Think this is the 2nd time to reply to you. Either way here ya go. And ignore any typos cause I don't have a publisher to proof read this book you just got.

So you got 2 options.

1. Adjust your generation settings until it works.

CFG 4 may work or higher may work for some lora. I've had some that work horrid below 14. And I had one insane one that wanted 27. I usually use 5.5 and lora usually work well enough.

Sampler makes a difference depending on the type of content you are making be it cartoon/anime 2D, 2.5D, or realistic. This may change depending on the checkpoint. And the steps taken may change greatly depending on the sampler, checkpoint, lora, and prompts. Know that higher steps sometimes is worse. Sometimes the more the better. There is no one answer because it's situational otherwise you'd have the number of steps as a constant and not an option for obvious reasons. I see a lot of people say beyond X is wasting time and usually this will be true but it isn't always. But you should always start low like 25 and work up by 10 a few times testing things out to see what happens. Some checkpoints/lora/prompt combo actually have some weird magic sweet spot. Guessing it's your computer's cGi Spot.

Highres.fix & different denoising values vastly alter things. Some content will go to shit if it's above something like 0.44 while others look like shit unless you jack it up to 0.75 or so. This also changes based on checkpoint, lora, embedding, controlnet, and ect.

weight values change between both checkpoints and prompts. Sometimes you need to adjust them changing just the sampler, CFG, and so forth. Lora values can also be above 1 tho rare. It's uncommon that you'd use above 1.0 but I've done it before and gotten better results. And keep in mind, this sometimes includes trigger prompts. Lora:0.75 + PromptTrigger:1.2 with a low CFG can sometimes help when a higher CFG throws an image off and increasing lora weight hurts it as well. Don't be shy and learn your checkpoints & lora fully if you enjoy them.

Drop your prompts. Use just the base prompt to trigger it and nothing more and tweak settings. Then slowly add some in. If you're using webui, I promise you need the dynamic prompts ext installed to comment out your prompts. If you also use save state, you can easily just turn a whole lines into "#prompt, prompt, prompt" to save that won't trigger and make it fast and easy to check conflicts (note, embeddings can be commented out but a lora or anything in <> can not be and will activate). Or you can also make your own dynamic prompt wildcard, instead of commenting a line out, so a single line of text only in the file will always use that line thus it works great for things like age, race, or other concepts since some checkpoints are heavy into a certain area and those help. In your case, a text file for "TNPmavisanyloraP" & "TNPmavisanyloraN" with a single line of text to use as a wildcard if you find certain needed prompts to fine-tune it regardless of how else you prompt.

Use controlnet. Setting up a pose does wonders and sometimes you can really do some interesting concepts by using the lineart feature, of anything, and dropping how much it effects it and/or control amount and so forth.

2. Use another checkpoint. There isn't ANY checkpoint that will make what you want 100% of the time using a lora itself. Anylora is just a catchy name and not a promise. You may enjoy it as a goto checkpoint but none will work 100% of the time flawlessly sadly.

I seen your comment, downloaded this, ran it on a checkpoint that does nicely (for me) with probably 80% of lora that I used that are based on artwork/animation 2D-3D. Not that it works with any lora, I'm just lucky that it works well when I use it and adjusting all those settings is sometimes needed but it works well enough for me personally. (photorealistic anime , curvy or thin) https://civitai.com/models/61539

Now, I'll give you a run by of this testing.

DPM++ 2M SDE Karras and using swin2SR_realworldSD_4x_* as upscaler using additional network rather than prompting the lora. Using steps 75 highres 25 denoise 0.5 as it was last used this way. Testing CFG 5.5 - 14 made no big difference, highres.fix helped a lot in both, additional network weight at 1.0 vs 0.7 didn't alter the results greatly but pre-highres was better at 0.7 thus I'd normally drop denoise down to 0.44 after I verify that if I stopped there. Adjusting steps to 35 highres 25 resulted in not much change at lora weight 0.7 for quality but gave it a 2.3D look I guess I'd say. Take lora back to 1.0 and pre-highres is bad again, hands are way worse, but looks more 2.5D. sample steps back to 75 and hands improve a lot, face improves a bit and it's still 2.5D. So crank it to 150, watching it update each step shows me a MUCH higher quality on everything, so much that I'll actually try this once more by dropping the highres denoise to 0.4. Dropped to 0.4 denoise and ran 150 again perfect. Ran at 0.7 lora weight and who the hell is that? Wild random girl appears so back to weight of 1. One more test and perfect again.

I may get better results messing with prompts and adding/removing certain things that are left in by default when I do things but I'm not doing all that. I'd also run through all that a 2nd time on around 6 other sampling methods to find which works the best for this lora. I'd be adjusting the prompts to make her clothing maybe work as well. But I was just testing this is all. Below is the details, dynamic prompts is why you see new lines as I'll comment embeddings out to quickly access them. (also, may wanna read my note to the maker at the end because it may not be your fault)

Positive prompt: high quality, best quality, masterpiece, highly detailed,

(8K:1.1), (hyperrealistic:1.1),

mavis dracula, dress, striped thighhighs,

Negative prompt: (bad-hands-5), easynegative, badv5, bad_pictures,

(watermark:1.1), (copyright:1.1), (signature:1.1), deformed, distorted, disfigured, poorly drawn, lowres, bad anatomy, wrong anatomy, (extra finger), extra eyebrows, (extra hand), (extra arm), (extra limb), missing limb, floating limbs, (missing fingers),(mutated hands and fingers), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation, (monochrome:1.1),

Steps: 150, Sampler: DPM++ 2M SDE Karras, CFG scale: 5.5, Seed: 2249062652, Size: 512x768, Model hash: dc7820422a, Denoising strength: 0.4, Clip skip: 2, Hires upscale: 1.6, Hires steps: 25, Hires upscaler: Swin2SR_RealworldSR_X4_64_BSRGAN_PSNR.v2, AddNet Enabled: True, AddNet Module 1: LoRA, AddNet Model 1: hoteltransylvania_mavisdracula-10(17742687bb0f), AddNet Weight A 1: 1, AddNet Weight B 1: 1, Version: v1.3.1-RC-6-gb6af0a38

Also, a note for @justTNP

DON'T use something like "Mavis Dracula" as a prompt trigger. If you unload your lora, keep your trigger prompt, and crank up CFG. I got a girl with shark teeth thanks to mavis being a female name & dracula being a vampire. Load your lora up, lower it's weight to 0.7 and even a 5.5 CFG will produce that same shark girl every so often due to that issue.

If the dracula prompt does help on some checkpoints, then you can simply ask users to add "dracula" as an optional prompt and thus they are given better control over the prompt & weight of both. Otherwise, your lora will have issues on checkpoints that thing vampires/dracula have rows of sharp teeth rather than 2 longer fangs.

If you wish to adjust the triggers, I suggest TNPMavis or even MavieWavie if you want to have fun with it. Something that is impossible for your prompt to result in anything happening without a lora is what you need. Please keep in mind, I'm telling you because even with this altering the result, it still looks good as long as you keep that CFG quite low. I do hope you remake this because @stapfschuh may be having issues due to a higher CFG & that flaw in the lora. And I'm also hoping you're be more aware when making content in the future so you don't hamper any future lora you release due to non-unique trigger prompts.