In the previous article, we introduced the method of building a comfyui workflow to quickly build scene renderings for products. How about drawing inferences from each other this time and replacing the product image with a font?
Let's take a look at the finished product first:
This is a set of four seasons posters, from the designer's point of view, the creativity of the poster is also simple, the use of calligraphy font fusion of seasonal elements as the main vision, with ancient poetry copywriting to echo, to create an elegant sense of literary and artistic seasonal atmosphere. The difficulty is that how to combine the glyphs and elements? How is the scene graph arranged? How do you match the colors? How do you create realistic light and shadow effects? Usually, if there is no suitable material for this picture effect, if you want to design it from scratch, it is estimated that it will not be able to be done in a few days (maybe in the end, it is likely to be an illustration version =. =！）
Now, AI has taken care of all of the above questions! All you need is an idea, have a certain aesthetic ability, and understand the basic operations generated by AI.
The whole process of creation is very efficient, you can create a basic comfyui workflow in a few minutes, then find some fonts and styles, and then draw cards to create ...... with parameters and promptsIn the end, I picked ...... in a bunch of generated **After selecting the picture, use PS to carry out simple text typesetting and color grading in the later stage, and it will not take half a day to complete a set of calligraphy font poster design scheme for the four seasons.
Set up a comfyui workflow
If you don't have any experience, you can go to station B to find relevant tutorials, including the deployment and basic operation of Comfyui, etc. It is recommended to directly use the image package deployed by others on cloud hosts such as autodl, which does not occupy the performance of the local host (mainly because the performance of my host is too poor =. =！), and secondly, to save time and study costs.
The basic idea of the workflow is to control the text part and the background part separately through the mask on the basis of the diagram, and then generate the final picture according to the prompt words. Here is a display of the whole process of construction through **, and the plug-in used has a display name in the upper right corner of the node panel, and the relevant model can be used accordingly.
Process Essentials
1. The font is black and white, and some fonts can be generated directly, and you can choose different styles of calligraphy fonts. Node processing is used to generate a ** and two positive and negative masks for the diagram. Here, the Tusheng diagram has been adjusted to reduce the contrast and brightness, and reduce its control intensity, so that the resulting ** has more variety.
2. Use the third-party plug-in Advanced ControlNet, which can add a mask to make ControlNet only affect the font part. The depth model can be combined with the scribble model to enhance the three-dimensional effect while maintaining a certain glyph outline.
3. The clip vision model control of ipadapter can migrate the style and picture elements of the reference ** to the generated **, but the control and influence of ipadapter is strong, so the parameters should not be set too high.
4. Prompt words are also crucial. For example, Chinese fonts have many different strokes, and the glyphs are ever-changing, and it is generally difficult to correspond to the shape of a specific thing. Therefore, when writing the prompt words of the font part of the picture, try to use the nouns of some malleable objects as the main prompt words, such as liqiud, metal, wood, plastic, plaster, glass, fabric and other material words, or objects with rich shapes, such as branch, sculpt, rubber clay, jade carving, wood carving, stone carving, paint, plant, etc. This makes it easier for the AI to understand when it is generated, without creating a cluttered and illogical picture effect.
Comparison of the generation effect of different prompt words and style maps
The magic of AIGC is that if you give it a simple idea, it will always give you unexpected surprises.
I wonder what form artificial intelligence will evolve into in the Year of the Dragon?