Prompting Glossary

Hey there!
I just bought the Veras plugin for a rapid prototyping I needed yesterday, and I got really good, and also - really bad :smile: results too, but non the less - the plugin is really interesting, and will be a tool for me.

What I would love to explore though is a prompt glossary as I saw that the algo reacts to certain prompts, while completely dismisses others, no matter how much emphasis you put on them. So my assumption is that it does not understand them.

Is there a way to see a prompt glossary, or at least a general guidance for prompt buiding other than the prioritizing with parenthesis?

1 Like

@G.Iliev - welcome to the forum and thank you for purchasing Veras!

Here are some notes on prompt engineering:


Currently prompts are only supported in English. There might be some overlap with other languages, but at the moment other languages are not supported.


Statements should be separated by commas, to denote separate ideas. Example:

good: brown carpet, back sofa, living room
bad: a living room with a black sofa and a brown carpet


The way you can emphasize part of the prompt is using parenthesis. You can add multiple parenthesis for specific text. Example:

red house, (((on the moon))), (with view of earth in the background)


It’s good to keep the prompt clear & concise, instead of vague. Example:

good: timber building
bad: timber construction

This does not mean to keep the prompt short. The prompt should be as detailed as possible.


It’s better to define elements in a specific way, instead of a vague way. Example:

good: white walls, white ceiling, forest seen through large windows
bad: white room, in a forest

The bad prompt above might make everything in the room white, as it’s not specific. Additionally, the room itself might have foliage & vegetation, instead of showing the forest outside the building.

Avoid Over-constraining

If the prompt is too specific (like a specific furniture piece that the ML model is not aware of), and this part of the prompt is emphasized, you can get bad pixelated results. It’s better to keep the prompt more open-ended. Example:

good: office chair with a black frame and a black mesh, office
bad: ((steelcase chair series 2 model B08L8J6BYV)), office

The prompt strength also works in tandem with the prompt - you can think of it as a big parenthesis over the entire prompt with a multiplier based on the slider value. If the prompt strength is maxed out AND the prompt is too constrained, you can get pixelated outputs.

Prompt Anatomy

These are some characteristics that can make up a prompt. The examples are not exhaustive.

  • Subject - what is shown in the scene
    • use the Is Interior toggle for interior renderings
  • Season
    • summer
    • winter
    • spring
    • autumn (better than fall, as fall can be more vague)
  • Weather
    • sunny
    • overcast
    • blizzard
    • rain
  • Time of Day
  • Lighting
    • interior lights (useful for exteriors)
    • soft lighting (useful for interiors)
    • cinematic lighting
    • dark
  • Color - add colors to the subject
    • glass with green hue
    • chair with gold frame
    • white walls
  • Style & Details
    • Ornate details
    • minimal
    • monolithic
    • rustic
    • modern
  • Aesthetic
    • not defining the aesthetic will default to photo-realistic
    • charcoal drawing
    • pencil hand drawing
    • watercolor
    • oil painting
    • rendering
    • surrealist
    • hyperrealistic
    • 8k
  • Effects
    • bokeh (for blurring the backgrounds - less effective with low geometry override values)
    • sharp focus
    • sharp edges


It’s good to test your prompts with at least 4 batch renders, as the random seed can produce very different results using the same settings

Prompt Assist with ChatGPT

You can use ChatGPT to generate prompts :smile: ! Check out this post: Veras for Pottery

Examples of Prompts

modern design with large windows, interior lights, timber building, during winter, snow, blizzard

villa in a city, white paneling, blue glazing, golden hour

cabin in desert, large blue glass windows, during summer, sandstone

beach house, large windows, during summer, ocean in the background, sand foreground

designed by SOM, during spring, glass facade, interior lights, aluminum panels, forest in background

cabin in the woods with large windows, interior lights, timber building, at night, night shot

large windows, forest outside, sky, wood flooring, book shelves, built-ins, chairs, soft lighting, dynamic lighting, fireplace, brick, wood ceiling, blue sky, timber, wood floor, designed by Joanna Gaines

living room, white walls, white ceiling, gray wood floor, forest outside seen through the larger glass windows

yellow modern volkswagen beetle car, white walls, white ceiling, forest in the background during autumn


The camera view that frames the shot is correlated to the prompt & quality of the output. For example, if the camera is set to a really obscure angle, this constraints what the prompt can interpret, as less ML data would be trained on images for such angles. A good though is to frame the shot a way a photographer would, as there is more data on this framing:

  • straight on shots
  • horizon to be shown
  • closer to 2 point perspective, instead of 3 point perspectives
  • do not use a super wide field of view or highly distorted views

Building Prompts

  1. start with a simpler prompt of the subject
  2. test the prompt with at least 4 renders
  3. gradually add other parts of the prompt anatomy: mood, weather, style
  4. test again with 4 renders between each new comma separated statement

Prompt Order

The prompt order matters. The text that is in the beginning of the prompt automatically receives more attention from the AI.

Let me know if this is helpful.


amazing. Exactly what I needed, thanks a ton for providing me the glossary :blush:

1 Like