Recently, the 150 years old art fair in Colorado chose Jason Allen’s work of art named “Théâtre D’opéra Spatial” for the first prize link to article . As per the news articles,” the judges chose the painting as it looked like as if it were a scene from a space opera where classical figures are staring at the beautiful landscape outside”. However, even though the painting is a master piece it’s win is not considered positively by the illustrators and artists. Reason is that this art work was generated based on a text to image generation, AI model.
I was fascinated by the art generation capabilities of AI and started playing around and spending time on this phonemical machine learning algorithm to know more about the newly minted “stable diffusion models”. Stable diffusion models belong to the category of generative deep learning models that can generate media from input textual prompt. What it means is that given text of my choice, I can get the computer to generate an image based on the textual prompt. Stable diffusion is much more efficient then the regular diffusion models as they provides inference on the latent space rather than the complete pixel space and hence has lower computational and time capacity requirements than the diffusion models.
I recently read the news of the lions in the Sydney zoo that escaped their enclosure only to be brought back in after much trouble newspaper article.. I was curious to see what image will be generated for the prompt” “lions in the streets of Sydney”. See the banner of the post to see what I got. As we can see the noise gets processed to carve out the image that I was looking for! This is the magic of stable diffusion that too without the need for massive GPU computing capacity!
Comments