<p>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 <a target="_blank" rel="ugc" href="https://civarchive.com/articles/7945">Flux LoRA Training article for Urban Decay</a>.</p><p>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 <a target="_blank" rel="ugc" href="https://civarchive.com/models/846120">Abandoned Parks</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/852935">Abandoned Streets</a>.</p><p><span style="color:rgb(193, 194, 197)">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 </span><a target="_blank" rel="ugc" href="https://civarchive.com/collections/5150982">KiWelten Halloween Collection</a><span style="color:rgb(193, 194, 197)">.</span></p><h3 id="version-2.0-oj0jbytvy">Version 2.0</h3><p>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.</p><h3 id="version-1.0-8itzuuvmz">Version 1.0</h3><p>Version 1.0 was trained using <a target="_blank" rel="ugc" href="https://civarchive.com/articles/7777/detailed-flux-training-guide-dataset-preparation">my dataset guide</a> 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. </p><p>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 <a target="_blank" rel="ugc" href="https://civarchive.com/models/785197/victorian-gothic-horror">Victorian Gothic Horror LoRA.</a></p>

Urban Decay: Abandoned Interiors

urban_decay_v20_rank64_bf16-step02016.safetensors

v2.0

v1.0

<p>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 <a href="https://civitai.com/articles/7945" rel="ugc noopener noreferrer nofollow" target="_blank">Flux LoRA Training article for Urban Decay</a>.</p><p>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&nbsp;<a href="https://civitai.com/models/846120" rel="ugc noopener noreferrer nofollow" target="_blank">Abandoned Parks</a> and <a href="https://civitai.com/models/852935" rel="ugc noopener noreferrer nofollow" target="_blank">Abandoned Streets</a>.</p><p><span style="color:rgb(193, 194, 197)">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 </span><a href="https://civitai.com/collections/5150982" rel="ugc noopener noreferrer nofollow" target="_blank">KiWelten Halloween Collection</a><span style="color:rgb(193, 194, 197)">.</span></p><h3 id="version-2.0-oj0jbytvy">Version 2.0</h3><p>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.</p><h3 id="version-1.0-8itzuuvmz">Version 1.0</h3><p>Version 1.0 was trained using <a href="https://civitai.com/articles/7777/detailed-flux-training-guide-dataset-preparation" rel="ugc noopener noreferrer nofollow" target="_blank">my dataset guide</a> 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.&nbsp;</p><p>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 <a href="https://civitai.com/models/785197/victorian-gothic-horror" rel="ugc noopener noreferrer nofollow" target="_blank">Victorian Gothic Horror LoRA.</a></p>

Files with SHA256: f2b76524e8b15461615bb656eddb56824c9de651f8934ea54e108912079cf7a1