A prompt is the text input that is sent to the AI generator.
There are 2 ways to send prompts to the Decohere image generator: Real-time Mode and Guided Mode.
The prompting mode can be switched at any time:
Guided mode allows you to talk to the AI generator in simple language. You can send simple instructions like “Make an image of a golden retriever” or “city of Atlantis underwater”. Under the hood, Guided Mode uses AI to rewrite your instruction so that it can be better understood by the image generator. You must submit your instruction to begin generating images.
Guided mode allows you to follow-up on instructions with additional commands. After viewing your first set of images, you can write additional instructions to make changes to the image.
Guided mode is more intuitive to use, but if you want more control over the exact prompt you use, you should switch to Real-Time mode. z
Real-time mode allows you to send prompts directly to the image generator. Prompts in this mode are not rewritten in any way.
When writing a prompt, you should attempt to write a description of your desired image.
The AI generator uses all of the words that are present in the prompt. This makes it important to remove any unnecessary words in the prompt that are not useful for describing your desired image.
Don’t: “Please generate an image of a cabin in the woods”
Do: “Cabin in the woods”
Removing the words “please generate an image of a” from the prompt above makes it better. The AI generator does not respond to instructions like other AI systems (ex: ChatGPT). Instead, the AI generator evaluates all words that are in the prompt and generates an image that is visually similar to the words you provide.
Because the AI generator looks at all words present in the prompt, it can create some unexpected results when you try telling the generator what not to generate. Generally, you should only describe the things that you want to see in your final image.
If you type “A living room with absolutely zero elephants”, you will almost always get a picture with an elephant in it. You would be better off typing “A living room” to get your desired result.
A living room with absolutely zero elephants
Being specific about the details in your image is the best way to gain control over what you generate. There are many ways to specify details in your prompt:
A well-formed prompt uses many/all of the techniques listed above. Here’s an example of how prompt writing can improve your images:
“aesthetic portrait of a woman wearing a green outfit in a sci-fi dystopian city at golden hour, storms in the background. character design by dayer, diego 5, detailed, inked, western comic book art, award winning film poster painting”
“woman in dystopian city”
Using well-known styles in your prompt is a great way to influence the direction of your generated image. The AI has an understanding of most well-known styles and artists. Experiment with your favorites and see how they influence the images.
Some examples:
Mustached man
Mustached man in Pixar style
Mustached man in Salvador Dali style
Mustached man in renaissance painting style
You should generally avoid writing extremely long prompts. The more words there are in the prompt, the less impactful each individual word is for the final image result.
The AI generator typically weights the words at the beginning of the prompt stronger than words at the end of the prompt, so you should put the more important information earlier.