yeap, in combination with HassakuXL cranks the anime characters a lot, but if you crank it past 1.25 it starts to change your characters to another preseted face, sometimes it retains the hairstyles.

i can try guessing: it being the flux version

I tried a lot of different things but i dont know if its my model, my other loras, and so on.
When i use it with HazzakuXL model and put this lora in 1.25 it results in a combination retaining anime features, if i crank it it changes the faces to a a different character.

Yes, it's because either the filename or the "output name" have two underscores (prob both) which dynamic prompts use (at least in Automatic1111, I don't know about other UIs).

I rename the Lora files, search-replace two underscores for just one. I found this extension: adjust Lora filename, which can then update the Lora to use the new filename as its "output name".

Basically, what I do:

- Search for "__" in Auto111's Lora browser
- Find offending Lora's in a file manager (just copy path from Lora browser)
- Rename Lora fles (search - replace, replacing two underscores for one)
- Go to "Adjust Lora Filename" extension
- Hit refresh next to the drop-down list
- Start typing (or pasting part of) the newly renamed Lora name
- Hit "Adjust" button
- Hit "Update" button to veriffy it's updated/changed.
- Hit Refresh in the Lora browser

(I also usually delete the .bak file it generates, you might want to keep it, or not).

It changes the Lora so at gereation time I get a warning / error in terminal about Lora "deserializatin" or some such, guessing there's a chechsum that isn't updated (or maybe something else), but it still works and generates images. Dynamic prompts also works again.

(It can be a pita to find and fix any offending Lora's though)

put the LoRA in your webui\models\Lora\ folder ... in Forge, below your negative field there are 4 tabs "Generation - Textual Inversion - Checkpoints - Lora" click LoRA ... on the same line to the right hit the refresh button ... and your lora should show up. hover over the lora and push the toolbox icon - adjust it "trigger word and strength" and save it ... then just click on the pic of the lora ... and it will be in your prompt field ... working?

@WackyArt Hey, thanks for your help. I've setup the Lora correctly but when it renders out, I don't get the results like the samples. I wonder if my forge is older what is why I don't see the correct results. is that what you have on your forge?

version: f2.0.1v1.10.1-previous-634-g37301b22  •  python: 3.10.6  •  torch: 2.3.1+cu121  •  xformers: N/A  •  gradio: 4.40.0  •  checkpoint: a5f4c4616e

@JamesMccay1995Ai sorry, few days delay :) ... something same versions ... but that's not the problem, i guess it's 'cause of the different seed?! If you look for a picture here and copy the prompt, you need the same seed not "-1" and of course all of the loras, same checkpoint etc ... get inspirations and render what you want :).

I use forge and it works fine. Best Niji lora I ever used!

Same, the pony version does not work on reforge.

Put cross in your negative prompt.

I've tried several options and haven't had any issues with the lip color. Your problem may be your prompts, checkpoint, or some other lora you use. I can tell you without a doubt that this lora doesn't have any issues with this lip color.

I did the tests on WaiNSFWIllustrious v14.

Yes indeed, realy hard to have an indoor wide shot! Have to play with strenght and compensate with a really good prompt

Kink_Lily can you give me example

It happens quite often but I'm not so sure it's the Loras fault, personally it's something that happens to me more with Illustrious than with Pony, an effective way to combat it is to change the resolutions, for example in vertical resolutions like 1024x1280 you have more chances of ending up in a closed space, but if you invert it to 1280x1024 it will probably be solved, even so sometimes it's not enough and there's no other option than to start experimenting with prompts like "hallway", "alley", and things like that, things that have to do with the infrastructure, in short you have to be more detailed/specific when describing the room or place you want to generate, I recommend trying loras that improve backgrounds, a pretty good one is "detailed backgrounds", something that can also help are loras that help you increase the level of detail of what you generate, you know loras like "Detail Tweaker" or "lazipos/lazyneg", you have no choice but to experiment until you get what you want.

MorganalOTP Thank so much, I will try what you said.

It happens to me as well, try not use indoors or outdoors prompts

Same.

Downloading before it get snuked.

@ransei_ what we can do, what we'll do????

@ggiga9 Download the model and it's metadata to your computer if necessary. There's also Civitai Archive, look it up :) You can find deleted models there ^^

Seeing as that thing has been up there since I began a year ago, it’s a bit of a joke. I really don’t think they’re going to remove this model or that message.

If i get weird eyes i switch between DPM 2M Karras and Euler A