Desde que inicie mi recorrido con los modelos de difusión siempre he tratado de obtener resultados increíbles y en la búsqueda de esos resultados he tenido aciertos y desaciertos. En especial con el modelo stable diffusion el cual es increíble pero obtener buenos resultados es bastante complicado. Y es que yo hago una comparativa entre dall-e que es muy fácil de usar y SD que es muy complicado con el violín y el piano. Dall-e vendría a ser el piano; un instrumento que ya viene afinado y con el cual es complicado que algo que toques suene mal y que de entrada es accesible para su aprendizaje. Por otro lado SD vendría siendo el violín; uno de los instrumentos más difíciles de tocar ya que tu lo tienes que afinar y requiere de muchas más habilidades para que suene bien.
Pues bien, dada la dificultad para crear buenos prompts me di a la tarea de buscar algo para facilitarnos (aún más) la vida y encontré una nueva IA en huggingface llamada MagicPrompt-Stable-Diffusion capaz de construir los mejores prompts para SD y los resultados dejan sin aliento.
Para comenzar cabe aclarar que ni siquiera necesitas una idea terminada ya que la IA es capaz de agregar elementos interesantes que puedes usar para tu prompt final. Yo comencé escribiendo "a dog" y me arrojó el prompt "A dog wearing headphones, 8k octane beautifully detailed render, post-processing, extremely hyperdetailed, intricate, epic composition, grim yet sparkling atmosphere, cinematic lighting + masterpiece, trending on artstation, very detailed, vibrant colors, Art Nouveau, masterpiece, full shot, full body, science fiction background" el cual me dio los siguientes resultados alucinantes.
Después escribí "an owl" y la IA terminó escribiendo un gran prompt con excelentes resultados el cual fue "a magic owl rising from the sea of blood, foggy background, wlop, concept art, digital painting, trending on artstation, highly detailed, epic composition, official media, 8 k uhd".
Y por último escribí "an eternal robot" y conseguí este prompt que me arrojó resultados que me dejaron sin aliento "an eternal robot mayan warrior, intricate, detailed, volumetric lighting, scenery, digital painting, highly detailed, artstation, sharp focus, illustration, concept art, ruan jia, steve mccurry".
Y hasta aquí el blog de hoy, si les gustó este contenido regálenme un voto arriba y un rehive para seguir trayendo contenido sobre IAs, hasta el próximo blog.
Since I started my journey with diffusion models I have always tried to obtain incredible results and in the search for those results I have had successes and failures. Especially with the stable diffusion model which is incredible but obtaining good results is quite complicated. I make a comparison between dall-e which is very easy to use and SD which is very complicated with the violin and the piano. Dall-e would be the piano; an instrument that comes already tuned and with which it is complicated that something you play sounds bad and that is accessible to learn from the beginning. On the other hand SD would be the violin; one of the most difficult instruments to play since you have to tune it and it requires a lot more skills to make it sound good.
Well, given the difficulty to create good prompts I decided to look for something to make our life easier and I found a new AI in huggingface called MagicPrompt-Stable-Diffusion capable of building the best prompts for SD and the results are breathtaking.
To start with, you don't even need a finished idea as the AI is able to add interesting elements that you can use for your final prompt. I started by typing "a dog" and I got the prompt "A dog wearing headphones, 8k octane beautifully detailed render, post-processing, extremely hyperdetailed, intricate, epic composition, grim yet sparkling atmosphere, cinematic lighting + masterpiece, trending on artstation, very detailed, vibrant colors, Art Nouveau, masterpiece, full shot, full body, science fiction background" which gave me the following amazing results.
Then I typed "an owl" and the AI ended up typing a great prompt with excellent results which was "a magic owl rising from the sea of blood, foggy background, wlop, concept art, digital painting, trending on artstation, highly detailed, epic composition, official media, 8 k uhd".
And finally I typed "an eternal robot" and got this prompt which gave me breathtaking results "an eternal robot mayan warrior, intricate, detailed, volumetric lighting, scenery, digital painting, highly detailed, artstation, sharp focus, illustration, concept art, ruan jia, steve mccurry".
And so far today's blog, if you liked this content give me a vote above and a rehive to continue bringing content about IAs, until the next blog.
Translated with www.DeepL.com/Translator (free version)
Imagen hecha por 
