Foreword: Brand symbol poster as a very important means of brand operation, with the continuous development of AI technology has gradually replaced the traditional design method, the improvement of quality and efficiency makes it a popular design method of super symbol extension, today we will learn about the SD super symbol case tutorial, analyze the use of stable diffusion, an AI tool to make super symbol posters.
We can see the symbol posters made by AI displayed by the major designs, including text, graphics, etc., most of them are rich in design styles, displayed in a matrix, the picture is very exquisite, full of imagination, and it takes very little time and has a rich output, so how is such a poster generated and produced?Here's a step-by-step analysis.
2.1. Basic tutorial analysis
Basic Theoretical Analysis:In the Wensheng diagram mode, the brand symbol poster is designed through the combination of large model + prompt + controlnet (lineart controls the outer contour), which can assist LoRa and the designer's creativity at the same time.
To summarize the basic formula:logo+lineart/canny
Basic Formula Diagram:
The poster of the 100 symbols designed under the basic formula:
3.1. Introductory SD symbol generation mode
In the basic formula, we need to add a deep controlnet to control the relationship between the front and back of the picture.
Advanced Introductory Formula:logo+canny/lineart+depth
The degree plus symbol poster designed under the advanced version of the introductory formula:
3.2. Advanced threshold - depth map
As an advanced introductory, we need to understand the concept of "depth maps".
Depth map
A depth map is a two-dimensional image that stores the depth values of all pixels in a single view, and is a unit of space, such as millimeters.
1) A depth map is a two-dimensional image, the same size as the original image.
2) Each position of the depth map saves the depth value of the pixels at that location.
3) The depth value is the Z-coordinate value in the camera coordinate system.
The shallower the area on the depth map (depth value close to 1) and the farther away it is, the darker it is (depth value close to 0). That is, the depth value from 0 to 1 represents the distance from far to near.
SD In-depth Analysis:
There are four kinds of processors in the ControlNet plug-in, namely Midas, ZOE, LERS, LERS++, the fineness is incremental, and the resulting model generation time is also increased exponentially, we can choose according to the needs, among which MIDAS and ZOE are more suitable for symbols.
Advanced Introductory Formula Diagram:
We select the depth map that can be automatically generated by the depth in the SD ControlNet plugin, we can run the map through the pre-processed effect, and then select a more satisfactory one to generate a poster, of course, making our own depth map will have better control over the accuracy of the picture.
3.3. Primary SD symbol generation mode (modeling aid).
When the three-dimensionality of AI-generated symbols cannot meet the needs of our design creativity, we need to assist in modeling.
Advanced Beginner Formula:Three-dimensional logo + depth + canny
Modeling Aids 10 - Advanced version of the primary formula:
Modeling Aids 20 - Advanced version of the primary formula:
We can also generate a symbolic perspective of a fallen two-point perspective, but the corresponding perspective needs to take into account the occlusion and recognition problems that will occur when creating the image.
3.4. Intermediate SD symbol generation mode
In the process of daily drawing, we will often spend a lot of time drawing cards, lucky to draw S card, the time of drawing the picture is halved, if you are unfortunate enough to draw the S card, the time of the picture is doubled, here I will introduce you to a self-developed symbol generation method, which effectively helps to improve the probability and draw the S card pool.
3.5. Advanced threshold - semantic segmentation
As an advanced intermediate generation pattern, there is a concept that needs to be understood, and that is "semantic segmentation".
Semantic segmentation
Semantic segmentation is an important direction in computer vision. Different from object detection and recognition, semantic segmentation implements pixel-level classification of images. It is able to divide an image into multiple chunks according to the similarities and differences in categories.
Semantic segmentation is the ability to divide each part of the whole graph, so that each part has a certain category meaning. Unlike object detection, object detection only needs to find the target in **, put a box on it, and then divide the category. Semantic segmentation is in the form of strokes, dividing the whole image into each area without leaving gaps, each region is a category, and the default background without category is background.
Comparison of pictures with semantic segmentation:
Intermediate SD symbol design degree plus symbol poster:
Here we can draw a black, white and gray depth map according to the design plan, the depth map here only has the outline of black, white and gray, and has no meaning of specific elements, so the generated picture is only composition and light and shadow.
Intermediate SD symbol design degree plus symbol poster:
On the basis of the black, white and gray depth map, we will divide each part into colors, you can refer to the semantic segmentation table on the Internet, at this time, in addition to the composition and light and shadow, we also give each element a specific meaning, and further control the composition of the picture.
+lora
We can use this method to add a variety of lora to stably generate a very high-quality poster picture, select the appropriate picture in the s card pool for further refinement, and add the typography is a set of brand symbol promotional poster images.
3.6. Advanced SD symbol generation mode (style control).
When we want a certain style, but there is no suitable lora, we need to research our own way to extract the style, which is the self-developed advanced symbol generation method, here are two special controlnet controls, shuffle and refrence.
shuffle is shuffling, that is, scrambling a picture and keeping only the style;The refrence is the reference, and the whole style, composition color, etc. follow the reference, for example, the following two pictures, we can extract the style through these two plugins:
shuffle and refrence:
Using the advanced symbol generation mode, we can run a series of brand symbol posters at the same time
3.7. Post-processing & drawing
The post-production is mainly carried out through the diagram mode, and the common post-production methods include enlargement algorithm, tile plus detail, post-compositing, local redrawing, etc.
Here's a plug-in model that enhances the details of the tile:
The tile model is essentially a redraw model, but it will add more details to the picture and improve the image quality while maintaining the original composition, painting style, and main body of the character.
1) Add details, such as high-definition upscaling, repair**, complete sketches, increase image quality, etc.
2) Ultra-high-definition amplification with other amplification algorithms (the tile model can suppress the pollution of local images by global prompts).
3) Automatically infer the content based on the screen to eliminate the sense of disobedience.
Degree plus, versatile, new original plan logo interpretation:
The MEUX content ecosystem design team is committed to researching more ways to control the creativity, style, quality and efficiency of AI, exploring more possibilities in AI graphics, and making AI a useful design tool for designers.
Enter the design gurus"AI Painting Library".Get your free SD+MJ drawing primer resources!
AI Painting Library Address:
After clicking on the link to log in, click to join, it is currently free and open, if it is deactivated, you can contact WX "Commyleung" to apply for a new channel to join.