import torch
from diffusers import DiffusionPipeline
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("xiaolxl/GuoFeng3", torch_dtype=torch.bfloat16, device_map="cuda")
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt).images[0]
ๅบไบSDXL็ๅฝ้ฃ4ๅทฒๅๅธ๏ผ- GuoFeng4 based on SDXL has been released! : https://huggingface.co/xiaolxl/GuoFeng4_XL
ๆฌไบบ้้ๅฃฐๆ๏ผๆฌๆจกๅ็ฆๆญข็จไบ่ฎญ็ปๅบไบๆๆใๅ ฌไผไบบ็ฉ่ๅ็้ฃๆ ผๆจกๅ่ฎญ็ป๏ผๅ ไธบ่ฟไผๅธฆๆฅไบ่ฎฎ๏ผๅฏนAI็คพๅบ็ๅๅฑ้ ๆไธ่ฏ็่ด้ขๅฝฑๅใ
ๆฌๆจกๅๆณจๆ๏ผ่ฎญ็ป็ด ๆไธญไธๅ ๅซไปปไฝ็ไบบ็ด ๆใ
| ็ๆฌ | ๆๆๅพ |
|---|---|
| GuoFeng3.4 |  |
| GuoFeng3.3 |  |
| GuoFeng3.2_light |  |
| GuoFeng3.2 |  |
| GuoFeng3 |  |
ไป็ป - GuoFeng3
ๆฌข่ฟไฝฟ็จGuoFeng3ๆจกๅ - (TIP:่ฟไธช็ๆฌ็ๅๅญ่ฟ่กไบๅพฎ่ฐ),่ฟๆฏไธไธชไธญๅฝๅไธฝๅค้ฃ้ฃๆ ผๆจกๅ๏ผไนๅฏไปฅ่ฏดๆฏไธไธชๅค้ฃๆธธๆ่ง่ฒๆจกๅ๏ผๅ ทๆ2.5D็่ดจๆใ็ฌฌไธไปฃๅคงๅน ๅบฆๅๅฐไธๆ้พๅบฆ๏ผๅขๅ ไบๅบๆฏๅ ็ด ไธ็ทๆงๅค้ฃไบบ็ฉ๏ผ้คๆญคไนๅคไธบไบๆจกๅ่ฝๆดๅฅฝ็้ๅบๅ ถๅฎTAG๏ผ่ฟๅขๅ ไบๅ ถๅฎ้ฃๆ ผ็ๅ ็ด ใ่ฟไธไปฃๅฏน่ธๅๆ็ๅดฉๅๆไธๅฎ็ไฟฎๅค๏ผๅๆถ็ด ๆๅคงๅฐไนๆ้ซๅฐไบๆ้ฟ่พน1024ใ
ๆ นๆฎไธชไบบ็ๅฎ้ชไธๆถๅฐ็ๅ้ฆ๏ผๅฝ้ฃๆจกๅ็ณปๅ็็ฌฌไบไปฃ๏ผๅจไบบ็ฉ๏ผไธๅคงๅคด็ ง็ๆๆ่กจ็ฐๆฏไธไปฃๆดๅฅฝ๏ผๅฆๆไฝ ๆ่ฟๆน้ข้ๆฑไธๅฆจ่ฏ่ฏ็ฌฌไบไปฃใ
2.0็ๆฌ๏ผhttps://huggingface.co/xiaolxl/Gf_style2
GuoFeng3:ๅๅงๆจกๅ
GuoFeng3.1:ๅฏนGuoFeng3ไบบๅ่ฟ่กไบๅพฎ่ฐไฟฎๅค
GuoFeng3.2:ๅฆๆไฝ ไธ็ฅ้้ๆฉGuoFeng3่ฟๆฏGuoFeng2๏ผๅฏไปฅ็ดๆฅไฝฟ็จๆญค็ๆฌ
GuoFeng3.2_light:้่ฟGuoFeng3.2่ๅไบๅบไบ Noise Offset ่ฎญ็ป็Loraไฝฟๅพๆจกๅ่ฝๅค็ปๅบๆดๆผไบฎ็ๅ ๅฝฑๆๆ(Lora:epi_noiseoffset/Theovercomer8's Contrast Fix)
GuoFeng3.2_Lora:ๅฝ้ฃ3.2 Lora็ๆฌ
GuoFeng3.2_Lora_big_light:ๅฝ้ฃ3.2_light Lora็ๆฌ ็ปดๅบฆๅขๅคง็ๆฌ
GuoFeng3.2_f16:ๅฝ้ฃ3.2 ๅ็ฒพ็ๆฌ
GuoFeng3.2_light_f16:ๅฝ้ฃ3.2_light ๅ็ฒพ็ๆฌ
GuoFeng3.3๏ผๆญค็ๆฌๆฏๅบไบ3.2็ไธๆฌก่พๅคง็ๆดๆฐไธๆน่ฟ๏ผๅฏไปฅ้้ full body๏ผๅณไฝฟไฝ ็tagไธๅคชๅฅฝ๏ผๆจกๅไนไผๅฏน็ป้ข่ฟ่ก่ชๅจไฟฎๆน๏ผไธ่ฟๅ ๆญคๆจกๅๅบ็่ธไผๆฏ่พ้ทๅใๆญคๆจกๅไผผไนไธ้่ฆ่ถ ๅ๏ผๆ็ๅบๅพๅคงๅฐๆฏ768*1024๏ผๆธ ๆฐๅบฆ่ฟไธ้ใๅปบ่ฎฎ็ซๅพ๏ผๆจชๅพๅฏ่ฝไธๆธ ๆฐใEuler aๅณๅฏใ(DPM++ SDE Karras, DDIMไนไธ้)
GuoFeng3.4:ๆญค็ๆฌ้ๆฐ่ฟ่กไบๆฐ็่ฎญ็ป๏ผ้้ ๅ จ่บซๅพ๏ผๅๆถๅ ๅฎนไธไธๅๅ ไธช็ๆฌๆ่พๅคงไธๅใๅนถ่ฐๆดไบๆดไฝ็ป้ฃ๏ผ้ไฝไบ่ฟๆๅ็จๅบฆ๏ผไฝฟๅ ถ่ฝไฝฟ็จๆดๅค็loraๅฏน็ป้ขไธๅ ๅฎน่ฟ่ก่ฐๆดใ
--
Welcome to the GuoFeng3 model - (TIP: the name of this version has been fine-tuned). This is a Chinese gorgeous antique style model, which can also be said to be an antique game character model with a 2.5D texture. The third generation greatly reduces the difficulty of getting started, and adds scene elements and male antique characters. In addition, in order to better adapt the model to other TAGs, other style elements are also added. This generation has repaired the broken face and hands to a certain extent, and the size of the material has also increased to the longest side of 1024.
According to personal experiments and feedback received, the second generation of the Guofeng model series performs better than the third generation in terms of characters and big head photos. If you have this need, you can try the second generation.
Version 2.0๏ผhttps://huggingface.co/xiaolxl/Gf_style2
GuoFeng3: original model
GuoFeng3.1: The portrait of GuoFeng3 has been fine-tuned and repaired
GuoFeng3.2: If you don't know whether to choose GuoFeng3 or GuoFeng2, you can use this version directly
GuoFeng3.2_Light: Through GuoFeng3.2, Lora based on Noise Offset training is integrated to enable the model to draw more beautiful light and shadow effects (Lora: epi_noiseoffset/Theovercolor8's Contrast Fix)
GuoFeng3.2_Lora: Guofeng3.2 Lora version
GuoFeng3.2_Lora_big_Light: Guofeng3.2_Light Lora Version Dimension Increase Version
GuoFeng3.2_F16: Guofeng3.2 semi-refined version
GuoFeng3.2_light_f16: Guofeng3.2_Light semi-refined version
GuoFeng3.3: This version is a major update and improvement based on 3.2, which can adapt to full bodies. Even if your tag is not good, the model will automatically modify the screen, but the faces produced by the model will be quite similar. This model doesn't seem to require supersession. My plot size is 768 * 1024, and the clarity is quite good. Suggest vertical view, horizontal view may not be clear. Euler a is sufficient. (DPM++SDE Karras, DDIM is also good)
GuoFeng3.4: This version has undergone new training to adapt to the full body image, and the content is significantly different from previous versions.At the same time, the overall painting style has been adjusted, reducing the degree of overfitting, allowing it to use more Lora to adjust the screen and content.
ๅฎ่ฃ ๆ็จ - install
ๅฐGuoFeng3.ckptๆจกๅๆพๅ ฅSD็ฎๅฝ - Put GuoFeng3.ckpt model into SD directory
ๆญคๆจกๅ่ชๅธฆVAE๏ผๅฆๆไฝ ็็จๅบไธๆฏๆ๏ผ่ฏท่ฎฐๅพ้ๆฉไปปๆไธไธชVAEๆไปถ๏ผๅฆๅๅพๅฝขๅฐไธบ็ฐ่ฒ - This model comes with VAE. If your program does not support it, please remember to select any VAE file, otherwise the graphics will be gray
ๅฆไฝไฝฟ็จ - How to use
TIP๏ผ็ป่ฟไธๅคฉ็ๆต่ฏ๏ผๅ็ฐๅพๅคไบบ็ฉๅฏ่ฝๅบ็ฐ็บข็ผ้ฎ้ข๏ผๅฏไปฅๅฐ่ฏๅจ่ด้ข่ฏๆทปๅ red eyesใๅฆๆ่ฒๅฝฉ่ณไธฝๅฏไปฅๅฐ่ฏ้ไฝCFG - After a day of testing, we found that many characters may have red-eye problems. We can try to add red eyes to negative wordsใTry to reduce CFG if the color is bright
็ฎๅ๏ผ็ฌฌไธไปฃๅคงๅน ๅบฆๅๅฐไธๆ้พๅบฆ - Simple: the third generation greatly reduces the difficulty of getting started
======
ๅฆๆไฝ ็ๅบๅพๅ จ่บซๅพๆถๅบ็ฐ่ธ้จๅดฉๅๅปบ่ฎฎๅ ้คfull bodyๅ ณ้ฎ่ฏๆ่ ไฝฟ็จ่ธ้จ่ชๅจไฟฎๅคๆไปถ๏ผ
ๅฝๅคๆบๅฐๅ๏ผhttps://github.com/ototadana/sd-face-editor.git
ๅฝๅ ๅ ้ๅฐๅ๏ผhttps://jihulab.com/xiaolxl_pub/sd-face-editor.git
If you experience facial collapse during the full body image, it is recommended to delete the full body keyword or use the facial automatic repair plugin:
Foreign source address: https://github.com/ototadana/sd-face-editor.git
Domestic acceleration address: https://jihulab.com/xiaolxl_pub/sd-face-editor.git
=====
- ๅ ณ้ฎ่ฏ - key word:
best quality, masterpiece, highres, 1girl,china dress,Beautiful face
- ่ด้ข่ฏ - Negative words:
NSFW, lowres,bad anatomy,bad hands, text, error, missing fingers,extra digit, fewer digits, cropped, worstquality, low quality, normal quality,jpegartifacts,signature, watermark, username,blurry,bad feet
้ซ็บง๏ผๅฆๆๆจ่ฟๆณไฝฟๅพ็ๅฐฝๅฏ่ฝๆดๅฅฝ๏ผ่ฏทๅฐ่ฏไปฅไธ้ ็ฝฎ - senior๏ผIf you also want to make the picture as better as possible, please try the following configuration
Sampling steps:50
Sampler:DPM++ SDE Karras or DDIM
The size of the picture should be at least 1024 - ๅพ็ๅคงๅฐ่ณๅฐ1024
CFG:4-6
ๆดๅฅฝ็่ด้ข่ฏ Better negative words - ๆ่ฐข็พคๅๆไพ็่ด้ข่ฏ:
(((simple background))),monochrome ,lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, lowres, bad anatomy, bad hands, text, error, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, ugly,pregnant,vore,duplicate,morbid,mut ilated,tran nsexual, hermaphrodite,long neck,mutated hands,poorly drawn hands,poorly drawn face,mutation,deformed,blurry,bad anatomy,bad proportions,malformed limbs,extra limbs,cloned face,disfigured,gross proportions, (((missing arms))),(((missing legs))), (((extra arms))),(((extra legs))),pubic hair, plump,bad legs,error legs,username,blurry,bad feet
- ๅฆๆๆณๅ ็ด ๆดไธฐๅฏ๏ผๅฏไปฅๆทปๅ ไธๆนๅ ณ้ฎ่ฏ - If you want to enrich the elements, you can add the following keywords
Beautiful face,
hair ornament, solo,looking at viewer,smile,closed mouth,lips
china dress,dress,hair ornament, necklace, jewelry, long hair, earrings, chinese clothes,
architecture,east asian architecture,building,outdoors,rooftop,city,cityscape
ไพๅพ - Examples
(ๅฏๅจๆไปถๅ่กจไธญๆพๅฐๅๅพ๏ผๅนถๆพๅ ฅWebUiๆฅ็ๅ ณ้ฎ่ฏ็ญไฟกๆฏ) - (You can find the original image in the file list, and put WebUi to view keywords and other information)
- Downloads last month
- 3,574