Sign In

Urban Decay: Abandoned Interiors

106
731
160
44
Verified:
SafeTensor
Type
LoRA
Stats
572
154
Reviews
Published
Oct 8, 2024
Base Model
Flux.1 D
Training
Steps: 2,016
Usage Tips
Clip Skip: 1
Strength: 1
Trigger Words
urban_decay_rooms
Hash
AutoV2
F2B76524E8
The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.
IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

Urban Decay is a new LoRA series that is both a continuation of my Flux training experiments and another fun Halloween-themed LoRA that can add some freaky options to your generated images. You can see the full details of the experiment's results in the Flux LoRA Training article for Urban Decay.

Abandoned Interiors focuses on the inside of various rooms and locations. By adding the LoRA and trigger to your prompt, you’ll instantly transform an otherwise normal location into a decaying and abandoned shadow of its former self. Walls peel, ceilings fall, and the floors become covered in rubble and debris. For more urbay decay, check out Abandoned Parks and Abandoned Streets.

This LoRA is made as part of the KI Welten Discord Group Collaboration. For the other Great Halloween-themed Flux LoRAs, follow the link to the KiWelten Halloween Collection.

Version 2.0

Version 2.0 is much better at making people without as much distortion, and images tend to have better overall composition. Both v1.0 and v2.0 were trained similarly, but v2.0 used the d2 flux fine-tune rather than the d1.

Version 1.0

Version 1.0 was trained using my dataset guide and my current default LoRA training setup: 31 images, 512 resolution, GPT captions (human edited), batch size 2, for ~2000 steps. It’ll serve as a control for future versions. Use the trigger word “urban_decay_rooms” along with the interior space you want to transform. Options trained into the LoRA include “room,” “bedroom,” “hallway,” “stairs,” “prison,” “hospital,” and “church” – but other locations should work as well. 

The majority of training images did not include people (being abandoned places and all), and I’ve noticed that people prompted into the images can be distorted/disfigured. I’m not sure if that can be improved by altering the training settings, but it is something to improve in future versions. It does make for some freaky-looking images. For additional fun, combine this with my Victorian Gothic Horror LoRA.