I created this background LoRA around the underground parking garage of a modest three-star city hotel. I wanted to turn everyday painted concrete, safety yellow markings and humid neon light into cinematic spaces with room for your own character.
The set explores the entry ramp, main aisle, warm VIP corner, EV charging corner, motorcycle and bicycle bays, elevator and stair lobby, utility and trash room, car wash bay, security booth, raw deepest level, and valet and key stand. A final architectural view looks up through the spiral ramp. Each location uses its own floor, lighting and camera compositions while remaining part of the same believable hotel garage.
Trigger: und3rgr0und g4r4g3
Start your scene prompt with the trigger, then describe the location, camera angle, materials and light. Keep output dimensions and generation instructions outside the scene prompt. Put exclusions in your negative prompt or the separate instructions supported by your workflow.
Example background:
und3rgr0und g4r4g3, underground hotel parking aisle, painted concrete columns with yellow bands, parked generic cars, damp gray floor, cold neon rows, cinematic lighting
Example with a character:
und3rgr0und g4r4g3, adult woman, cowboy shot, looking at viewer, standing in a hotel garage elevator lobby, rubber matting, warm recessed lighting, cold neon through the garage opening
Suggested starting points only, not tested results: for a compatible locally trained SDXL-family version, try 25-35 steps, CFG 4-7 and a LoRA weight around 0.6-0.9, then adjust to your checkpoint and sampler. No training or inference tests have been performed for this dataset. SDXL and Illustrious require their own appropriately trained versions; compatibility is not established by these images. Krea is a separate workflow: use the scene description where supported, and use this LoRA only if that workflow actually supports the trained version. These step, CFG and weight suggestions do not establish Krea support or apply automatically to it.
arrssenne
J'ai cree ce LoRA de background autour du stationnement souterrain d'un hotel urbain modeste trois etoiles. Je voulais transformer le beton peint, les marquages jaunes de securite et les neons humides du quotidien en decors cinematographiques pouvant accueillir votre personnage.
Le set explore la rampe d'entree, l'allee principale, le coin VIP chaleureux, les bornes de recharge, les places pour motos et velos, le hall des ascenseurs et de l'escalier, le local de service et des dechets, la baie de lavage, le poste de securite, le niveau profond brut, ainsi que le comptoir du voiturier et ses cles. Une derniere vue architecturale regarde vers le haut de la rampe en spirale. Chaque lieu possede son sol, sa lumiere et ses cadrages tout en appartenant au meme garage credible.
Trigger : und3rgr0und g4r4g3
Commencez le prompt de scene par le trigger, puis decrivez le lieu, le cadrage, les materiaux et la lumiere. Gardez les dimensions et les instructions de generation hors du prompt de scene. Placez les exclusions dans le negative prompt ou les consignes separees de votre workflow.
Exemple de background :
und3rgr0und g4r4g3, underground hotel parking aisle, painted concrete columns with yellow bands, parked generic cars, damp gray floor, cold neon rows, cinematic lighting
Exemple avec un personnage :
und3rgr0und g4r4g3, adult woman, cowboy shot, looking at viewer, standing in a hotel garage elevator lobby, rubber matting, warm recessed lighting, cold neon through the garage opening
Points de depart suggeres seulement, sans tests : pour une version entrainee compatible de la famille SDXL, essayez 25-35 steps, CFG 4-7 et un poids LoRA autour de 0.6-0.9, puis adaptez selon le checkpoint et le sampler. Aucun entrainement ni test d'inference n'a ete effectue pour ce dataset. SDXL et Illustrious demandent chacun une version entrainee appropriee; les images ne prouvent pas cette compatibilite. Krea est un workflow distinct : utilisez la description de scene si elle est prise en charge, et ce LoRA seulement si le workflow accepte reellement sa version entrainee. Les suggestions de steps, CFG et poids ne confirment aucun support Krea et ne s'y appliquent pas automatiquement.
arrssenne
Description
FAQ
Comments (6)
Thanks for these, this should be good for action scenes.
If you are willing take requests for krea, i'd love an olympus city inspired by your previous marble temple ones.
And an Atlantis inspired by the water goddess temples.
Hi, Can I ask something? I noticed that some of the recent LoRAs don't have an Illu version. Is it because you were short on Blue Buzz, or could it have something to do with the Yellow Buzz I sent ? I want this and this, too.recently? 🥲
Hi! 😊
As I explained in my recent announcement, I’ve been having a lot of issues with failed LoRA trainings lately, especially with Illustrious.
Since training is quite expensive, whenever a training fails, I request a refund from Civitai. While waiting for the refund, if I don’t see much interest in that particular model, I usually don’t bother retraining it.
However, Water Park is definitely one I was planning to retrain! I was just waiting for Civitai to refund me, and if I’m not mistaken, they’ve already done so. I was actually planning to retrain it this morning.
Normally, I always create three versions of each LoRA: SDXL, Illustrious, and Krea.
If one of these versions is missing, there are usually only two possible reasons:
I’m still working on it. Creating preview images takes quite a bit of time. It takes me around an hour just to prepare the Krea previews, and then I still have to create the SDXL and Illustrious previews. The whole process can take several hours.
The training failed. If several hours have passed and a version is still missing, there’s a good chance the training was defective and I’m waiting for Civitai to issue a refund before trying again.
So don’t worry, it has nothing to do with the Yellow Buzz you sent me! I really appreciate your support. ❤️
I’ll be giving Water Park another try!
@arrssenne Thank you for explaining everything! That makes sense, and I’m relieved to know it had nothing to do with the Yellow Buzz. I was just wondering because I noticed several Illustrious versions were missing recently.
Your LoRAs are always so easy to use and full of unique ideas, and I really love them. I’ll keep enjoying your work, so please take care of yourself and keep working at your own pace without pushing yourself too hard.
And thank you for planning to retrain Water Park! I’ll look forward to the Illustrious version. ❤️
@hentai_pl @hentai_pl I just retrained the Water Park LoRA for the second time, and unfortunately, it has the exact same issues again. The model is completely unusable.
At this point, I honestly don't know what to do anymore. I think I'll wait a few days before trying again because this is getting ridiculous.
I'm not even sure Civitai will refund this second attempt. Technically, the training did generate images, but the quality is so terrible that the LoRA is completely unusable.
It's really frustrating to keep wasting Buzz on failed trainings, especially when I have no idea what's causing these problems.
@arrssenne I feel like Civitai has been getting worse in certain ways every time they update their site policies or introduce changes like Green Buzz.
I've also experienced several issues myself. At one point, my Civitai training took several days to complete, only to be canceled in the end. When that happened, I found other users experiencing the same problem in the Civitai Help channel on Discord.
https://discord.com/channels/1037799583784370196/1130679274999910490
Maybe you could ask there to see if anyone else is experiencing similar issues or knows what might be causing them.
It must be really frustrating to keep spending Buzz on training runs that fail, especially when you don't even know what's causing the problem. I hope you can get some answers and that your next attempt goes smoothly. Please don't push yourself too hard!



















