FIRST AND FOREMOST – SOME GROUND RULES
Image AIs can and will produce inappropriate and undesirable outputs in some cases. The prompt you use should be constructed thoughtfully and without any obvious slant towards visualizations that might be considered inappropriate. Use good judgment. We do log all image generations and they will show in a gallery. If you accidentally produce an image that needs to be removed from the gallery, reach out to Matt. Eventually it will be self-serve.
BASIC IDEA
Prompt = image. You write a prompt, the AI generates an image. The same prompt will produce a unique image every time you execute it. So, 2 people could submit “white lamb” and they would get different results. This is by-design for these AIs.
WHY THIS IS IMPORTANT?
Artificial Intelligence in all of its forms is having a huge impact on the world and business. Learning to leverage AIs is a key objective for Bellomy and especially for the tech team at Bellomy. Learning how they work, why they work, how to leverage them, strengths and weaknesses as well as knowledge of their limits and possibilities is key to maintaining a positive position in the industry. It is also an important professional skillset that can only benefit us individually.
In the case of image AIs, some use cases for the market research industry and for Bellomy include:
- Generating custom imagery to go with proposals and presentations
- Matching imagery and iconography to segmentation analysis (e.g. each segment/cluster has a custom icon that represents the segment)
- Generating background imagery to pair with text analytic summaries (e.g. improve story telling)
CONSTRUCTING A PROMPT
Image AI prompts generally consists of up to 5 components.
- Basic prompt
- Style specification (Optional)
- Rendering specification (Optional)
- Artist specification (Optional)
- Customizations (Optional) (Not all engines support this)
The BASIC PROMPT can be one word or a long phrase or even paragraphs.
Some examples:
- yorkshire terrier
- black cat
- dinosaur
- fearsome dragon
- etc.
The STYLE portion of the prompt contains information the AI uses to determine stylistic attributes.
Some examples:
- oil painting
- watercolor painting
- pencil sketch
- concept art
- digital art
- gothic
- illustration
- elegant
- matte
- traditional animations style
- cozy
- delightful
- beautiful
- fantasy
- epic
- Etc.
The RENDERING portion of the prompt contains information the AI uses to render the image.
- highly detailed
- intricate details
- cinematic lighting
- volumetric lighting
- clear white background
- unreal engine
- sharp
- dramatic lighting
- etc
The ARTIST portion of the prompt tells the AI to attempt to “paint” the result in a style consistent with that artist or artists. There is no definitive list of artists, there are 100s in the training data, so a variety exist. It is often interesting to combine artists.
- in the style of Leonardo Da Vinci
- Combining the style of Leonardo Da Vinci and Pablo Picasso
- by Vincent Van Gough
- etc.
The CUSTOMIZATIONS portion of the prompt tells the AI things like desired width and height of the image, how many variations to generate, how many processing steps to perform. Different engines support different customizations. At the moment, the DeepAI Stable Diffusion engine supports no customizations. Other engines will support options and they will be listed here as those engines are added.
SOME EXAMPLES
oil painting of a yorkshire terrier, highly detailed, intricate details, cinematic lighting, photo realistic, in the style of starry night by vincent van gough
water color paining of a crazy lady dancing in the rain in the moonlight by rembrandt
Dogs and cats walking together in traditional animation style
white house lawn, highly detailed, unreal engine, cinematic lighting, by thomas moran and claude money combined with styling from Edgar Degas
epic, fantasy, highly detailed, intricate details, fearsome European medieval dragon, flying over medieval village at sunset by Alfred Bierstadt
abstract art, fine art, highly detailed, volumetric lighting, fantasy, smoke trails, swirls and lines, lightning flashes
