3 services based on neural networks that will help create optical illusions

What optical illusions can you create?

Variants of optical illusions that users are now actively distributing are images based on reference forms or pictures with words “sewn” into them. Such compositions became popular after the advent of the ControlNet plugin for the Stable Diffusion neural network.

It allows you to create pictures based on other images, while preserving the details or shapes of objects from the original frames. The process is reminiscent of generating files using a stencil or template. This principle allows us to form optical illusions.

A black and white image is used as a basis. It could be a geometric figure, a word, or something similar. Objects that are too complex will not be read in the final result.

The neural network creates a picture based on the user’s entered request, but at the same time takes into account the proposed reference form. For example, a cityscape is arranged in accordance with the shape of a flower, or trees form a given word, which is noticed when the viewer moves away from the image.

To use Stable Diffusion in conjunction with the ContolNet plugin, you need a very powerful computer, which not everyone has. But alternative online services have already appeared with the same capabilities, which can be accessed by anyone.

These systems differ in the number of settings and the quality of the results, but after a series of experiments you can get successful and impressive shots. You don't need any special knowledge or skills to do this. Let's look at a few of these tools below.

How to generate optical illusions yourself

1. Using Illusion Diffusion

The Illusion Diffusion service uses ContolNet and allows you to fine-tune the desired result. Here's how to generate an image:

1. Go to the site Illusion Diffusion. To register, you will need a GitHub account. If you don't have one yet, you can quickly create one using email and then return to the neural network.

2. In the editor, select one of the ready-made patterns or upload your own version of a black and white picture. To do this, click “Select file” and select the one you need. You can download or prepare an image for reference in any graphic editor. It is important to use a monochrome shape on a white background.

3. In the Prompt field, enter a request in English. It is better to indicate all the main details of the future image to get a more accurate result. To compile a promt, you can use one of the auxiliary services.

4. The text in the Negative Prompt field does not need to be changed. It contains parameters that should not be in the final result (for example, low-quality objects).

5. Next, you can configure the Guidance Scale and Controlnet Conditioning Scale parameters. They affect how close to your request neural network will generate the frame and how much the loaded reference form will affect the result. Large values ​​will highlight the stencil, and with a lower coefficient the form will be almost unreadable.

6. All other parameters can be left unchanged. Click Run. You will have to wait about a minute for your request to arrive. The time depends on the server load. The finished image will appear to the right of the options panel. Click on the link icon in the top right corner of the generated image to open the file for download.

2. With FusionArt AI

FusionArt AI is very similar in its operating principle to Illusion Diffusion. This service allows you to generate images based on ready-made patterns, user-uploaded templates, or entered inscriptions. The number of free requests is up to 25, but you can bypass the limitation if you use multiple accounts.

Proceed as follows to get a picture with an optical illusion:

1. Open the site FusionArt AI, select a starting template from those offered and click Generate Free Now. You will need to sign in with your Google account or Discord.

2. In the editor that opens, on the Pattern tab, select the appropriate option from the list. Or switch to the Text or Upload tab. There you can enter text to use as a reference or upload your own black and white file.

The Fill container parameter will stretch the reference shape to the entire frame of the generated image. In the Position item, you can mark the location point from which the illusion will be built in the image.

3. In the Prompt field you must indicate a detailed request in English. If you don’t have ready-made ideas, click on the Inspire button - the tool will offer its own option promta for the test.

4. Next, specify the aspect ratio of the image. There are 9:16, 1:1, 16:9, 3:4 and 4:3 options to choose from.

5. The Weight parameter determines how strongly the template shape affects the final result. You will most likely have to experiment with the value of this setting to find the optimal balance between the readability of the reference and the quality of the content built on it.

6. In the Advanced section, you can specify several options for the image to be generated, and also activate the Enhance face option if your request contains people’s faces. The Negative prompt field is open for your instructions on those moments that should not be in the picture. This setting works identically to Illusion Diffusion.

7. Next, click Generate. It may take about a minute to generate the result. The finished image is downloaded using the Download Art button.

3. Using Glif Controlnet Any Word

The Controlnet Any Word service runs on the Glif platform. The tool is built on Stable Diffusion and ControlNet. It allows you to quickly generate images with built-in text, other types of optical illusions cannot be created with its help.

1. Go to the website with the service Controlnet Any Word. To start, you need to register an account. This can be done, for example, through Google or Discord.

2. In the Prompt field, write the request text (in English), which will be used to generate the basis of the content in the picture.

3. In the Single short word to render field, write down the word or phrase that you want to embed in the generated image.

4. Use the Font size parameter to adjust the size of the caption text in the frame. For one word it is recommended to use a value of 130, and for phrases it is better to set it to 100 or even less.

5. Next, click the Glief it button. It will take about 20–30 seconds to generate the result.

You can download the finished image using the appropriate button or copy a link to it via Share. Remember that your requests and those created based on them Pictures are displayed publicly for all users.