Face work: Emotions
Welcome to the third practice lesson! Learning how to work with faces is divided into two parts, namely working with emotions and hairstyles. Within this lesson, we will break down the different emotions and how they affect the final result.
Emotions: their purpose and influence on the result
First of all, I would like to draw your attention to the fact that with the help of emotions we can solve one of the rather serious problems of any model and algorithm, namely poor eye mapping. Problems with eye mapping go to the same level as hand problems, they are constant. Today all existing models have problems with eye mapping ! One option to solve this problem is to use emotions. With their help you can get a more acceptable result and hide the flaws of the model.
Here's a list of the emotions I use in my work process:
For demonstration I will use this prompt (separately marked what was changed)
- A girl, wearing a short-sleeved skirt, upper body, (laughs: 1.1 Меняем только тут. Увеличенный вес указан для лучшего эффекта), (open eyes: 1.2), bright light, background bokeh, depth of field, blurred background, light particles, strong wind, (heart particles: 1.1)
You can also work through the emotions on your own. To do this, use this link to a ready-made algorithm.
Important specks when working with emotions
For a better understanding of this lesson, I want to draw your attention to these aspects when dealing with emotions:
- Write the emotion clearly. Try not to add additional words to it. For example, "aqua crying face" will not give you the same result as just using "cry".
- You can't combine too different emotions in one prompt. If you try to combine "cry" + "laughs", the result will be completely unpredictable.
- Increase the weight for such an element as emotions, for example (laughs: 1.1). This is to make the AI more likely to draw the emotion we need. This tip applies to both small queries and very long queries. If you follow these simple rules, you will get very good and, most importantly, predictable results.
Results
Let's move on to practice. First I want to consider the emotion of joy. Emotions of this group are very often interpreted in the same way by the AI and you can get very similar results for "laughs" and "excited".
The same principle applies to the emotion of sadness. The AI treats sadness emotions like "ominous", "expressionless" and "sad" very similarly. So if you need to get one of these emotions, which is quite rare because everyone prefers to see only smiles, you can try each of them and find for yourself the saddest face and what the AI will generate.
The next set of emotions that can be distinguished is the emotion of embarrassment. From them "embarrassed", "shocked" and "confused" are distinguished (a specific emotion can be interpreted differently in different models)
In addition, I would like to emphasize specific emotions, namely "nervous" and "maniac". Why are they specific? First of all because of the unpredictability of the result. It all has to do with the fact that they will be interpreted differently depending on the model. In the examples that I give you, I specifically looked for models that will work with these emotions with the highest percentage. If you want to make such emotions on your own, you need to experiment with different models. "maniac" in our particular case is not much different from joy or embarrassment. And "nervous" AI worked very well, because you can see some nervousness through the smile.
And then we still have anger. This is a classic emotion, which will show different results depending on the model. In our particular example anger is well expressed, but without unnecessary details, it is very successful. Sometimes anger can be displayed through grinning teeth, strong distortion of eyes and mouth area.
Homework
Your homework will be to try out the different parameters for working with emotions on your own. You can use this prompt for this purpose
- A girl, wearing a short-sleeved skirt, upper body, (laughs: 1.1 Меняем только тут. Увеличенный вес указан для лучшего эффекта), (open eyes: 1.2), bright light, background bokeh, depth of field, blurred background, light particles, strong wind, (heart particles: 1.1)
Or you can go straight to the link and start generating.
You can also see examples of how to use prompts in my telegram group, where I share ready-made algorithms and image creation tweaks every day.
Download the results of your experiments and share them within the course. Remember that the main thing is your creativity and imagination! Good luck, and enjoy the process of creating unique images with artificial intelligence.