Huge Midjourney Update: Consistent Characters, Step-by-Step Tutorial

Video Statistics and Information

Video
Captions Word Cloud
Reddit Comments
Captions
mid Journey just released one of its most requested features consistent characters now it is easy to create consistent characters like these and like these in this video I will show you how you can do the same first you will need a base image that will be a reference for all of the other images I'm going to use this one to make it easier I will create an option for the new parameter rightclick on the image and select copy link start typing SL prefer you want the prefer option set command so press enter click on plus one more and click value type-- CF which stands for character reference add a space rightclick and paste the URL of the image enter the name of the option and press enter now I'll enter a prompt for a scene in which I would like to place Tom I think I'll have him riding a bike in a park I want the image to be wide so I'll change the aspect ratio to 16 to9 in order to add Tom to the scene consistently I need to add our custom option d-t one if you look at the prompt you'll notice that our custom Option d-t 1 was replaced with its value-- CRF followed by the image link now look at these images I think they all look like the original image of Tom they are even wearing the same outfit some with slight variations for the most part I would say these images are consistent what do you think and here are some more images I generated in their prompts when Examining The Prompt for this generation you'll notice I requested Tom to be depicted riding a horse in a cowboy outfit however his clothing Bears a closer resemblance to the attire from the original image albeit not identical rather than typical cowboy clothes to rectify this mid Journey introduced a new parameter D- CW representing character weight ranging from 0 to 100 with a default value of 100 a value of 100 will try to recreate character in the image reference including the face hair and clothes a value of zero will focus only on the face which is great if you want to change outfits or hairstyles so I re-entered The Prompt and added the character weight parameter with a value of zero in this instance I don't think any of these images look like Tom to me it seems like the Pixar style is taking over and making Tom look more like Woody from Toy Story I'm going to try try increasing the character weight to 25 the top two bear a stronger resemblance to Tom while the bottom two appear less like him Additionally the clothing carries a cowboy influence but still preserves elements of the original images style like a blue shirt you can incorporate multiple image references as well let's say I want to include this image for character reference I'll utilize the prefer option set command once more more choose your option now it appears I can't simply add another link because the value is empty so I'll need to copy the value from one of the prompts and paste it into the value box copy the link of the image you wish to add then paste it ensuring there's a space between the links you can also experiment with creating a character sheet using your character reference however it took me numerous generations to achieve one that somewhat resembles Tom I had to reduce the stylization to 50 as indicated in the prompt next I experimented with a new prompt for generating a character sheet simply input the number of images along with multiple poses and facial expressions the results varied while the face showed some consistency the shirts displayed less uniformity Additionally you can explore incorporating character and style references style references enable you to effortlessly replicate a style while maintaining its Aesthetics to include it in your prompt simply type-- sref followed by the link of the image whose style you wish to emulate I'll upload an image into Discord and press enter I will make Tom a comic book hero adopting the style from this image you can further refine the style reference by adding weight using the D- ssw parameter this parameter accepts values from zero disabling the style to 1,000 closely following the reference Style with a default value of 100 and here's the result all of them bear some degree of resemblance to Tom and the bottom two appear to depict an older version of him apart from that the images look fantastic and perfectly capture the style of the reference image after trying out this feature I must say it's quite impressive however it's not Flawless as it may produce inconsistencies from the reference image mid Journey advises that it works best with images generated by mid journey and isn't intended for real people or photos additionally it might not replicate exact styling across Generations such as freckles dimples or intricate clothing details as evident in these images nevertheless I'm eager to see the potential developments of this tool in the future have you given this feature a try yet if so share your thoughts by leaving a comment below if you found this video helpful please consider liking and subscribing to our Channel thank you for watching
Info
Channel: AI Concoction
Views: 7,210
Rating: undefined out of 5
Keywords: midjourney, midjourney ai, texttoimage, ai images, midjourney news
Id: D4dHfy_snjE
Channel Id: undefined
Length: 6min 28sec (388 seconds)
Published: Sun Mar 17 2024
Related Videos
Note
Please note that this website is currently a work in progress! Lots of interesting data and statistics to come.