Part of this bounty request https://civarchive.com/bounties/317/scuba-diving-gear-lora
Honestly one of the most challenging ones, as the ai is just refusing to recognize certain parts of the gear.
But I made it work most of the time, the weight is pretty flexible , you can go as low as 0.4 and you will still get decent results as long as you use the right prompts. I recommend 0.7 for best results.
It's a good idea to prompt things like diving mask, scuba tank, goggles, etc , etc. Its not required but it does help guide the AI.
You will still get some abomination results due to the ai mistaking body parts for gear parts.
Works well with either Dreamshaper 8 ,M4rvels checkpoints or my own merge Wolf Systems.
Some anime checkpoints tend to struggle with it. and it works best in portrait, as landscapes seem to confuse the ai and body parts go flying in all directions.
Description
FAQ
Comments (9)
HOW? HOW can it be SO good??? 👍👍👍
Thank you, I thought it was average at best, but I appreciate the feedback
Oh i know the pain, i attempted to make one before, it's pretty rough.
Yours look better than mine... good job ;)
Thank you kindly
@Wolf_Systems To both legends, would be nice to see this expanded to separate scuba mask (or gas mask, with or without the air tanks) from the wetsuits, since those as LoRAs tend to be rarer, but with a lot more data online with more variation (full bodysuit with multiple colors vs two-piece and pitch black).
is it possible to make few lora then combine it? i mean we make some focused lora like scuba tank first, then focused with bcd, fins etc, making one equipment as one lora. and then remix them in one lora?
I guess you could, but you are probably going to get the same results in the end, It will be similar if not the same as using one Lora.
Basically a nice idea for a Lora, but I struggled because it seems to make very ugly faces. Lots of fails and no real good result. Sorry.
The faces are not actually due to the lora. I recommend using adetailer if you generate images locally on your own machine, which fixes faces for you that are distorted mostly due to resolution. You can see the difference in my results, compared to yours. Or if you are training on the site, which doesn't currently support adetailer, you have to bump up the resolution. Basically smaller resolutions tend to distort faces. Also I noticed discoloration and anomalies in some of your results, I can't really tell, but sometimes that means you are trying to use multiple loras which sometimes don't work well together.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.