A colleague had recently provided me with pass-through access to OpenAI’s DALL-E platform.
The DALL-E platform generates images based on natural language input using something called a diffusive AI trained model. It is able to create unique images from what appears like random noise and then iteratively makes small changes to the image until it starts to ‘look’ like what was asked. The ‘looked’ like portion is tested against a trained neural network to recognize objects and context. Once it scores with high probability, the image is complete.
Here’s an example:
Training prompt: Puff the magic dragon with a background cannabis leaf
Output:
Some variations:
Let’s try changing the input prompt.
Training Prompt: Puff the magic dragon with a background cannabis leaf in the style of a cartoon
Output:
How about impressionist styling?
Training Prompt: Impressionist style puff the magic dragon with a background cannabis leaf
Output:
Training Prompt: Puff the magic dragon in a garden with tomato plants, cannabis plants, and cucumbers
Output:
How about … A cannabis plant that grows cherry tomatoes
Output:
How about … A cannabis plant that grows cherry tomatoes illustration
Output:
Do you want to give it a try? Output could be just about anything.
For instance, An astronaut on a space walk reaching towards the sun produces the following:
Or: An astronaut in space surrounded by alien spacecraft
Amazing. Controversial. Scary. All machine generated. The generation isn’t free but not too much.
Wanna try? For now, one phrase per individual until I run out of the credits that’s been shared with me.
Phrase can be anything that makes sense to a human (keep it clean). As you can see, the output sometimes is amazing, sometimes not the great, sometimes unexpected. Pick your phrase carefully. Enter it below.