Other download: https://drive.google.com/uc?id=1990Jo1FnMuwbBZfWYypBdtecMPFBW0at https://drive.google.com/uc?id=1zZzJ7jhDQgKGJXnN4BvIzxe7Rtd05TmH
\/ READ THE RECOMMENDED SETTINGS FOR BETTER RESULTS \/
Recommended settings:
Sampling method: Euler A, (found the best results with Eular A, 2nd best DMP++ SDE Karras)
Sampling steps: 24 and higher
CFG Scale: 1-3 (go as low as possible)
Resolution: if you want to generate girls consistently do portrait 768 which is 624x936 you can mess around with the ratios, landscapes make sure not to go above 1200 res on either height or width and you're all good
Try using portraits for girls and wide for landscapes
Using VAE might make the image quality worse
If the style doesn't show try adding "a painting of..."
Try use "text, signature, watermark, username, artist name, trademark," for negative prompts to not get the signature thing.
This model was trained on 624x936 936x624 just so you know.
This model was original a pretest but ended up pretty good. It has a pretty limited dataset only trained on (Landscape, girl) so if you try generating a man it'll look weird. The final release v2.1 might be out within this month if I gain access to a good GPU to train. The model after that will be v3, it will have easier prompting.
This was trained on Niji expressive images I specifically selected then color graded them.
(edit the result might not be great if you use these) prompt regularly and you'll get good results, be simple.
Here are some prompts you can try:
Horizontal:
a woman with a sword in her hand and a hairdow in her hair, with a large, feathered headpiece, looks into the distance, NIJIFTB
a painting of a rocky path with flowers and rocks on either side of it and a sky background with clouds and sun shining through the trees and flowers, NIJIFTB
a painting of a mountain with a lake in front of it and a forest in the background with a few clouds in the sky above it and a few trees, NIJIFTB
a painting of a mountain valley with flowers and trees in the foreground and a stream running through the valley below, with a mountain in the background, NIJIFTB
a painting of a mountain lake with trees and a house in the foreground and a snow capped mountain in the background with a blue sky and white, NIJIFTB
a painting of a man walking through a field with a mountain in the background and red flowers on the ground in the foreground, and a blue sky with clouds, NIJIFTB
a painting of a man riding a horse in a mountain valley with a stream running through it and a mountain range in the background with snow capped mountains, NIJIFTB
a painting of a landscape with a stream and trees in the background and a house on a hill in the distance with a blue sky and white clouds, NIJIFTB
a painting of a house in a field with flowers and a stream running through it, with a sky background and clouds in the sky above it, NIJIFTB
a painting of a couple of horses in a field with mountains in the background and a stream running through the grass and flowers in the foreground, NIJIFTB
a digital painting of a cat in the snow with snowflakes on it's fur and eyes, with a blurry background of snow, NIJIFTB
Vertical:
a cartoon character with a pink hair and blue eyes wearing a blue dress and pink earrings and a blue dress with pink flowers on it's shoulders, NIJIFTB
a cartoon character with pink hair and a cat ears on her head and a black cat outfit on her body and a pink collar and tail on her head, NIJIFTB
a doll with red hair and blue eyes wearing a green shirt and pink tie and pink pants and a pink and green jacket with a pink collar, NIJIFTB
a woman with blonde hair and blue dress sitting on a floor with her hand on her chin and her hand on her chin, looking to the side, NIJIFTB
a woman with blue hair and a blue jacket sitting on a chair with her hands on her face and her hands on her chin, her chin, NIJIFTB
a woman with blue hair and a flower shirt on, with her arms crossed and her eyes closed, is smiling at the camera, with a blue sky in the background, NIJIFTB
a woman with blue hair and a green shirt is sitting down and looking at the camera with her hand on her chin and her hand on her chin, NIJIFTB
a woman with green hair and a green jacket sitting on a stool with her hand on her chin and her hand on her cheek, with her chin resting her chin, NIJIFTB
a woman with long white hair and a cat ears hat is sitting on the floor with her hand on her chin and her hand on her chin, NIJIFTB
Description
FAQ
Comments (13)
This one is looking much better already :)
If anyone has any tips for generating models better id be happy to hear. Here's discord server: https://discord.gg/GCD7eQveK3
Putting a keyword / trigger word in a model seems weird.. It just makes it more confusing to use.
I can't think of any value then to have to remember it..
@DreamExplorer idk how this stuff works, thought using trigger words and keywords was the only way to get some sort of style into a model. apparently this model just works regardless. i started using dreambooth yesterday and i have no clue what im doing. well my previous model only worked when keywords and triggers were used which kinda make me confused.
@DreamExplorer also may you know the reason as to why when i try to merge the model with another or when i try to use a lora on top they dont work?
@aresisok191 Trigger words are good for applying certain themes within a dataset..
AKA If you made a Lora based on a set of characters from a game or anime, you could use a trigger to add bias towards the different characters..
Some people also like to add Triggers to LORA's so you can control in prompts when you apply / stop applying a LORA.
--
When you have a generic model that represents a single cohesive style though, adding a trigger word that "turns on" the entire model with no differentiation doesn't add much value. IMO.
Thanks for reading / replying. ♥
@DreamExplorer Yuh I found the reason why the LoRa's don't work, most are trained on stable diffusion 1.5, and same thing with merging models, merging with a 1.5 model doesn't work.
great model
look forward the 2.1 ver
looking for niji style in sd yesterday and this pop out
great model. but does anyone know why i keep getting this error msg when use with controlnet? RuntimeError: mat1 and mat2 shapes cannot be multiplied (154x1024 and 768x320)
I think the controlnet models were made for 1.5 stable diffusion models, and this model was made on the 2.1 stable diffusion model
Is there any solution@aresisok191
@fishwang no 2.1 and 1.5 are just not compatible with each other
XL, please.



















