Lora for generating Draft Horses in various poses, styles, view angles, skin colors and genders (male and female).
I trained in various configurations, below are the best results in Civitai Generator:
Model (base)
FurryBlend (FurryBlendv1.0)
(The point here is that I trained this LoRA in Civitai based on a Pony model in order to have greater diversity, even though it was based on SD 1.5.)
LoRA (base)
FurryDalle (v1.0) - 0.8 ~ 1.00 (preferably 1.00)
Embedding
Boring_e621 Negative Embedding: Enhance Images Stylistically AND Topically (v4.0)
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 - 25 ~ 50 (preferably 25)
Clip Skip - 1 / 2 (preferably 1)
VAE - vae-ft-ema-560000-ema
Prompt (Positive)
farm background, anthro, furry, (draft horse), solo, solo focus, female, anthro, (gray body), hair, nude, big breasts, equine pussy, puffy anus, wide hips, thick thighs, dynamic posePrompt (Negative)
deformityv6, boring_e621_fluffyrock_v4, bad-hands-v5, low quality, watermark, text, signature, blur, jpeg artifacts, compression artifacts, low resolution, symmetrical, extra legs, extra fingers, extra arms, extra ears, bad anatomy, humanHere are the best results I've obtained using ComfyUI + DDetailer + Upscale:
Checkpoint
ckpt_name - furryblend_v10.safetensors
CLIP Set Last Layer
stop_at_clip_layer - (-1)
LoraLoaderModelOnly
lora_name - DalleStyleTest-08.safetensors
strength_model - 1.00
KSampler
steps - 25
cfg - 7.0
sampler_name - dpmpp_2m
scheduler - 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 - 24
cfg - 7.0
sampler_name - dpmpp_2m
scheduler - 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 - dpmpp_2m
scheduler - 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)
farm background, anthro, furry, (draft horse), solo, solo focus, female, anthro, (gray body), hair, nude, big breasts, equine pussy, puffy anus, wide hips, thick thighs, dynamic posePrompt (Negative)
deformityv6, boring_e621_fluffyrock_v4, bad-hands-v5, low quality, watermark, text, signature, blur, jpeg artifacts, compression artifacts, low resolution, symmetrical, extra legs, extra fingers, extra arms, extra ears, bad anatomy, humanDescription
FAQ
Comments (14)
Thanks, but I don;t have SDXL
Seriously, I'm still stuck on a TI-84.
@Big_Soda Oh well
hey have take request lora you creatinng plz?????
Depending on the type I can try to create, I have more affinity at the prompt with furrys than with humans
@homer_26 when promp try creating for sex kindred female x wolf male sex both (both for league of legends)a for lora try creating lora for artist amocin or bbmbbf (but preffer second)
@fog1 It's going to be an interesting challenge, I'm going to try to create this Lora
@homer_26 ok tagme when end
Hey @fog1 Take a look to see how this LoRA turned out, if possible send feedback. This is the first time I've trained a LoRA based on images
@homer_26 ok I see that despite the potential of lora style this needs a lot of fine tuning and I still know the images to add there because you can see the missing parts and small artifacts know for now I will wait for a newer version of this lora but thanks for the chenci.
@homer_26 baytheway not problem more request?
@fog1 unfortunately for this type of LoRA (external images) I need a very powerful personal computer, because it requires a very large image bank for the images to be the same as the author's, I depend on Civitai, so it is very difficult for the results to be the same in the author's style
@homer_26 when need more image for autor why nor use page rule34
@homer_26 plus have always use civitai when need better pc a return request i request you creating bakugan wavern anthro female for sdxl