In this step-by-step guide, we will explore how the use of a prompt can influence the generation of images using the ControlNet feature
We have three distinct scenarios and will be utilizing three different modes: Character Mode, Interior Mode, and Depth Mode. We will generate an image in each scenario, starting with no prompt, followed by a basic prompt, and finally, a more advanced prompt
Example 1
For our first scenario, we aim to create an enchanting image of a red-haired sorceress. I've handpicked a photo that I believe effectively captures the essence of this theme. This image is then dragged and dropped into the Reference Image field, and the Character Mode is selected from the ContolNet feature.
Model: TTRPG Realistic Fantasy Style
ControlNet Mode: Character mode
.png?width=688&height=459&name=Untitled%20(1).png)
No Prompt
The prompt-less generation is more random and perfect for generating new ideas and brainstorming.
Sampling Steps: 50
.png?width=688&height=459&name=Untitled%20(2).png)
Basic Prompt
Prompt: a beautiful powerful witch girl with redhead
Guidance: 7
Sampling Steps: 50
.png?width=688&height=459&name=Untitled%20(3).png)
Advanced Prompt
Prompt: a beautiful powerful witch girl with redhead, high detailed, high quality, 12k, beautiful eyes, high quality face
Negative Prompt: Disfigured
Guidance: 9
Sampling Steps: 50
In this case the result without the prompt was already very good, but random. With a basic prompt it is enough in most cases, but if you want something more it is essential to create a more complex prompt, defining some details and fixing some problems.
Example 2
Our second scenario involves the creation of a hospital room, but not just any room - a room from the future. Here, I've chosen a photo of a rather ordinary hospital room to serve as our base image. Like before, this image is dragged into the Reference Image field. However, this time, we switch the ControlNet to Interior Mode.
Model: Asset Diffusion(Beta)
ControlNet Mode: Interior mode.png?width=510&height=684&name=Untitled%20(4).png)
.png?width=688&height=459&name=Untitled%20(5).png)
No Prompt
Sampling Steps: 50
.png?width=688&height=459&name=Untitled%20(6).png)
Basic Prompt
Prompt:a futuristic hospital room
Guidance: 7
Sampling Steps: 50
.png?width=688&height=459&name=Untitled%20(7).png)
Advanced Prompt
Prompt: a futuristic hospital room, high detailed, 12k, micro details
Guidance: 9
Sampling Steps: 150
The hospital room is a great example of how the prompt can help us improve and reduce distortions.
Example 3
Lastly, in our third scenario, our objective is to craft a turtle monster, suitable for a gaming environment. To achieve this, I've selected a picture of a turtle, specifically chosen for its pose and perspective. This image is dragged into the Reference Image field, and the Depth Mode is chosen in the ControlNet feature.
Model: Anime Horror Creature Generator
ControlNet Mode: Depth mode.png?width=490&height=678&name=Untitled%20(9).png)
.png?width=688&height=459&name=Untitled%20(10).png)
No Prompt
Here we already have a very scary monster that somehow resembles a turtle, but I am sure that with the prompt we can improve this.
Sampling Steps: 50
.png?width=688&height=459&name=Untitled%20(11).png)
Basic Prompt
Prompt: a monstrous giant turtle
Guidance: 7
Sampling Steps: 50
.png?width=688&height=459&name=Untitled%20(12).png)
Advanced Prompt
Prompt: a red eyed monstrous giant turtle with open mouth full of sharp teeth
Guidance: 9
Sampling Steps: 50
Here, we have managed to create our monster, keeping the characteristics of a turtle, using the provided prompt.
Wrapping up our little adventure here, it's pretty clear how much of a game-changer this ControlNet tool is for our art projects. Throw in the right prompts and you've got yourself a super tool that lets you tweak and fine-tune your artwork.