DALLE 2 is the name of OpenAI’s most recent AI model. If you’ve seen some of its works and believe they’re incredible, stay reading to see why you’re both completely correct and incorrect.
A blog article and a paper titled “Hierarchical Text-Conditional Image Generation with CLIP Latents” from OpenAI were published on DALLE 2. The paper and the post are both useful for learning the technical details and viewing the results, but neither clearly conveys why DALLE 2 is so good in simpler terms.
What is DALLE 2?
The updated version of DALLE, a generative language model that creates unique graphics from sentences, is called DALLE 2. DALLE 2 is a huge model with 3.5B parameters, but it is interestingly smaller than GPT-3 and not quite as large as its predecessor (12B). Despite its size, DALLE 2 produces photos with a 4x higher resolution than DALLE, and human judges favor it +70% of the time for caption matching and photorealism.
OpenAI did not release DALLE 2 as they did with DALLE (you can always be on the waiting list that never ends). They did, however, open-source CLIP, which, albeit only loosely associated with DALLE, served as the foundation for DALLE 2. (CLIP is also the foundation for the apps and notebooks that DALLE 2 users who are unable to access it use.) The DALLE models are currently only available to a select few (they’re opening the model to 1000 people each week), but OpenAI’s CEO Sam Altman promised that they would eventually make them available through their API.
How DALLE 2 works?
- Text/image embeddings are created via the CLIP model using image-caption pairs to provide “mental” representations in the form of vectors.
- Old model: creates CLIP picture embeddings using a caption or CLIP text embedding.
- The Decoder Diffusion Model (unCLIP) creates images using a CLIP image embedding.
- DALLE 2: Prior + diffusion decoder model combination (unCLIP).
DALLE 2 is a specific example of a previous and a decoder-based two-part model (figure 1, bottom). We can convert a statement into an image by concatenating both models. That is the way we communicate with DALLE. The “black box” accepts a text as input and returns a precise image.
Notably, the decoder is termed unCLIP since it performs the opposite process of the original CLIP model, producing an original image from a general mental representation instead of establishing a “mental” representation (embedding) from an image.
People, animals, objects, styles, colors, backgrounds, etc. are the major qualities that are semantically significant and are encoded in the mental representation so that DALLE 2 can create a new image while keeping these characteristics while changing the non-essential features.
DALLE 2 Examples
1. Stage, scenic, and set design
DALLE 2 can design stage sets based on themes, artists, or other criteria that you provide. A good prompt might be something like: “Award-winning scenic design for “PROMPT!
Prompts:
The set design of Game Of Thrones
2. Creating theatrical costumes
What about the costumes now that the set is ready? Recruit DALLE to create costumes. Try once more to use language like “Award-winning costume design for “PROMPT!”
Prompt:
Award-winning costume design from the press release for the Broadway production of “ABANDON SHIP!”; the costume is based on an orange survival suit.
3. Nail art, cosmetics, and hairstyles
Ask DALLE 2 for pictures of hairstyles, makeup looks, tattoos, or other body art to help bring them to life as you get closer. (Remember that sharing photos with actual faces is prohibited.)
Prompt:
Macro photography of intricate nail art inspired by Mandlebrot Set fractals.
Airbrush
Airbrush.ai is a revolutionary AI technology enabling users to create original stock photos, NFTs, art, and more in seconds. With Airbrush, users can save time and money by eliminating the need for photoshoots and allowing them to generate high-quality images for every use-case. Airbrush gives a wide range of DALLE 2 visuals which are available for use in various projects, including commercials, websites, and presentations.
Airbrush makes it easy for you to choose from a variety of price possibilities to find the perfect image for your project. You may also store your favorite photos for easy and quick access and search for photos using tags or keywords. You only need to sign up for a free account to use it. Therefore, setting up requires less time.
Airbrush can help you with these four important challenges
- Spend less: Producing your stock photographs saves time and money.
- Usage anywhere: Make your personal or professional computer produce results that look professional.
- Professional: Professional-quality images can be obtained without the need for a photo shoot.
- Simple to use: You may create photos, sketches, and artwork with Airbrush’s simple interface.
Ready to take your image game to the next level? Sign up now for Airbrush and explore all the amazing features powered by AI technology that will transform your images! Join the Airbrush community now and take your creativity to new heights!
Leave A Comment