karla-photo-sd

Karla photo sd

MykosX/karla-photo-sd is a Stable Diffusion model that can be used both for:

• text-to-image: generates quite good photo images, may generate bad faces for people far from viewer
• image-to-image: tends to improve the quality of images generated by this model, does a good work on images from other models

Image show-case

<table> <tr> <th></th> <th>(seed=300)**</th> <th>(seed=400)**</th> <th>(seed=500)**</th> </tr> <tr> <td> text-to-image </td> <td> <img src="images/couple-having-fun-(generated)-300.jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(generated)-400.jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(generated)-500.jpg" width="600"/> </td> </tr> <tr> <td> image-to-image </td> <td> <img src="images/couple-having-fun-(from-generated)-300.jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(from-generated)-400.jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(from-generated)-500.jpg" width="600"/> </td> </tr> </table> <table> <tr> <th></th> <th>(seed=300)**</th> <th>(seed=400)**</th> <th>(seed=500)**</th> </tr> <tr> <td> text-to-image </td> <td> <img src="images/girl-posing-photo-(generated)-300.jpg" width="400"/> </td> <td> <img src="images/girl-posing-photo-(generated)-400.jpg" width="400"/> </td> <td> <img src="images/girl-posing-photo-(generated)-500.jpg" width="400"/> </td> </tr> <tr> <td> image-to-image </td> <td> <img src="images/girl-posing-photo-(from-generated)-300.jpg" width="400"/> </td> <td> <img src="images/girl-posing-photo-(from-generated)-400.jpg" width="400"/> </td> <td> <img src="images/girl-posing-photo-(from-generated)-500.jpg" width="400"/> </td> </tr> </table> <table> <tr> <th>Base image (from another model)</th> <th>image-to-image (seed=300)**</th> <th>image-to-image (seed=400)**</th> <th>image-to-image (seed=500)**</th> </tr> <tr> <td> <img src="images/couple-having-fun-(original).jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(from-original)-300.jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(from-original)-400.jpg" width="600"/> </td> <td> <img src="images/couple-having-fun-(from-original)-500.jpg" width="600"/> </td> </tr> <tr> <td> <img src="images/girl-posing-photo-(original).jpg" width="600"/> </td> <td> <img src="images/girl-posing-photo-(from-original)-300.jpg" width="600"/> </td> <td> <img src="images/girl-posing-photo-(from-original)-400.jpg" width="600"/> </td> <td> <img src="images/girl-posing-photo-(from-original)-500.jpg" width="600"/> </td> </tr> <tr> <th>Base image (from another model)</th> <th>image-to-image (seed=300)*</th> <th>image-to-image (seed=400)*</th> <th>image-to-image (seed=500)*</th> </tr> </table>

** using these defaults unless specified:

<table> <tr> <th>Setting</th> <th>Default value</th> </tr> <tr> <td>prompt (landscape)</td> <td>landscape image, a boy and girl having fun on the beach</td> </tr> <tr> <td>prompt (portrait)</td> <td>portrait image, a girl in a nice dress posing for a photo</td> </tr> <tr> <td>negative prompt</td> <td>deformed iris, deformed pupils, bad anatomy, cloned face, extra arms, extra legs, missing fingers, too many fingers</td> </tr> <tr> <td>size (landscape)</td> <td>1024 x 768</td> </tr> <tr> <td>size (portrait)</td> <td> 768 x 1024</td> </tr> <tr> <td>seed</td> <td>300</td> </tr> <tr> <td>guidance scale</td> <td>12.0</td> </tr> <tr> <td>strength</td> <td>0.5</td> </tr> <tr> <td>inference steps</td> <td>30</td> </tr> </table>

Diffusers

For more general information on how to run text-to-image models with 🧨 Diffusers, see the docs.

1. Installation

pip install diffusers transformers accelerate

1. Running example for text-to-image generation

import torch

from   diffusers import AutoPipelineForText2Image

pipe = AutoPipelineForText2Image.from_pretrained('MykosX/karla-photo-sd', torch_dtype=torch.float32)
pipe = pipe.to("cpu")

prompt = "portrait image, a girl in a nice dress posing for a photo"

image = pipe(prompt).images[0]  
image.save("./images/text-to-image.png")

1. Running example for image-to-image generation

import torch

from   diffusers import AutoPipelineForImage2Image
from   PIL       import Image

pipe = AutoPipelineForImage2Image.from_pretrained('MykosX/karla-photo-sd', torch_dtype=torch.float32)
pipe = pipe.to("cpu")

base_image = Image.open("./images/girl-posing-photo-(original).jpg")
prompt = "portrait image, a girl in a nice dress posing for a photo"

image = pipe(prompt, image=base_image).images[0]  
image.save("./images/image-to-image.png")

PS

Play with the model and don't hesitate to show off