Hola amigos, pues después de estar experimentando con SD v 2.0 por fin puedo mostrar algunas imágenes decentes. Y es que en esta nueva versión por lo que me he dado cuenta es que se priorizan los prompts negativos ¿pero que son exactamente? Bueno son lo opuesto a los prompts; mientras que un prompt indica que quieres en la imagen, el prompt negativo indica que no quieres en la imagen. Esto da como resultado imágenes filtradas con los errores típicos como mala anatomía, que este mal dibujado etc.
Aun así los resultados no son tan visualmente atractivos como los de su antigua versión porque se han filtrado artistas y muchas cosas del dataset pero aun así creo que los resultados no son tan malo como esperaba. Pero bueno, para este experimento usaré una web llamada pormpthero donde buscaremos imágenes lindas que tengan prompts negativos para experimentar con SD v 2.0 en un [colab](https://colab.research.google.com/github/qunash/stable-diffusion-2-gui/blob/main/stable_diffusion_2_0.ipynb#scrollTo=gId0-asCBVwL) (ya que es más rápido y gratis).
La primer imagen que me ha gustado ha sido esta y vienen con su respectivo prompt y prompt negativo.
[Fuente](https://prompthero.com/prompt/08c45c7b755)
Prompt: The personification of the Halloween holiday in the form of a cute girl with short hair and a villain's smile, (((cute girl)))cute hats, cute cheeks, unreal engine, highly detailed, artgerm digital illustration, woo tooth, studio ghibli, deviantart, sharp focus, artstation, by Alexei Vinogradov bakery, sweets, emerald eyes
Prompt negativo: bad anatomy, extra legs, extra arms, poorly drawn face, poorly drawn hands, poorly drawn feet, fat, disfigured, out of frame, long neck, poo art, bad hands, bad art, deformed, gun, double head, flowers,asian,hyperrealistic,child
Y después de meter todo en SD v 2.0 obtenemos los siguientes resultados que no están tan mal aunque no es lo que pedí:
La siguiente imagen que he escogido para la prueba ha sido esta:
[Fuente](https://prompthero.com/prompt/49e5598346a)
Prompt: complex 3d render ultra detailed of a beautiful porcelain profile woman android face, cyborg, robotic parts, 150 mm, beautiful studio soft light, rim light, vibrant details, luxurious cyberpunk, lace, hyperrealistic, anatomical, facial muscles, cable electric wires, microchip, elegant, beautiful background, octane render, H. R. Giger style, 8k
Negativo: poor quality resolution, incoherent, poorly drawn, poorly drawn lines, low quality, messy drawing, poorly-drawn, poorly-drawn lines, bad resolution, deformed, disfigured, disjointed, asymmetrical face, cross-eyed
Y lo que obtuve:
Y por último escogí esta:
[Fuente](https://prompthero.com/prompt/eb21a9e8cf8)
Prompt:Pixar style little girl, 4k, 8k, unreal engine, octane render photorealistic by cosmicwonder, hdr, photography by cosmicwonder, high definition, symmetrical face, volumetric lighting, dusty haze, photo, octane render, 24mm, 4k, 24mm, DSLR, high quality, 60 fps, ultra realistic
Negativo: black and white, blur, blurry, soft, blush, filter, noise, deformed, defective, incoherent, twisted, extra limbs, extra fingers, poorly drawn hands, messy drawing
Y mis resultados:
Y bueno, pues eso es todo por hoy, espero que les haya gustado, hasta un próximo blog.
---
# English
Hello friends, after experimenting with SD v 2.0 I can finally show some decent images. And the thing is that in this new version, what I have noticed is that the negative prompts are prioritized, but what exactly are they? Well they are the opposite of prompts; while a prompt indicates what you want in the image, a negative prompt indicates what you don't want in the image. This results in filtered images with the typical errors such as bad anatomy, poorly drawn etc.
Still the results are not as visually appealing as the old version because they have filtered out artists and many things from the dataset but still I think the results are not as bad as I expected. But well, for this experiment I will use a website called pormpthero where we will look for nice images that have negative prompts to experiment with SD v 2.0 in a [colab](https://colab.research.google.com/github/qunash/stable-diffusion-2-gui/blob/main/stable_diffusion_2_0.ipynb#scrollTo=gId0-asCBVwL) (since it's faster and free).
The first image I liked was this one and it comes with its respective prompt and negative prompt.
[Source](https://prompthero.com/prompt/08c45c7b755)
Prompt: The personification of the Halloween holiday in the form of a cute girl with short hair and a villain's smile, ((((cute girl))))cute hats, cute cheeks, unreal engine, highly detailed, artgerm digital illustration, woo tooth, studio ghibli, deviantart, sharp focus, artstation, by Alexei Vinogradov bakery, sweets, emerald eyes
Prompt negative: bad anatomy, extra legs, extra arms, poorly drawn face, poorly drawn hands, poorly drawn feet, fat, disfigured, out of frame, long neck, poo art, bad hands, bad art, deformed, gun, double head, flowers,asian,hyperrealistic,child
And after putting everything in SD v 2.0 we get the following results which are not too bad although not what I asked for:
The next image I chose for the test was this one:
[Source](https://prompthero.com/prompt/49e5598346a)
Prompt: complex 3d render ultra detailed of a beautiful porcelain profile woman android face, cyborg, robotic parts, 150 mm, beautiful studio soft light, rim light, vibrant details, luxurious cyberpunk, lace, hyperrealistic, anatomical, facial muscles, cable electric wires, microchip, elegant, beautiful background, octane render, H. R. Giger style, 8k
Negatives: poor quality resolution, incoherent, poorly drawn, poorly drawn lines, low quality, messy drawing, poorly-drawn, poorly-drawn lines, poorly resolution, deformed, disfigured, disjointed, asymmetrical face, cross-eyed
And what I got:
And finally I chose this one:
[Source](https://prompthero.com/prompt/eb21a9e8cf8)
Prompt:Pixar style little girl, 4k, 8k, unreal engine, octane render photorealistic by cosmicwonder, hdr, photography by cosmicwonder, high definition, symmetrical face, volumetric lighting, dusty haze, photo, octane render, 24mm, 4k, 24mm, DSLR, high quality, 60 fps, ultra realistic
Negatives: black and white, blur, blurry, soft, blush, filter, noise, deformed, defective, incoherent, twisted, extra limbs, extra fingers, poorly drawn hands, messy drawing
And my results:
And well, that's all for today, I hope you liked it, see you in the next blog.
Translated with www.DeepL.com/Translator (free version)
Imagen hecha por @fclore22
hiveblocks