sref (style reference) is a reference method that allows users to specify one or more images as style references. Midjourney will use these reference images to help generate images with similar styles.
Introduction to sref styles
sref can be used to control the overall appearance and feel of the image. For example, you can use sref to generate images in the following styles: realism, surrealism, abstraction, expressionism, impressionism, fauvism, cubism, dadaism, surrealism, and pop art.
cref (content reference) is a reference method that allows users to specify one or more images as content references. Midjourney will use these reference images to help generate images with similar content. The working principle of cref is somewhat similar to image prompts, focusing more on image features, but the biggest difference is that sref generates images with the same style, while image prompts are for basic creation.
Introduction to cref styles
The MJ official states that the cref function is more suitable for anime or games, and is relatively less applicable to real photos. cref can be used to control the content included in the image. For example, you can use cref to generate images containing the following content: anime and game characters, animals, plants, landscapes, buildings, vehicles, food, and objects.
First, we click the top left corner ➕ to upload one or more images. Here we choose a character image to generate. When selecting images, be careful to choose a clear, frontal or side profile, well-lit, and preferably unobstructed portrait photo of a real person as a character reference. Click send, use imagine to input prompt words and dimensions, then add –cref URL and input the image address, and then use –cw to adjust parameter settings. Remember, –cref is currently only applicable to niji 6 and v6 models.
The default parameter is 100 (–cw 100), if you pursue a high similarity to the image, you can adjust the parameter (–cw 80 to –cw 100), if you want to generate more closely aligned with keyword descriptions, you can adjust the parameter (–cw 0 to –cw 30). –cref can restore higher fidelity for character accessories and clothing, but cannot fully restore facial features. However, there are always ways to improve. After repeated testing, I found that providing detailed descriptions of the facial features of the reference image can effectively enhance the similarity of the character’s face. Additionally, the quality of the reference image greatly affects the quality of the generated image. Therefore, when selecting reference images, be sure to choose high-quality images. High-quality images should have the following characteristics:
· High resolution
· Clear and sharp
· Well-lit
· No blurriness or distortion
Examples of cref
(–cw 0) when
The girl wearing a denim suit tied a ponytail –ar 9:16 –cref https://s.mj.run/s9ahb_ilI0s –v 6.0 –cw 0
(–cw 50) when
The girl wearing a denim suit tied a ponytail –ar 9:16 –cref https://s.mj.run/s9ahb_ilI0s –v 6.0 –cw 50
The girl wearing a denim suit tied a ponytail –ar 9:16 –cref https://s.mj.run/s9ahb_ilI0s –v 6.0 –cw 100
Keyword translation: The girl wearing a denim suit tied a ponytail
It is not difficult to find that the image generated with (–cw 0) has a relatively low restoration rate, but the description of the keywords is quite accurate. The image generated with (–cw 100) has a high restoration rate, but the extraction of keywords is not precise enough. The range of (–cw 30 to –cw 60) can achieve a more stable performance in both the restoration rate of the image and the controllability of the keywords.
Similar to the usage of cref, first we click the top left corner ➕ to upload one or more images, here we choose images of styles we like to generate, such as cyberpunk style, polka dot style, pixel style, etc., then click send, use imagine to input prompt words and dimensions, then add –sref URL and input the image address. Remember, –sref is currently only applicable to niji 6 and v6 models.
Advanced usage of sref
The steps for sref remain consistent with cref, except that after inputting the image address, you need to use –sw to adjust parameter settings. The default parameter is 100 (–sw 100), if you want to turn it off, you can set the parameter to 0 (–sw 0), and if you want to restore the maximum strength of the style, you can set the parameter to 1000 (–sw 1000).
Examples of sref
(–sw 100) when
Girl, wearing a beret, a windbreaker, cool boots, holding a baseball bat in hand –ar 9:16 –sref https://s.mj.run/KG-wdaBGNdo –v 6.0 –sw 100
Girl, wearing a beret, a windbreaker, cool boots, holding a baseball bat in hand –ar 9:16 –sref https://s.mj.run/KG-wdaBGNdo –v 6.0 –sw 500
(–sw 0) when
Girl, wearing a beret, a windbreaker, cool boots, holding a baseball bat in hand –ar 9:16 –sref https://s.mj.run/KG-wdaBGNdo –v 6.0 –sw 0
Keyword translation: The girl wearing a beret, a windbreaker, and cool boots, holding a baseball bat in hand
The image generated with (–sw 0) has a relatively low style restoration rate, but the description of the keywords is quite accurate. The images generated with (–sw 100) and (–sw 500) not only have a high style restoration rate but also can extract keywords more accurately.
Cyberpunk style, girl, wearing beret, windbreaker, cool boots, holding baseball bat in hand –ar 9:16 –cref https://s.mj.run/s9ahb_ilI0s –v 6.0 –cw 20 –sw 500 –sref https://s.mj.run/KG-wdaBGNdo
Keyword translation: Cyberpunk style, girl wearing a beret, windbreaker, cool boots, holding a baseball bat
First, we click the top left corner ➕ to upload one or more character images, then upload a reference style image, click send, use imagine to input prompt words and dimensions, then add –cref URL and input the character image address, then use –cw to adjust character parameter settings, then put the reference style image link in, and finally add –sw to adjust the style parameters. The above image is a case I tested, and the overall effect is quite good.
Today’s sharing ends here. Overall, the sref and cref functions are quite powerful, and everyone needs to practice more to flexibly apply them to their work scenarios~
Zhang Junfeng
Using design as a pen, painting the colors of life.
-END-
▼
点这里，学习更多设计知识！
👇