#Tip 9 – Using emotional cues
7 messages · Page 1 of 1 (latest)
Every word in a prompt will affect the result, sometimes in ways you don’t expect. Adjectives describing emotions often produce side-effects based on associated training data. It can be very hard to fight training data biases. There is an easy way to explore what the AI will want to do on its own when adding an emortional keyword. The trick is to check first for the common result for the emotion just by itself, with no other subject or art style. Most of the time you will get back a set of photos. Don’t worry if the photos have nothing to do with what you want to create. All we are doing in this step is to notice the body language, setting, or props that are associated with the emotion.
Let’s try this with the adjective “thoughtful”:
Notice that in all these, body language is very significant. In three of them, the hand is touching the face, and in the fourth the hand is making a gesture near the face. The lips are pursed or downturned. We did not ask for these gestures, they are just associations in the training data.
What will we get if we ask for a “thoughtful robot”?
The hand and arm positions are similar to the photographic samples, and in two of them the mouth is downturned. Two have a question mark. Unexpected text or symbols is sometimes due to the underlying training data having them.
What does the keyword “happy” give us?
Lots of smiles, and arm position is noticeable in at least three of them. We also got a banner, background scenes, and a guitar. What will happen if we ask for a “happy robot”?
Notice that the arm postions are energetic, and two of the four have big smiles. We didn’t ask for these features in the prompt. They are indirect attributes found in the training data for the word “happy”.
A “bored robot” tends to sit down.
Once you figure out what the AI wants to do with emotional cues, you can simplify your prompts. You don’t need to ask for features that the AI will probably produce anyway. Reducing the complexity of your prompt makes it easier for the AI to get a coherent result.
For more tips, see the guides in #📚┃resources or ask about specific prompts in #❌┃prompt-help.