How to create a logo using AI
In this short and easy guide I will explain of how to utilize current state of the art Artificial Intelligence technology to generate a professional looking logo. For this guide I’ll assume that you already have installed a working version of Stable Diffusion and a capable GPU device. More guides on how to correctly set up and install Stable Diffusion will be released in the near future.
Generate a first version using txt2img
Before you start to generate images you have to think about how your logo should look like and what should be contained in it. In my case I wanted it to have a brain on a shield like background, so I came up with the prompt:
# webui:
flat logo design of brain on a shield, flat 2d, sharper, company logo
# command line:
python txt2img.py --prompt "flat logo design of brain on a shield, flat 2d, sharper, company logo"Using the above prompt, I generated 128 images and selected my favorite one for further variations. The selected image will be used as initial image to generate further variations or correct smal details using inpaint.
Generate variations using img2img
If you use the webui of Stable Diffusion you can easily move it to img2img using the button ‘Send to img2img’ otherwise if you use the command line version of Stable Diffusion you have to specify the input image using this command:
# command line:
python img2img.py --prompt "flat logo design of brain on a shield, flat 2d, sharper, company logo" --init-img "your_path"If you look at the grid of resulting images above, you see that the overall style of the image remained the same. Depending on how much you want the new images to vary from the input image you can set in the webui the Denoising strength slider to a value from 0 to 1. Zero means that everything should be kept as it is and one means that the model has absolute freedom of how the generated image can vary from the input, these values are the same for the command line approach. For the latter you should use the additional command strength:
# command line:
python img2img.py --prompt "flat logo design of brain on a shield, flat 2d, sharper, company logo" --init-img "your_path" --strength 0.75Change details using inpaint
This feature is only available for users of webui. The inpaint feature is located in the img2img tab and requires you to again provide an initial image. Afterwards you can use the brush tool to select a mask to specify the area of the image that should be changed. In my example I selected the brain-part of the logo and adjusted the prompt to:
flat logo design of a cat on gold background, flat 2d, sharper, company logoThe inpaint feature is perfect to remove artifacts that are sometimes generated by Stable Diffusion. An example would be if you would want to generate a face and the eyes are really blurry and pixelated but the rest looks really good. In this case you would just paint over the eyes and maybe change the prompt a little and run the program again.
Summary
If you want to design a logo but don’t have money to hire a professional artist Stable Diffusion might be what you are looking for. Keep in mind that the model’s outputs can only be as good as the prompts that you provide. It is very important that you know beforehand what you want to generate. Something very helpful is to look up good prompts on i.e. the Stable Diffusion Subreddit. In my case I read that flat 2d, sharper, company logo are really good phrases to use in the prompt for a professional looking logo. Other nice looking brain-themed logos that I have generated are:
