm pleased, but I am getting random artifacts in some, the odd extra limb etc. Slowly getting there.
Is that with the base dev model or are you training a lora for them?
I've been having a lot of fun with Flux, but I can't find a good pie in the face LORA. Something simple with just whipped cream. I tried to train one off of youtube video screenshots but the quality was terrible. Good coverage and rendering, but everything was blurry. If anyone has suggestions for either training data, LORAs or prompting I'd appreciate it.
For now, some simple pie tease photos created using Flux of a messy mistress and her submissive WAM girl.