These lora were trained on about 160 images from the CoMix Wave films, Your Name, Weathering With You, and Suzume. My goal was to capture the background art style from the films. I did not include any of the characters (or any person) from the films in the data. There are two main reasons for this.<ol><li>I really like background art and the background art from these films are some of my favorites.</li><li>I wanted to use this lora to greatly improve the backgrounds of the gens for my other loras.</li></ol>I am happy to say I have achieved my goal. It's not perfect, but I hope this lora will bring some life to your gens as well.V2 was trained on AutismMix. It was an attempt to increase the density of detail in the background art. What I actually managed to accomplish was make a version with greater prompt comprehension, more detailed and accurate characters and a bit more coherent backgrounds. You can compare the two loras and decide which one you like best. To sum it up, V2 is like V1 but easier to prompt accurate results.Some things to take note with these lora. The tag Scenery is extremely over powering in base Pony. You will struggle to gen characters if you include it in your prompt. The more detail you get with explaining the background, the harder it will become to add characters. You must work with a balance or focus on characters mainly. Prompting accurate backgrounds is a super gatcha in stable diffusion and XL is no exception. V1 was trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/257749/pony-diffusion-v6-xl">Based Pony</a> for the ultimate compatibility.To get the results you see in the V2 images above, these are the settings I used.Prompt: score_9,score_8_up,score_7_up,score_6_up,source_anime,anime screencap,&lt;lora:CoMix_BGXL_V2:1&gt;, (more scores will give you more detailed backgrounds)Negatives: mosaic censoring,bar censor,censored,source_cartoon,source_pony,source_furry,simple background,Sample: DPM++ 2S aSteps: 26 (Whatever gives you the best results)CFG Scale: 7Extension: [sd-webui-color-enhance] set to 0.5 (this increases the saturation a bit)Upscaling: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/4560">Follow this method</a>The V1 includes images using another lora of mine <a target="_blank" rel="ugc" href="https://civarchive.com/models/269772/memaxl-flat-anime-style-pony-lora">MeMaXL</a> since the 2d style is similar to the film's character art. The rest of the images are just the lora alone. V2 didn't seem to need the extra lora at all. It captured the 2d style all on it's own. Very nice. Even though V1 was trained on Pony, I exclusively use <a target="_blank" rel="ugc" href="https://civarchive.com/models/288584?modelVersionId=324619">AutismMix</a> and I recommend you do the same. All of the examples were made with that model.That's all folk, enjoy!

CoMix BGXL | Anime Scenery Pony Lora

V2 was trained on AutismMix. It was an attempt to increase the density of detail in the background art. What I actually managed to accomplish was make a version with greater prompt comprehension, more detailed and accurate characters and a bit more coherent backgrounds. You can compare the two loras and decide which one you like best. To sum it up, V2 is like V1 but easier to prompt accurate results.

Files with SHA256: bc75289a12b4b7643ebb7767b39c47bc74efd94a89e514894c7f19520225cbec