How to use stable diffusion for brand visual extension

After completing the basic environment construction, we entered the preliminary material preparation stage. Prepare the original brand symbol materials that you want to extend visually and the expected visual style prompt keywords.

Taking the Apple logo as an example, the original material image used is 512x512 pixels in size. The material size can be adjusted according to personal needs. In order to facilitate controlnet to recognize the control structure, the color of the material needs to be uniformly processed into black and white.

Here, epicrealism_pureEvulutionV5 is used as an example of the Stable diffusion model. If this model is not available, you can choose any other realistic model.

We use the hardware theme as an example to visually expand the Apple brand, and radiate Prompt keywords such as computer, chip, chip, space, and multi-dimensional around the hardware theme. Enter Prompt in the text box:

geometric multidimensional space, many three-dimensional light emitting blocks, blue light and black background, with a glowing polygonal computer crystal chip,

If you can't come up with suitable keywords, you can also collect some relevant design references on the Internet, insert the reference picture in the picture generation interface, click CLIP to infer the prompt words, select the generated words and then enter them into the prompt of the text generation picture.

In the SDXL Styles interface, check "Enable Style Selector" and check "Digital Art" in Style.

This selector can provide options to generate different types of styles. If this option is not available, you can also copy the following link to install it online in the extension.

https://github.com/ahgsql/StyleSelectorXL

Select the number of iterations, sampling method, width and height, and prompt the secondary bootstrap coefficient. The width and height are set to 512 here to facilitate the rapid generation of different effect diagrams.

After setting the parameters, click Generate to start drawing cards.

Select the direction that best matches your expectations in the generated images. Here we take the first image in the upper left corner as an example.

We input the image seed that matches the direction in the random number seed, and use high-resolution repair to enlarge the image precision. In the high-resolution repair (Hires. fix) interface, select the enlargement algorithm ESRGAN_4x, set the high-resolution iteration steps, redraw amplitude, and magnification. The iteration steps are generally selected between 10 and 20. If the number of steps exceeds 20, deformities may occur. The redraw amplitude is generally between 0.3 and 0.8. Adjust according to the actual effect. If you want a larger change, increase the value.

Click to generate the redrawn image.

The generated image is simply processed through PS to incorporate more luminous crystal elements into the center of the image.

We switch to the image generation mode and put the processed image into the image generation mode for secondary generation. Enter the same prompt as the text generation mode. Set the number of iterations, sampling method, width and height, and redraw amplitude.

The controlnet parameters are basically the same as those in the previous figure.

After setting the parameters, you can click Generate. During this process, you can adjust the redrawing amplitude according to the effect until a satisfactory picture is generated.

The final effect.

By setting the Controlnet parameters, you can increase the Control Weight strength and make the graphic outline more obvious.

After completing the parameter settings, start drawing cards to generate pictures.

We select the right picture to proceed to the next step, and make simple adjustments to the bottom and head structure in PS, replacing the flames with dust effects.

After completing PS, we switch to the raw image mode, insert the processed image, and enter the same prompt and parameter settings as before.

Then you can click to generate the final screen effect that meets your expectations.

With the main visual effect, related extended applications will come naturally, combining information content and adapting and expanding the visual effects according to the application environment.