<h1 id="live-in-architecture">Live in Architecture</h1>My first style LoRA — Live in Architecture.The idea is pretty simple: it combines a photographic subject with a colored architectural drawing of the surrounding environment. The result works especially well with wide cinematic and full-body framing, with realistic shadows that make the subject actually interact with the drawn environment.<h2 id="prompting">Prompting</h2>Keep it simple. Krea does most of the work. You mainly need to give it the frame.The most important thing: use all trigger words. Otherwise the LoRA tends to look too weak.My preferred prompt structure:object, subject, environmentFor example:<blockquote>female tourist, outside tosca village</blockquote>works much better than:<blockquote>female tourist, outside, villa, landscape, trees</blockquote>Why? Because tags like villa, landscape, trees can become part of the subject instead of the environment.The same applies to surface-related objects. If you want books and shelves to stay part of a drawn library environment, don't explicitly prompt books, shelves. Krea will create them on its own. If you add those words, they can get interpreted as the subject and become photorealistic instead of drawn.You also don't necessarily need to prompt pose or expression. In most cases, the subjects naturally fit into their surroundings, which is part of what makes the style work.<h2 id="the-slightly-different-function"><span style="background-color:rgb(40, 40, 40);color:rgb(232, 232, 232);font-family:DuckSansDisplay, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen-Sans, Ubuntu, Cantarell, &quot;Helvetica Neue&quot;, Arial, &quot;Noto Sans&quot;, sans-serif;font-size:18px">The slightly different function</h2><span style="background-color:rgb(40, 40, 40);color:rgb(232, 232, 232);font-family:DuckSansDisplay, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen-Sans, Ubuntu, Cantarell, &quot;Helvetica Neue&quot;, Arial, &quot;Noto Sans&quot;, sans-serif;font-size:18px">Don't stick to my prompt suggestions and just try the LORA as a supplement to your usal workflow: In this case, it smoothes the edges and sharpens the contrasts. The colors, especially the backgrounds, get a somewhat watercolor-like structure, coloring and lines become rather semi-realistic.It also gives your characters' poses a more natural swing, and their embedding in the environment seems much more natural to me.<span style="background-color:rgb(40, 40, 40);color:rgb(232, 232, 232);font-family:DuckSansDisplay, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen-Sans, Ubuntu, Cantarell, &quot;Helvetica Neue&quot;, Arial, &quot;Noto Sans&quot;, sans-serif;font-size:18px">Example:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/50963b61-f081-49e1-a7de-78e51ce8e9e5/original=true/50963b61-f081-49e1-a7de-78e51ce8e9e5.jpeg" />strengh:0.0 strengh:0.5 strengh:1.0 strengh:1.5 <h2 id="dataset-and-prompt-experiments">Dataset &amp; Prompt Experiments</h2>If you're interested in the prompt I used for the dataset, here's an <a target="_blank" rel="ugc" href="https://civarchive.com/images/140813630">example image</a>.The prompt is an absolute monster, but here's the funny part: 99% of it after the first line is exactly the same for every image of the dataset.The dataset itself contained exclusively SFW images of single women in environments based on architectural drawings. But the LoRA isn't locked to that concept at all.Try putting an elephant in a porcelain shop. Put two people in a room. Throw something completely different at it and see what happens.If you don't over-constrain the prompt, the LoRA leaves a lot of room for the model to do its thing. You don´t need to tell it what to do in every little detail or provide any details at all, even for NSFW. Just define the characters and the setting, then let them act. They will with the right checkpoint.<h2 id="style-and-lighting">Style &amp; Lighting</h2>Style tags like Art Nouveau, Bauhaus, Brutalism, etc. work really well and are absolutely worth using.Other styles like Gothic, Grunge, and similar styles also work just fine and are definitely worth experimenting with.Lighting works nicely too — neon, golden hour, sunset, and similar tags can push the atmosphere quite a bit. Just don't overdo the LoRA strength, otherwise stronger lighting can start creating a white vignette effect.There are example images with prompts if you want to see how little information is actually needed.For skin, additional quality/skin tags can work surprisingly well. They can also cause Krea to occasionally push the subject into a closer shot while keeping the environment in a wide cinematic composition.<h2 id="settings">Settings</h2>LoRA strength: 1.0 is the sweet spot for me.If the prompt isn't ideal, you can try pushing it gradually toward 2.0.No negative prompt needed.Character LoRAs are a bit hit or miss with this one. Some work perfectly and stay nicely focused on the subject, while others influence the environment too much.Have fun with it — this is my first style LoRA, so I'm curious to see what you guys do with it.And please feel free to post your images — I'd genuinely love to see what you create with it. Just keep it aesthetic and in the spirit of the LoRA. Nothing against X, but if things get too XXX, I reserve the right to remove the images from my gallery.

Files with SHA256: 8a30432237dfd930c45aeb5f821da44cf0a773297733f9d4429cd75461ae987d