Post by account_disabled on Mar 11, 2024 8:03:21 GMT
Today we often hear about DALL·E 2 from OpenAI, Imagen Muse, Midjourney, BlueWillow and AI Stable Diffusion. These are AI-based algorithms that are called Text-To-Image, and are capable of generating quality images starting from descriptive text in natural language. Alessio Pomaro Alessio Pomaro July 8, 2022 •14 min read From text to images through artificial intelligence... can we talk about art? From text to images through artificial intelligence... can we talk about art? In recent months, various algorithms based on artificial intelligence have been compared, which are defined as Text-To-Image because they are capable of transforming text ( formulated in natural language ) into an image.
The best known are OpenAI's DALL·E 2 , DALL·E mini ( an open-source project ), Google's India Mobile Number Data Imagen and Muse , Midjourney, BlueWillow and AI Stable Diffusion . What is DALL·E 2? DALL·E 2 is the new version of DALL·E, a generative language model that transforms simple sentences into images. It has 3.5 billion parameters and belongs to the category of LLM ( Large Language Model ) , although, for example, it is not as large as GPT-3 . It is curious that the model has smaller dimensions than its predecessor. Despite this, DALL·E 2 generates images with 4 times better resolution than DALL·E.
The following are images generated by asking the system.. "an astronaut riding a horse in a photorealistic style" An example of image generation by DALL·E 2 Another interesting aspect of DALL·E 2 is its ability to realistically edit and retouch photos . Users can select an area of the image, and use a text message to indicate the desired change. Within seconds, the algorithm produces different combinations of modified image. DALL·E 2 adds a sofa in the original image in position 1 DALL·E 2 adds a sofa in the original image in position 1 DALL·E 2 adds a sofa in the original image in position 2 DALL·E 2 adds a sofa in the original image in position 2 Note how the modified objects are placed with appropriate shadow and lighting .
The best known are OpenAI's DALL·E 2 , DALL·E mini ( an open-source project ), Google's India Mobile Number Data Imagen and Muse , Midjourney, BlueWillow and AI Stable Diffusion . What is DALL·E 2? DALL·E 2 is the new version of DALL·E, a generative language model that transforms simple sentences into images. It has 3.5 billion parameters and belongs to the category of LLM ( Large Language Model ) , although, for example, it is not as large as GPT-3 . It is curious that the model has smaller dimensions than its predecessor. Despite this, DALL·E 2 generates images with 4 times better resolution than DALL·E.
The following are images generated by asking the system.. "an astronaut riding a horse in a photorealistic style" An example of image generation by DALL·E 2 Another interesting aspect of DALL·E 2 is its ability to realistically edit and retouch photos . Users can select an area of the image, and use a text message to indicate the desired change. Within seconds, the algorithm produces different combinations of modified image. DALL·E 2 adds a sofa in the original image in position 1 DALL·E 2 adds a sofa in the original image in position 1 DALL·E 2 adds a sofa in the original image in position 2 DALL·E 2 adds a sofa in the original image in position 2 Note how the modified objects are placed with appropriate shadow and lighting .