Background concept + style by artist <a target="_blank" rel="ugc" href="https://www.pixiv.net/en/users/19106776">Ayuli2424</a><a target="_blank" rel="ugc" href="https://tensor.art/models/753735622205734999/Deep-urbanized-city-background-SDXL-v0.92-XL">Tensor.art</a> | <a target="_blank" rel="ugc" href="https://yodayo.com/models/c3221a18-be5c-409d-afdc-88f3728df125/?modelversion=07854581-d8bd-4461-93ae-086976d9d3af">Yodayo</a>/<a target="_blank" rel="ugc" href="https://yodayo.com/models/e4478e4f-fbfe-4f47-8f81-51162f066597/?modelversion=2523ceef-2103-4c46-9f82-f46724b9801f">SD 1.5</a>All images have a prompt you can look at and choose what you are interested in.I'd be happy if you'd like to follow my work.Minimum recommended steps: 28Trigger word: a0bSo looking at the artist's work you'll see what I mean by background concept, variants with depth were selected, so this lora makes the city deeper if you will. SD 1.5 does not hesitate to make the city as deep as possible going into the void. The rest of the models, still prefer a 1 level decrease in standard prompt. So it's both background and concept.Possible problems/Non-obvious points:I'd say that despite the detachment of some things, the model still likes to put them in, namely: stairs, railings, lanternsUse those words in negative prompt if you want less of that.Illustrious/NAI XLTrigger word: a0bFor NTR MIX XII Recommended weight: 0.8Recommend cfg scale: 4-5.8 maybe (5)works quite badly, maybe this is a model problem, not testedv0.92 Animagine XLRecommended weight: 0.7-0.9 (0.9)Recommended words: scenery, from above, outdoorsAlso: architecture, deep cityscape, lantern, lance vines, overgrown, night, deep wide shotDifferences from v0.901 PDXL:Many people know that pony v6 doesn't work well with backgrounds, so sdxl backgrounds tend to be better in everything, more stable, prettier. I haven't found any disadvantages to using in scenery variant, but maybe with characters, sdxl will handle worse. Specifically lora is absolutely identical, but:Training on Animagine XL v3.1 v0.901 PDXL Check Checkpoints out at Recommended Resources (right)! It lists models that do a good job with the background. Mugenmalumix v41 Delorean has been removed from civit, but it is available <a target="_blank" rel="ugc" href="https://yodayo.com/models/2d83a1c0-1e3e-4b45-bd46-5134852438d8/?modelversion=92765a1f-d13f-45fb-ab62-d3ddf0e3827f">here </a>and can be downloaded. Snowpony also good, I found both to be the best for working with my lora. Recommended weight: 0.7-0.9 (0.9)Recommended words (the same like at sdxl): scenery, from above, outdoorsAlso: architecture, deep cityscape, lantern, lance vines, overgrown, night, deep wide shotTraining on Pony diffusion v6I don't recommend using this lora with <a target="_blank" rel="ugc" href="https://civarchive.com/models/376395/worldly-xl-or-anime-scenery-lora">Wordly XL</a> you can try reducing the weight of both options for compatibility, but I have not been able to see any particular improvement. I recommend you to try different popular styles it can improve the quality. v0.921 SD 1.5 Recommended weight: 0.5-0.8 (0.65)some models have overcook, so it is recommended to use a weight of 0.6 or lower there.Minimum recommended steps: 26 Trained on AnyloraMix checkpoint.Differences from SDXL:Give SD 1.5 a chance, yes it is much less stable, but it has a rich variation. highly recommended to use hiresfix. Recommended words (the same like at sdxl): scenery, from above, outdoorsAlso: architecture, deep cityscape, lantern, lance vines, overgrown, night, deep wide shot(Not all of them can have a strong effect.) I like the way lora works, but I think future versions will focus on more depth in the scene, bypassing the plants and that style.

Deep urbanized city background in style of Ayuli2424 [Pony XL, SD 1.5, Animagine SDXL]

AyuliCAMENOOB0.95tx1-000010.safetensors

v0.95 NAI 1.1

qwen v0.1 wip

v0.92 Animagine SDXL

v0.921 SD 1.5

v0.901 PDXL

Background concept + style by artist <a target="_blank" rel="ugc" href="https://www.pixiv.net/en/users/19106776">Ayuli2424</a><a target="_blank" rel="ugc" href="https://tensor.art/models/753735622205734999/Deep-urbanized-city-background-SDXL-v0.92-XL">Tensor.art</a> | <a target="_blank" rel="ugc" href="https://yodayo.com/models/c3221a18-be5c-409d-afdc-88f3728df125/?modelversion=07854581-d8bd-4461-93ae-086976d9d3af">Yodayo</a>/<a target="_blank" rel="ugc" href="https://yodayo.com/models/e4478e4f-fbfe-4f47-8f81-51162f066597/?modelversion=2523ceef-2103-4c46-9f82-f46724b9801f">SD 1.5</a>All images have a prompt you can look at and choose what you are interested in.I'd be happy if you'd like to follow my work.Minimum recommended steps: 28Trigger word: a0bSo looking at the artist's work you'll see what I mean by background concept, variants with depth were selected, so this lora makes the city deeper if you will. SD 1.5 does not hesitate to make the city as deep as possible going into the void. The rest of the models, still prefer a 1 level decrease in standard prompt. So it's both background and concept.Possible problems/Non-obvious points:I'd say that despite the detachment of some things, the model still likes to put them in, namely: stairs, railings, lanternsUse those words in negative prompt if you want less of that.Illustrious/NAI XLTrigger word: a0bFor NTR MIX XII Recommended weight: 0.8Recommend cfg scale: 4-5.8 maybe (5)works quite badly, maybe this is a model problem, not testedv0.92 Animagine XLRecommended weight: 0.7-0.9 (0.9)Recommended words: scenery, from above, outdoorsAlso: architecture, deep cityscape, lantern, lance vines, overgrown, night, deep wide shotDifferences from v0.901 PDXL:Many people know that pony v6 doesn't work well with backgrounds, so sdxl backgrounds tend to be better in everything, more stable, prettier. I haven't found any disadvantages to using in scenery variant, but maybe with characters, sdxl will handle worse. Specifically lora is absolutely identical, but:Training on Animagine XL v3.1 v0.901 PDXL Check Checkpoints out at Recommended Resources (right)! It lists models that do a good job with the background. Mugenmalumix v41 Delorean has been removed from civit, but it is available <a target="_blank" rel="ugc" href="https://yodayo.com/models/2d83a1c0-1e3e-4b45-bd46-5134852438d8/?modelversion=92765a1f-d13f-45fb-ab62-d3ddf0e3827f">here </a>and can be downloaded. Snowpony also good, I found both to be the best for working with my lora. Recommended weight: 0.7-0.9 (0.9)Recommended words (the same like at sdxl): scenery, from above, outdoorsAlso: architecture, deep cityscape, lantern, lance vines, overgrown, night, deep wide shotTraining on Pony diffusion v6I don't recommend using this lora with <a target="_blank" rel="ugc" href="https://civitai.com/models/376395/worldly-xl-or-anime-scenery-lora">Wordly XL</a> you can try reducing the weight of both options for compatibility, but I have not been able to see any particular improvement. I recommend you to try different popular styles it can improve the quality. v0.921 SD 1.5 Recommended weight: 0.5-0.8 (0.65)some models have overcook, so it is recommended to use a weight of 0.6 or lower there.Minimum recommended steps: 26 Trained on AnyloraMix checkpoint.Differences from SDXL:Give SD 1.5 a chance, yes it is much less stable, but it has a rich variation. highly recommended to use hiresfix. Recommended words (the same like at sdxl): scenery, from above, outdoorsAlso: architecture, deep cityscape, lantern, lance vines, overgrown, night, deep wide shot(Not all of them can have a strong effect.) I like the way lora works, but I think future versions will focus on more depth in the scene, bypassing the plants and that style.

Files with SHA256: 589d8b06614722fd061a7f67288c25bd595b5b1e4ea87ff9f6bb129f6ac559df