deadman44's picture
Update README.md
81cbb94 verified
|
raw
history blame
19.6 kB
metadata
license: other
license_name: faipl-1.0-sd
license_link: https://freedevproject.org/faipl-1.0-sd/
language:
  - en
tags:
  - text-to-image
  - stable-diffusion
  - safetensors
  - stable-diffusion-xl

models



Zipang XL test3.1

-4000+ twitter images trained & 10000+ images merged model

Animagine XL 3.1 Base
photoreal like tag: shadow, flash lighting, backlighting, silhouette, sunset, night, day, bokeh, etc.

Download:test3.1

sample1
sample2
sample3
sample4
sample5
sample6
-refer to pnginfo

- base model

-Animagine XL 3.1 -great XL model!

- trigger

japanese
european
yo tag and myjs, myjc, myjk (e.g.:18yo, myjk)

- quality tags

masterpiece, best quality, very aesthetic, absurdres,

- negative tags

lowres, (bad), error, fewer, extra, missing, worst quality, low quality, extra digits

- sampler

DPM++ 2M SDE Heun Karras :24-28 steps
(Lightning)DPM++ 2M SGMUniform :8-14 steps
(Hyper-SD XL 8steps)DPM++ 2s a,DPM++ 2M SGM Uniform ;8-16 steps (Bad)

- cfg scale

3-5
(Lightning):1-3
(Hyper-SD XL):1

- Resolution

768 x1024, 1024 x 768
1024 x 1024
896 x 1152, 1152 x 896
512 x 768, 768 x 512 (hires.fix required)

Sample prompt

-SDXL Lightning (too fast: recommended)

masterpiece, best quality, very aesthetic, absurdres,
16yo, japanese, myjk, 1girl, solo, backlighting, silhouette, horizon, sunset, wind, water, outdoors, wading, scenery, white dress, photo background, skirt hold, realistic, long hair, standing, barefoot, from behind, facing away, close-up
Negative prompt: lowres, (bad), error, fewer, extra, missing, worst quality, low quality, extra digits

Steps: 24, Sampler: DPM++ 2M SDE Heun Karras, CFG scale: 3, Seed: 2770074115, Size: 896x1152, Model hash: 90bbe169ac, Model: zipang_XL_test3.1.fp16, Denoising strength: 0.3, Hires upscale: 2, Hires upscaler: 4x-UltraSharp, Version: f0.0.17v1.8.0rc-latest-276-g29be1da7


-train
base1:trained 4000+images:manual tagging (Prodigy:70epoch)
model:trained 10000+images:wd14 tagger(swinv2 tagger v3) (base model:base1)(Nadam:lr:1e-04:27epoch)


Zipang XL test03

-4000+ twitter images trained & 9000+ images merged model

I don't like the look a bit, but the controls are much improved.

Download:test03b (micro-adjustment version)
Download:test03

myJS LoRA (dedicated)
myJC LoRA (dedicated)
myJK LoRA (dedicated)

sample1
sample2
sample3
sample4
sample5
sample6
-refer to pnginfo

- base model

-Animagine XL 3.0 -great XL model!
-stable-diffusion-xl-base-1.0

- trigger

japanese
european

- quality tags

masterpiece, best quality, realistic, 

- negative tags

bad anatomy, worst quality, low quality

- sampler

DPM++ 2M SDE Heun Karras :24-28 steps
(LCM)DPM++ 2S a Karras :6-8 steps
(Lightning)DPM++ 2M SGMUniform :8-10 steps

- cfg scale

5-7 (LCM:3-5) (Lightning:1-1.3)

- Resolution

768 x1024, 1024 x 768
1024 x 1024
896 x 1152, 1152 x 896
512 x 768, 768 x 512 (hires.fix required)

sample prompt

masterpiece, best quality, natural lighting,
japanese, realistic, 1girl, solo, cute,
bangs, brown hair, ponytail, brown eyes, smile,
long white coat, zipper, snow fall,
v, looking at viewer, outdoors, photo background, upper body, front view
Negative prompt: bad anatomy, worst quality, low quality, nsfw
Steps: 28, Sampler: DPM++ 2M SDE Heun Karras, CFG scale: 3, Seed: 2206710877, Size: 896x1152, Model hash: 8a9978e263, Model: zipang_XL_test03.fp16, Version: v1.7.0

-LCM

masterpiece, best quality,
15yo, japanese, realistic, 1girl, solo, cute,
bangs, black hair, long hair, black eyes, smile,
cat ears, wristwatch, school uniform,
indoors, shop, looking at viewer
<lora:lcm-lora-sdxl:0.3>
Negative prompt: bad anatomy, worst quality, low quality
Steps: 8, Sampler: DPM++ 2S a Karras, CFG scale: 3, Seed: 246389822, Size: 896x1152, Model hash: 8a9978e263, Model: zipang_XL_test03.fp16, Lora hashes: "lcm-lora-sdxl: 2fa7e8e56b09", Version: v1.7.0


-SDXL Lightning (too fast: recommended)

masterpiece, best quality, cute, japanese,
14yo, 1girl, bangs, black hair, twintails, police uniform, tokyo street, wind, posing, smile, upper body
<lora:sdxl_lightning_8step_lora:1>
Negative prompt: bad anatomy, worst quality, low quality
Steps: 8, Sampler: DPM++ 2M SGMUniform, CFG scale: 1.3, Seed: 1747615192, Size: 896x1152, Model hash: 8a9978e263, Model: zipang_XL_test03.fp16, Version: f0.0.15v1.8.0rc-latest-229-g54c89503


-merge & train
base1:merge Animagine XL and stable-diffusion-xl-base-1.0
      mbw(0.1,1,1,1,1,0,0,0.1,0.1,0,0.3,0,1,0.5,1,0.5,1,1,1,1)
base2:trained 4000+images:manual tagging (base model:base1)(Adafactor:lr:1e-04)
model:trained 9000+images:wd14 tagger(swin v2) (base model:base2)(Adafactor:lr:1e-04)



zipang test03b

-micro-adjustment version
-natural face
-some of the prompts could work better.

test03b test03
sample1
sample1
sample1
sample1

-refer to pnginfo

my JS LoRA for test03

-natural face

Download

trigger:

myjs

original LoRA myjs
sample1
sample1
sample1
sample1

-refer to pnginfo


my JC LoRA for test03

-natural face

Download

trigger:

myjc

original LoRA myjc
sample1
sample1
sample1
sample1

-refer to pnginfo


Ponypang XL giveup (bad quality)

This is an incomplete model.

Mixed LoRA due to problems with training not being able to continue.
e621 tag might be better than danbooru tag.


-4000+ twitter images trained & 10000+ images merged model

-experimental
-not good. (bad quality, worst quality:2.0)
-I don't know how this happened.
-ADetailer and HiresFix Recommendation.

[Download](https://huggingface.co/deadman44/SDXL_Photoreal_Merged_Models/resolve/main/ponypang_XL_giveup.fp16.safetensors?download=true)
sample1
sample2
sample3
sample4
sample5
sample6
-refer to pnginfo

- base model

-Pony Diffusion V6 XL -great XL model!
-ebara pony 2.1 -great XL model! too

- trigger

japanese
european

- quality tags

score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up,

- negative tags

worst quality, low quality, normal quality, messy drawing, amateur drawing, lowres, bad anatomy, bad hands, source_furry, source_pony, source_cartoon, comic, source filmmaker, 3d


^Tested only with Hyper-SDXL-8steps-lora

- sampler

DPM++ SGM Uniform :8-16 steps

- cfg scale

1-1.5

- Resolution

768 x1024, 1024 x 768
1024 x 1024
896 x 1152, 1152 x 896
512 x 768, 768 x 512 (hires.fix required)

sample prompt

<lora:Hyper-SDXL-8steps-lora:1>
score_9, score_8_up, score_7_up, score_6_up, score_5_up, score_4_up,
15yo, myjc, japanese, 1girl, solo, realistic, long hair, squatting, black hair, plaid skirt, school uniform, smile, looking at viewer, loafers, bangs, indoors, black socks, kneehighs, white shirt, black eyes, school bag, long sleeves, pleated skirt, collared shirt, full body, teeth, window, grin, photorealistic, red skirt, hallway, looking at viewer, day, parted lips, classroom
Negative prompt: worst quality, low quality, normal quality, messy drawing, amateur drawing, lowres, bad anatomy, bad hands, source_furry, source_pony, source_cartoon, comic, source filmmaker, 3d
Steps: 8, Sampler: DPM++ 2M SGMUniform, CFG scale: 1, Seed: 368170729, Size: 896x1152, Model hash: 9a0a16a861, Model: ponypang_XL_giveup.fp16, Lora hashes: "Hyper-SDXL-8steps-lora: 4f494295edb1", Version: f0.0.17v1.8.0rc-latest-276-g29be1da7

-merge & train

base1:ebara pony 2.1 base Trained.
base2:base1 x ebara base LoRA (1,1,1,1,0.3,0.3,0.3,0.3,0.5,0.75,0,0,0,0.75,1,0.9,0.5,0.3,0.3,0.3,1,1,0,0,0,0)
base3:Merge Zipang test3.1 (0,0,0,0,0,0,0,0,0,0.5,0,0.3,0,0.1,0,0,0,0,0,0)
base4:base3 x pony base LoRA (0,0,0,0,0,0.1,0.25,0,0,0.7,0,0,0,0.5,0.25,1,0,0.1,0.75,0.65,0,0,0,0,0,0)