Type | |
Stats | 45 1,374 409 |
Reviews | (10) |
Published | Nov 24, 2024 |
Base Model | |
Training | Steps: 2,002 Epochs: 1 |
Usage Tips | Clip Skip: 1 Strength: 0.5 |
Trigger Words | moxxie suit tail hooves fingerless gloves white freckles |
Hash | AutoV2 CD0AAC0904 |
Its one helluva day to get slain.
Appearing stage left is Moxxie from Vivziepop's hit cartoon Helluva Boss! (Please go give Vivzie some love, the artist and Helluva Boss crew really have outdone themselves with this awesome, [still ongoing] instant classic cartoon animation.)
There will be multiple versions of this LoRA. One of them is trained on NoobAI.
The PonyXL version may get another version in the future.
Usage on Noob:
Also use "furry" when using furry-style artists.
Feel free to add quality tags. "masterpiece", "best quality" and "good quality" give different style compositions and add some spice to the image. Even "normal quality" and "worst quality" in the positive prompt can help (but you can put them in the negative prompt).
Always have something in the negative prompt when generating with noob (especially with this LoRA).
Noob tends to generate images without color, so: if you want to ensure that there is some color in the image: specify at least one color in the prompt. I recommend "red skin", which would enforce Moxxie's skin to be red.
Use (contrast:0.7) on the neg to remove ultra high contrast in colored gens.
Lower the model strength when less tags are present. I recommend starting off with 0.5-0.6
if you have just a few prompts.
The model works fine for 80% of the time with a strength of 0.7.
As you increase the amount of tags: you will need to increase the model strength so that it can continue to generate Moxxie as a character.
Usage on PonyXL:
The first version of this LoRA is very dependent on the LoRA strength. It can generate with Moxxie features at a strength of 0.1, but it is more an influence. With a strength of 0.3 the character looks like Moxxie in the original PonyV6 model and is very flexible and obedient to prompting. Other models might need a LoRA strength of 0.5 to gen Moxxie with all his features while still being flexible enough to change Moxxie, his wardrobe, his body type, his gender, his accessories, etc. The LoRA becomes more rigid at a strength of 0.7. Making Moxxie in other genders tends to be difficult in various PonyV6 checkpoints with a strength of 0.7. Changing his clothes and other features (like hair color/size, skin color, etc) is possible in strengths from 0.1 to 1 (in the various models I tested). The LoRA is still flexible with a strength of 1, but some models may require a lower strength because the LoRA tries to impose a cartoon style (even in realistic models). Lower the strength if style becomes an issue.
The PonyXL LoRA may automatically generate characters with nsfw parts. If this happens: one can get rid of them by mentioning them in the negative prompt.
"nose, snout, nostrils" in the neg works in some models to remove these face features. The original pony model tends to not generate them unless the strength has been lowered quite a bit.
All Models:
Use "moxxie" as the keyword.
You don't need to use all the keywords to generate Moxxie. But they are words that the LoRA responds to.
The models sometimes breaks off the tail from Moxxie, so I recommend that you put "tail" in the positive prompt when genning him. Unless you don't want a tail: then put it in the negative prompt.
These models were trained with images that have both SFW and NSFW content so that the model can understand what Moxxie looks like underneath his clothes and so that it does not try to gen the clothes on every gen.
If you want to gen Moxxie in his traditional suit put "suit" in the prompt.
A whole body image? Type "hooves" in the prompt and then add the pose ("standing", "sitting", "wariza", "cocked hip", etc)
"rear view" of also works.
Sometimes his skin turns another color (usually blue or black): specify "red skin" if you wish for Moxxie to have red-colored skin.
"red hooves" also works to turn his hooves red (if you like that).
If the images come out black and white: adding some color to the prompt tends to fix it (putting "red skin" in the positive prompt also tends to fix this issue).