|
--- |
|
license: openrail |
|
--- |
|
<h1>ITRobo2022 model. Trained on SD 1.5.</h1> |
|
|
|
 |
|
<br> |
|
I really like the Robo-Diffusion model (https://huggingface.co./nousr/robo-diffusion), but most of what you can get with it is robot heads. :)<br> |
|
In my model I tried to emphasize full-length images of robots. I also get good results on a homogeneous background, which makes it easier to cut out objects for further work.<br> |
|
However, good results are also obtained with mixed queries. Try it. Good luck! |
|
|
|
!!! For best result use this token at the beginning of the prompt: <b>itrobo2022</b><br><br> |
|
<b>itrobo2022.ckpt</b> - base trained model. It's a little hard to control the result, but good for generating a variety of robots, and for working with img2img.<br> |
|
<b>itrobo2022-40-with-v1-5-pruned-emaonly-60.ckpt</b> - 40% mixed with base SD1.5. Better manageability and control of results. |
|
|
|
<b>Example:</b><br> |
|
Prompt: <i>ITRobo2022 (a full body photo of pug)+, isolated, high resolution photo, cinematic lighting, trending on artstation, DOF, high resolution, 4 k, 8 k, solid background</i><br> |
|
Negative prompt: <i>(duplicate)+++, deformed, no leg, blurry, no head, headless, watermarks, writings, text, marks, ugly, a lot of fingers, mutation, too many legs</i><br> |
|
 |
|
<br> |
|
Prompt: <i>A realistic photograph of a 3d robot in a modern city. A glossy white and orange robot.</i><br> |
|
Negative prompt: <i>black and white robot, picture frame, a children's drawing in crayon. #Wholesale, Abstract Metal Sculpture. i'm leaving a bad review.</i><br> |
|
 |
|
|
|
Best results on:<br> |
|
DDIM<br> |
|
steps:20<br> |
|
CFG scale 7<br> |
|
512x512 |
|
|
|
 |
|
<br> |
|
img2img:<br> |
|
 |
|
<br> |
|
NSFW :)<br> |
|
 |
