I'm a fan of Ultra Heroines and wanted to create an image of a room collapsing due to a monster attack, with people desperately trying to escape amidst the sounds of footsteps and destruction. I tried using Stable Diffusion but couldn't get the image with flying debris I envisioned, so I've created Lora. It's set to depict a school in total collapse, but the place can change depending on a different Lora (character) used, as seen in tests. Indeed, with Asuna, it tends to be a wooden building room, and with Takina, a wooden café-like setting. Therefore, if the prompt doesn't specify a location, the image will show the associated place in ruins. If a specific location is desired, adjusting the prompt should suffice.
Description
FAQ
Comments (5)
Nice seeing LoRA for "escaping from collapsing building", and the sample images lead to a weird question: how hard is it to make "escaping from gunfire" (flying ammo and pierced surroundings)?
Probably could do the same. Though just beaware, similar to what @schsch said, it will probably break certain models.
[TIPS] It kinds of 'break' realistic models like epicPhotoGasm and Photon. It becomes very dark with low quality at weight 0.7 and higher. Perhaps you can get the background aspect if you put weight as low as 0.3 or even less, then use img2img without this LORA to improve the quality.
In 2.5D models like HASSAKU, it becomes better if weight 0.4, maintaining quality (still a bit darker), but you need to reinforce 'running' and 'fleeing' tags, or she will be just 'walking' or 'standing'.
Other than that, nice model!
On-site generation makes it really tough to work. Can you recommend any model?
I have use zenith, because I recommended to use it. So I introduce this model.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.