There are so many AI image generators out there- many of them still in their infancy, but the creative potential these platforms offer in building images and videos is phenomenally wide. There are still a lot of hallucinations these AI tools might do if you are not prompting or training the model enough. So for beginners, I am putting down a couple of pointers that might help you generate images and videos in the way you intend.
1. Try starting from a compositional reference.
Compositional reference is used to train the AI tool to understand the overall construction of the image (Layout, composition, scale, contrast, etc). Create a blueprint image of your idea. This can be as rough as a pencil sketch or as detailed as a photo montage. Do not focus on photorealism here, instead pay attention to putting the structure of the imagery in place. Make sure that your dimensions are right, the forms are close and the contours are clear. For example, I assembled some pngs to get my reference for composition.
2. Choose a style reference.
Depending on the AI tool that you are using, there are several ways to input a style reference. Style refers to the aesthetics of imagery, tone, mood, etc. Midjourney uses Style reference codes, Firefly uses Style reference uploads. In this example, I used Pinterest to look for an image with the style I wanted to replicate.
3. Be specific and logical in prompting.
Do not waste time by writing unnecessary operators like "create", "add" etc. Be descriptive and explain the visual you want to create. Use your compositional reference to drive the structure of the image and your style reference to create the aesthetics - play with its strength until you get the best outcome.
<end of brain dump>
Hozzászólások