This is my first Lora, or rather, the beta version. I originally created it for my own use, but I thought I’d share it with you all
About Lora:
I focused more on ruined city buildings than on details like burnt cars, destroyed tanks, rocket debris, or craters left by explosions. Out of the 160 photos dataset, I included some burnt cars and destroyed tanks, but their representation is more abstract and most likely will be distorted.
I also noticed that the images come out somewhat noisy due to tiny details like debris, grass or tree branches, and I’m not yet sure how to fix that yet.
If it makes too realistic/static images, try decreasing Lora's weight to 0.5.
A few common keywords used:
Sandbags, destroyed/burnt Russian tank, debris, detached turret (might not work), rubble, collapsed/broken bridge, drone footage, destroyed building, burnt car.
Trigger word - WarinUA
The photo descriptions were created with the help of ChatGPT, trying to match the Flux style (although short descriptions work just fine too):
WarinUA, the front facade of a building is blown apart, with massive damage to the entire structure. The floors have collapsed inward, with pieces of brick, wood, and metal hanging dangerously from the walls. The remnants of rooms and furniture can be seen through the broken windows. The sky is overcast, adding to the gloomy atmosphere. This is a snapshot of urban destruction in a warzone.
So yeah, I'd be very happy if you could upload the results of using this Lora just to see how it works for you and where I should focus on the V2.
Thanks!
Description
FAQ
Comments (6)
Destruction and decay render skills seem to be a work in progress for diffusion render models. I guess it's because we Human viewers bring so much semantic visual understanding knowledge, and complex structures can break/decay in such complicated ways. We notice the 'wrongness' in such output at many different levels.
Appreciate any LORA in this direction, and look forward to trying it.
Would really like to see some kind of SDXL version of this
Are there any difference between Russian and Ukrainian tanks and building in appearance?
Both countries use Soviet-based tanks (T62, T72, T80, T84) and BMPs. Some slight modifications have been made on both sides, but they don't make much of a difference visually. However, this Lora won't generate these tanks as the focus was on recreating an atmosphere in general.
Same with buildings. All post-Soviet countries still have the same-looking buildings as well
It is not Ukraine, you just downloaded random pics from internet, and the activation tags are a bit incorrect :)
The pics I've used to train this lora are the actual ones from Ukraine. The problem is, this was one of my first loras ever trained and because of the excitement I had at that time, I wanted to publish it ASAP. Now, I wouldn't even publish it as it's a proper fail and not even close to what it should be, including training settings, keywords, etc. And then, Civitai is banned in the UK, so I've lost interest in updating/creating loras and I only make them just for my own use. :)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.