homer_26
Lora based on the model Niji Pony 3D Styles (TensorArt), I approached the 3D style through the Pony 🦄FaeTality in conjunction with the LoRA VividColor 2.5d [PonyXL] Style, bringing a cleaner style, with slightly strong lines and a reflection of light on the skin/hair of the art.
I mixed around 250 images to create LoRA in Civitai, making it more refined, especially in various poses, viewing angles, close-ups, shadow and light styles, male and female gender.
I decided to choose the Pony style because it was more versatile in terms of using tags, including a cartoon/realistic style. Among the other style options, I didn't reach a satisfactory conclusion, it was too cartoony or too realistic, so I decided to use Pony (in the future I intend to extend it to other SD 1.5 models, etc.).
Observation:
Use the tag hippo and/or hippopotamus
The tag slight chubby can sometimes make the character too obese, if you want to make a character not too fat or too thin, use the tag curvy figure
When using the tag close-up, especially on faces, it can make the character childish, if it is an adult character, in my tests that were successful, I used the tags ultra detailed, adult, mature
The tag tail had some errors in the hippo, such as a tail feline or reptile, so it is up to you to use the tail tag, in my tests, when I used the tail tag in the negative prompt, the model was doing the correct tail for a hippo
Use the tag 3D in the negative prompt so that the art does not look "plasticized" or in the "3D blender" style
Below is my default configuration in Civitai:
Model (base)
Pony 🦄FaeTality (v1.0)
LoRA (base)
VividColor 2.5d [PonyXL] Style (2.5d Realistic v1.0) - 0.4 ~ 0.8 (preferably 0.60)
Advanced
CFG Scale - 7.0 ~ 12.0 (preferably 7.0)
Sampler - Euler a ~ DPM++ 2M Karras (Any other type of sampler can have unexpected results)
Steps - 24 ~ 50 (preferably 24)
Clip Skip - 1 / 2 (preferably 1)
VAE - none
Prompt (Positive)
score_9, score_8_up, score_7_up, gray background, solo, solo_focus furry, anthro, female, hippo, nude, big breasts, wide hips, thick thighs, curvy figure, pose sexy, dynamic angle
Prompt (Negative)
zPDXL-neg, 3D, text, watermark, horns, tail
Here are the best results I've obtained using ComfyUI + DDetailer + Upscale:
Checkpoint
ckpt_name - ponyFaetality_v11.safetensors
CLIP Set Last Layer
stop_at_clip_layer - (-2)
LoraLoaderModelOnly
lora_name - 2.5dcolor-000009.safetensors
strength_model - 1.00
KSampler
steps - 25
cfg - 7.0
sampler_name - euler_ancestral / dpmpp_2m
scheduler - normal / karras
denoise - 1.00
Upscale Model
model_name - 4x_foolhardy_Remacri.pth
Latent Image
width - 1024
height - 1024
batch_size - 1
SAMLoader (Impact)
model_name - sam_vit_b_01ec64.pth
device_mode - auto
FaceDetailer (DDetailer)
guide_size - 512
guide_size_for - bbox
max_size - 1024
steps - 25
cfg - 7.0
sampler_name - euler_ancestral / dpmpp_2m
scheduler - normal / karras
denoise - 0.60
feather - 5
noise_mask - enabled
force_inpaint - enabled
bbox_threshold - 0.50
bbox_dilation - 20
bbox_crop_factor - 3.0
sam_detection_hint - center-1
sam_dilation - 0
sam_threshold - 0.93
sam_bbox_expansion - 0
sam_mask_hint_thresold - 0.70
sam_mask_hint_use_negative - False
drop_size - 10
Ultimate SD Upscale
upscale_by - 1.50
steps - 24
cfg - 7.0
sampler_name - euler_ancestral / dpmpp_2m
scheduler - normal / karras
denoise - 0.20
mode_type - Linear
tile_width - 1024
tile_height - 1024
mask_blur - 4
tile_padding - 32
seam_fix_mode - Band Pass
seam_fix_denoise - 1.00
seam_fix_width - 64
seam_fix_mask_blur - 8
seam_fix_padding - 16
force_uniform_tiles - true
Prompt (Positive)
score_9, gray background, solo, solo_focus furry, anthro, female, hippo, nude, big breasts, wide hips, thick thighs, curvy figure, pose sexy, dynamic angle Prompt (Negative)
zPDXL-neg, 3D, text, watermark, horns, tail
