🧑‍🏫Prompt Engineering 101

Be specific

Explicit, detailed instructions produce the best results.

Some examples from existing glifs:

  • “Could you describe 4 panels of a comic about the above scene - the first should stir anticipation and intrigue in the reader, with the remaining 3 giving more substance - please include both a summary of the image and the text for each panel.” (Comics from the New Space Age glif)

  • “black and white coloring book drawing of a {input1}, empty background, clean coloring book for adults and children, low detail, black and white outlines of {input1}, Monochromatic line art illustration of a {input1}, blank backdrop, suitable for both adults and children. Minimalist design with crisp, black and white contour lines of {input1}, free from intricate details.” (Coloring Book Images glif)

  • “a top front view render of a bizarre tamagotchi with a rectangular screen, a device inspired by the style of {style}, with the word "{logo}" embossed on the device in small letters. 4K HD octane product render. The tamagotchi device is rendered on white background.” (Glifgotchi glif)

Give the AI a persona

Encourage the AI model to role play. This sounds a little silly, but is actually very powerful. Assigning a persona is a form of priming that will influence the ways in which the model handles the rest of your prompt. This tends to create more creative and less “bot-like” output.

Start with a phrase such as “You are a ____” or “Act as a ____” when assigning a persona.

Some examples from existing glifs:

Provide step-by-step instructions

When crafting your prompt, it can be useful to imagine that you’re speaking to a young child or an entity from another planet that has little prior experience with the concept you’re describing. Breaking things down into steps helps the model process your prompt.

Adding the phrase “Think step by step” to your prompt will assist with this, even if the rest of your prompt is not explicitly broken into steps.

Some existing glifs that use this method:

Provide examples

Adding in specific examples of your desired output can result in more consistent and accurate results.

Some examples from existing glifs:

Use negative prompts

Negative prompts exclude unwanted elements from the output. This can include anything from objects to words to styles to unsettling content.

Some examples from existing glifs:

  • “Blurry, pixel, low quality, grain, ugly, cluttered, pixel art, illustration, drawing” (Teenage Engineering Generator glif)

  • “multiple characters, blurry, photograph,low quality, character sheet, asset, black and white” (RPG NPC Anyone glif)

  • "human, drawing, cartoon, comic, low quality, blurry, bad quality, amateur” (Monster Designer glif)

Set limitations

Phrases such as “only” and “never” within the prompt can help to shape the output to be more precise.

Some examples from existing glifs:

  • “Only return a short visual description, nothing else.” (Insane Emoji Upscaler glif)

  • “Keep it brief, only return the 3 words, do not preface it with anything, just go:” (Low Poly Vector Art glif)

  • “List 3 additional visual details that add relevant features to this, keep it very brief, do not add numbers, only return a comma separated list, max 10 words, just go:” (Candid Style Photography glif)

For image output: Describe the device creating the image

If you have a very specific vision in mind for your visual output, it can be useful to specify the camera or software platform that you imagine being used to generate the image.

Some examples from existing glifs:

  • “shot on film, clean empty blackground, analog film photograph from a 90s yearbook, color photograph, washed out photography, 1990s high school fashion, glam photography of the 90s” (90s Yearbook Anyone glif)

  • “award-winning portrait, fine art, film photograph, analog photography, kodak portra 800, dark atmosphere, film grain, VHS tape” (Twin Peaks Characters glif)

  • “made in Octane, 3d object, 4K render, soft shading, soft-edged, playful, simplified. Isometric perspective, rendered on a white background.” (Clay 3D Icons glif)

Be Patient 😇

Our team is constantly working to speed up processing times, but you can also control processing times through your request. The more complex the prompt and its parameters, the longer the processing time will be.

Ways in which you can cut down on processing time include:

  • Reducing the number of steps

  • Reducing the number of tokens

Don’t Be Afraid to Experiment

Like with many things in life, perfecting your output requires a lot of trial-and-error and practice. If you don’t like the initial output, don’t be afraid to keep modifying your prompt and parameters until you’re satisfied with the results.

Last updated