Это видео недоступно.
Сожалеем об этом.
Midjourney CREF Deep Dive | Consistent Character Ultimate Guide | Midjourney v6 Tips
HTML-код
- Опубликовано: 12 авг 2024
- #midjourney #midjourneyv6 #midjourneytips
✨ Midjourney Cref Deep Dive - Midjourney V6 Character Reference Parameter!✨
Midjourney's Character Reference is a huge step toward creating images with characters that are truly consistent with a reference.
Watch this video to learn how to get the most out of Midjourney's Character Reference parameter --cref. I walk you through everything from the basics of Character Reference, unique applications, and creating images with multiple reference characters.
This video was a bit of a beast to put together. There are so many nuanced ways to use Character Reference, so please let me know if you have any questions about it.
Make Midjourney Images on the Website: • GOODBYE Discord? Make ...
Midjourney V6 Style Reference Deep Dive: • Midjourney SREF Deep D...
Midjourney V6 Prompt Series: • Midjourney Prompts for...
Midjourney V6 Features & Tips: • Midjourney V6 Tips & F...
FREE Midjourney Guides: woollyfern.gumroad.com
20 FREE SREF Random Styles for v6: woollyfern.gumroad.com/l/midj...
100 SREF Random Styles for v6: woollyfern.gumroad.com/l/midj...
FREE Midjourney v5 Style Codes PDF: woollyfern.gumroad.com/l/midj...
00:00 Intro
00:32 Basics: How to Use Cref (--cref)
01:31 Multiple Scenes and Poses
01:52 Character Weight (--cw)
03:00 Using Cref on Midjourney's Website
03:33 Transfer Character to Different Styles
04:24 Using Character Ref with Style Ref (--sref)
05:24 Non-human and Abstract Characters
06:09 Multiple Characters (Four Methods)
08:25 What Makes a Good Character Reference?
09:08 Tips for Character Posing
09:35 Wrap-Up
/ woollyfern
/ woollyfern
/ woollyfern.art
⭐ Don't forget to like, comment, subscribe, and share this video with anyone you think might find it helpful!
I hope you enjoy the video!
Support Woollyfern / Buy Me A Coffee
ko-fi.com/woollyfern
Tools [affiliate links]
AI Image Upscaler: geni.us/sh3QMig
AI Text-to-Speech: geni.us/wdqnz
Music from Epidemic Sound: geni.us/oWJUc
If you purchase through one of my affiliate links, I may receive a small commission at no cost to you. All commissions directly support costs related to creating videos for this channel.
Patreon Community ➡www.patreon.com/Woollyfern
What is your favorite Character Reference workflow? How are you getting your best results?
Simply wow, this is the best structured midjourney tutorial regarding the --cref feature. Thanks.
Thank you, glad you enjoyed it!
fainallyyy a clear, easy, simple tutorial -- Thanks
Glad it helped!
Thank you! Everything turned out great!
This video helps so much! Thank you!
Great examples! Thanks!
I think I've found the channel I was looking for. Thank you! :)
Great, thank you :-)
Brilliant video, loved the structure and all the examples. Thank you!!
Thanks!
Simply fantastic MJ tutorials! Subscribed!
Thank you!!
wow. you are such an amazing speaker, so EPIC this was. So clear. Thank You!
Thank you!!
Perfect vid, thX!
Luv your vids, M$ MJ master! 🎉👏🏻✨💯
Hi, thanks for a great deep dive video.
Thank you! Super helpful. Much appreciated.
Thank you so much!
Thank you so much! It was very helpful! ❤
Great to hear, thanks!
excellent video, chock full of info and in record time. well done. subscribed.
Thanks!
This is great! Thanks for these details!
Happy you found it helpful, thanks!
brilliant, thank you!
Terrific video. Subbed!
Thanks!
Pretty good video, thanks
Excellent work! I consider myself very knowledgeable, but took away some very useful ideas/approaches from your exploration.Thank you!
Awesome, thank you so much!
Best tutorial ever 👏🏽👏🏽. Thank you
Thanks!
thanks a lot. Your video is just awesome!
No problem, thanks for the support!
this is great---What I am struggling with, and perhaps you've covered this, is that ability to have characters doing specific actions. I am creating a video made up of images + narration and I'm struggling to get my characters to do specific actions. Seems like they are always just posing. I know I need to do a deep dive into mixing (1) images as prompts and (2) describe and play with image weights. But perhaps you have some best practices you recommend. Thanks!
Thanks!
thank you!!
best tuto in youtube thanks you ❤
Glad you enjoyed it, thanks!
Hi Woollyfern!
I've been spending entire afternoons watching your videos and I absolutely love the quality of the content, especially the impeccable pacing. Wonderful lessons!
I have a question for you: are you monetizing your images in any way?
I'm asking because I would love to become a MidJourney Master (I've been an art director for 34 years and a generative AI enthusiast), but I still can't envision a business model where I can turn this into money.
Thank you for sharing such a value content! 👌
Thanks for the support!
I am not monetizing my images at the moment, but have looked into using images for print-on-demand products in the past. There are pros and cons to the process, but I think that's the direction a lot of people choose to go for monetizing AI images.
AMAZING TUTORIAL!!!!!!!
Thanks! Glad you found it helpful.
You are the best RUclipsr on Text-to-image stuff
Thanks for the support! 🙂
nice tutorial :) Thank you
Glad you liked it, thanks!
nice work.
Thanks!
Great video, clear as always and very encouraging! So you weren't having the distorted features problem? Your photo-realistic characters all look super consistent. (Mine have NOT. 😄)
Thanks so much! I still notice some distortion in the results that I got 🙂. Slightly thinner or rounder faces, bigger nose, hair not as thick or curly, etc. Some characters definitely worked better than others though. I notice the most significant distortion when using external non-midjourney images.
@@woollyferncreative A lot of my characters got decidedly more "cro-magnon." It was definitely not cute.
no matter what i do (even with 2 photo uploads) im getting someone older and NOT me and its always same character !
Yeah, unfortunately the Character Reference feature doesn't do well with externally uploaded images (I'm assuming that's what you're doing from how you worded it). I tried uploading a selfie and what I got didn't resemble me at all 😆. Cref works best with images made by Midjourney.
@@woollyferncreativeI have the same problem with an older woman that I created in midjourney, every time I try to recreate her older. I probably wasted about $30 working on one picture so far. 😢
@@woollyferncreative yes i did upload a selfie with clear bg, busy bg,white bg, black… 😅
I took selfie with my laptop,mobile,…
The funny thing is ALL the photos generated is same person no matter what prompt i use !!
But Its not me for sure
I even generated a model then tried to edit the face
Guess what? It gave me same person! (Not me!)
I assume this is because external reference images do not have a descriptive prompt attached, whereas MJ images can
@@Kelvinapplegateexperience is not a waste . Once u get it you will progress with greater knowledge. 30 60 90 all cheap for what you’ll learn & earn . Blessings!
Amazing tutorial. Can you provide guidance on how to maintain a consistent character when using photos (family portrait)? I've tried uploading single/ multiple pictures to create some artwork, but the --cref option isn't accurate with photographs.
Yes, --cref works best with images that were created by Midjourney.
With some effort though, you might be able to get a Midjourney AI version of each person you upload. Use /describe on the uploaded image to see how Midjourney describes the image of the person. Then, use your uploaded image as an image reference (URL at the front of the prompt), followed by a Midjourney-inspired description of the person. With this approach you might be able to create an image that will give more successful results when it is then used as a --cref. No guarantees, but it's one way you can try to improve your --cref results when using an image of a real person.
Thank you for a brilliant tutorial about using cref. - as well as your other videos. Something you didn’t mention, and is something I’ve also addressed directly with midjourney, is an issue when using cref and wanting a full body imagine. I have tried various prompts over various hours to no avail. The images midjourney produces distort the body proportion or they only show 3/4 of the body in the result. Any tips or suggestions would be appreciated. Thank you again.
Yeah, including "full-body" in a prompt doesn't always work. My first tip would be to full describe the outfit and include the shoes. This will try to force Midjourney to make a full-body image. I'll try to address things like camera angles and portrait vs full-body images in a future video.
Thank you for replying. Yeah, it’s something I do is describe the outfit and the shoes, but strangely the figures still come out disproportional.
great video, well explained. I'm creating a story in video format using images and I'm struggling a bit getting my characters to do specific actions. I'm able to use text prompts + an image reference that shows the action but I lose all the stylistic consistency for the story. Even when using SW of 1000. Any ideas? I realize this might be too much to ask of MJ at this point...maybe version 10? ;-)
@BigBrainBlitz if you are losing stylistic consistency, try using some of the images that do have the style you want as Style References with the --sref parameter. Showing specific poses and actions is a lot of trial and error, unfortunately. Hopefully, they will come out with a better way to do pose references someday.
@@woollyferncreative Yes--I used a reference image with the style and even pumped up the SW to 1000. It just refuses to pick up the style. Also played with image weight. Even at 0 it favors the position over the style--it really feels I have to choose the pose I want OR the style I want....maybe in version 10? ;-).
There should be an option that reduces the probability that the character is looking directly at the camera. E.g. --clook [0 - 100], where clook is camera look.
I know there could be way better combination of words than "camera look", but just an example. Couldn't think of anything lol. I know there are ways to get them to look at other objects, it's just way inconsistent and you slam through GPU hours trying to get to the image you want.
Yeah, I haven't done a thorough experiment to see which words or phrases give the most consistent correct/expected results. I run most of my jobs in relax mode, otherwise I'd burn through my hours in a few days or less 😆.
if cref already
can cref again with vary(region) because face doesn't match 100% then must cref again
What I think is really needed in the community is 1) style consistency - we have that. 2) character consistently - we have that. And 3) picture to pose, which….I feel like I keep struggling to find a good tutorial on that for Midjourney.
Without camera angles, without perspective and capacity to form the angles of each image, we can really generate proper compositions that are designed by us, the creator.
Do you think, if such a thing exists, you could combine these three things into a tutorial? Because I feel like that is the missing link from Ai creator to finished creation.
What MOST of us want from Ai is the illustration craft capabilities. But as creators we want to be explicitly involved in the creative process. If we can use pictures of posed models, toys, even ourselves that we can then upload into Midjourney and then apply that style consistency and character consistency, we will have arrived at the new era we’re all searching for.
Agreed! And I think we're heading that direction quickly. Midjourney will likely be entering the 3D space this year and with that should come the ability to better control camera angles. If I come across a more consistent way to get specific character poses, I'll definitely put that into a tutorial.
@@woollyferncreative love it! Keep us posted! :)
Hi Woolly, Wich soft do you use to avec your little Icon in the vidéo when you screen shot your video Please?
I use the software ShareX for screen recording and edit all of my videos in Adobe Premiere. Hope that helps!
This is a great tutorial. Thank you so much. I’m struggling with the pan function, when I pan left or right and try to input the information to add a person with a cref, it pans, but there is no new person in the shot…any tips?
Sounds like you are following the right approach. Try adding --chaos 15 to the end of your prompt when you are panning. This will force Midjourney to make the results in an image grid more different from one another. Sometimes this helps get it out of a rut, if it's not doing what you want. The chaos parameter takes values 0 to 100.
@@woollyferncreative I’ve managed to get a second person when I pan. Thank you so much! I’m now trying to pan to the other side to add a third person with cref, but the scene stays empty, do you know if its possible to use this function to add a third?
In theory, adding a third person should be doable by panning in the other direction. I haven't tried putting more than 2 crefs in a scene yet. Hopefully you can get it to work!
Thanks honey
thanks for the amazing tutorial but I tried so much and the character dont look the same with cref, I dont know why, Ive tried cref, characterweight but I dont know what Im doing wrong
Are you using a Midjourney image as your cref or an image that was not created with Midjourney?
For some reason the cref parameter isn't respecting my character's age. I placed the picture of a young woman as a character reference, which was made in MJ, yet for some reason it keeps de-ageing her to that of a preteen, even younger. It's not respecting the bone structure or maturity of the face at all. I even specified in the prompt that the photo is of "a woman". She looks eight or nine at most, down from perhaps twenty. I'm not sure what I'm doing wrong.
Ugh, that can be frustrating. Do you have any other images of that character that you can try? If not, try making some Variations of your original image and use those. Sometimes including multiple image URLs with --cref can help. I'm not sure how old the woman in your character reference looks, but maybe try including "college-aged woman". Or if Midjourney is really intent on significant de-ageing, perhaps "middle-aged woman" could result in a young woman 😄.
It seems like when you click pan right you can't edit the prompt anymore. Any other creative solutions for two characters outside of custom Zoom?
It should work. Make sure you have Remix Mode turned on in your settings. In Discord type /prefer remix, or go to your /settings.
@@woollyferncreative That was the problem. Thanks!
@woollyfern how did you get the option to change the text when panning right?? I've clicked it and it just generates.. no option to type in new info. Also I've held the button for a couple seconds, nothing. 🥺🤷🏽♂️
Oh! Found it! You have to type in /prefer remix to turn on remix mode... That'll give you the option to type in new info when pressing pan
You got it! Yep, you can also check your default settings anytime by typing /settings in Discord. Enabling Remix Mode allows you to change text input when you pan 👍
Now the issue I'm having is that, after typing in new info and adding the cref of the second character, I get nothing. Just the initial character on one side and the newly panned side is just a continuation of the scene 😮💨
It can be tricky sometimes. Try adding --chaos 15. Chaos values go from 0 to 100. Sometimes Midjourney needs a little push getting out of a rut when it's not doing what you want and adding chaos can help.
After several attempts (like 50)... I discovered that adding the words "standing next to her/his friend" and then pressing "pan" finally created the second person
What if i already have my own drawn character? How can i load it and use as the base character?
Yes, you can upload an external photo in Discord or on Midjourney's website and then use the link to the image as your --cref. You might not get an exact replica of your uploaded character, but should be able to get something similar.
There's a short section of this video ruclips.net/video/RYcGacwrnPA/видео.html that shows you how to upload to the website. For Discord, just drag and drop your image into a Discord channel, right click on the image, and copy link.
I uploaded the my desired MJ image into my discord. How do you get the link correct, it's not liking it.
You should be able to right click on your image and select Copy Link (on Desktop app). It will be a really long link. MJ will shorten it after you submit your prompt. Also make sure you have a space between --cref and your link.
@@woollyferncreative as mine is only coming up as copy message link or copy media link I don’t have any other alternative. As I’m using my iPad, do you think that this could be the reason?
Yes, I think the menu is slightly different depending on if you're using the desktop app, mobile, or web.
How do you reference scenery to get the same in multiple photos. anything besides sref?
Things to try:
1) Use an image as an image reference by including the image URL at the beginning of the prompt, followed by a space, and then your prompt text. This is different from --sref because you are not using the image URL with a parameter, it just gets added to the beginning of the prompt.
2) Alternatively, you can create the image of your scene and then use Vary(Region)/inpainting to select the area you want to change. For example, you could create an image of a park, select an area with Vary(Region) and add a character by changing the prompt text within the Vary(Region) window.
Hope some of that helps!
@@woollyferncreative thank you!
when i use the vary rigion,it doean't allow me to use cref ,I want to replace one of the two characters,so need to use vary region and put cref URL,but can't add the uil on,I don;t know why ,I used to use this function but now it stopped,anyone else here met such problem?
@lucy-917 double check the spaces and formatting in your prompt when you have the vary region window open. For example: my prompt text here --cref imageURL
Needs a space before the two dashes. No space between the dashes and cref. A single space after cref before the image URL.
@@woollyferncreative thanks very much,yes you are right,i just did as you said ,but it didn't work,I wonder if there are other restrict?i want to make a screen shot to you but can't add here,thanks anyway
Using Midjourney website, having real trouble creating a cref after dropping in a MJ created image. HELP PLEASE!
Can you provide a little more info?
ON DISCORD,
how can I get the URL of an image created ON DISCORD?
It constantly give me this message: "Image denied Request cancelled due to image filters".
An it works it does not recreate the same character.
To get the URL of an image you created with Midjourney on Discord or one you have uploaded to a Discord channel, you can right click the image and select "Copy Link". Then you can use that link to your image in a prompt.
Sometimes the image filters can be overly sensitive, and unfortunately the only way around it is to try a different image.
Re: results not looking like the CREF - if you are using a non-Midjourney image (ex. a photo of yourself) as a Character Reference, it does not work as well. It's best if your CREF image was created by Midjourney.
Hope some of that helps!
@@woollyferncreative Thank you very much, it turns out that when you want to copy the URL for the image, it should be copy from the image as it is on the chat (once is uploaded), because if you chose the option "open on the web" or if you make the image bigger by tipping on it and then copy the ULR it struggles to get the info. In addition to the fact that you get more accurate results IF first you create a value for that image through the "/prefer option set" function.
Thanks again!
- - cw 0 doesn't work for me at MJ website :| it makes error
What's the error? Make sure there are no spaces between either of the dashes and the cw. There should only be a space between --cw and 0. So like this: your prompt text --cref myURL.jpg --cw 0
@@woollyferncreative I commented before I realized I could hover mouse over the inscription to read the expansion :D I had to use picture as cref to use sw in prompt. It doesn't do well with my pictures. I want to try with illustration now.
Your video is very helpful; however, even though I have uploaded many images of my face from different angles, the generated image still doesn't resemble my face-I'm Asian, and the resemblance is only about 40-50%. After using face swap, it only improved to around 70%. I hope you can help me, or create a step-by-step tutorial on how to generate a face that looks exactly like mine, or find some way to contact you directly.
Unfortunately, --cref doesn't work well with character references that were not created by Midjourney. So it's a difficult task to try to use your own images as crefs with accuracy.
You can first try to create a Midjourney version of yourself using your selfie as an image reference. Once you have a Midjourney version of yourself, use that as a Character Reference.
Possible workflow: 1) run /describe on your selfie to see how Midjourney describes you, 2) submit a prompt using your selfie image URL followed by text using your /describe results for ideas (place URL at beginning of the prompt), 3) get a suitable Midjourney-created version of yourself from step 2 and then use THAT as a Character Reference.
It won't be perfect and it's a tedious process, but should give you a better chance at creating a suitable Character Reference of yourself.
Hopefully that's not too confusing. Best of luck!
@@woollyferncreative That's so helpful, thank you. Let me try
hi I really have an important question, whenever I create a male character they always turned out to have wrinkle lines on forehead how to make midjourney understand not to give that line? I just can't seem to figure it, you never have that problem as I saw your video, would you please help me out, thanks
Can you give an example prompt? What types of age-related words are you trying to use? Alternatively, you can try to use Midjourney's inpainting feature called Vary (Region) to select the forehead area and it will redo it.
@@woollyferncreative oh I didn't know inpainting vary could do that, will check it out .I use the prompt like like this
"a cinematic still of a man in his 30s is sitting in a coffee shop wearing black coat and pant with blue tie, coffee and laptop on the table, shot on blackmagic ursa mini pro, muted color grade."
That's just one eg. But even if I make a person in his 20s, same with forehead lines.
I would love to know if you face that issue from your end after trying my prompt. Thanks for the reply really appreciate 🙏
I tried your prompt. I do see that in some of the results, but for most of them it looks natural. I would try out the vary (region) tool or try using different words to give him a more youthful appearance. Sometimes MJ can exaggerate some descriptors such as age.
Hm... Strange...that for the most part it's realistic for you, I'm trying the vary, not the satisfying result but it's ok.
whats the prompt for 1:15? how did you introduced a man to the subject that’s awesome, please the prompt.
Did you have the wrong time stamp? The prompt used at 01:15 is shown.
@@woollyferncreative opps my bad 1:50
@@amitnatty4923 here you go :-)
Prompt: 35mm color photo of a barista standing outside of a coffee shop. she is talking with a male friend. --ar 3:2 --sref s.mj.run/pTBMUHtyJzY s.mj.run/otfzIs9AMR0 --cref s.mj.run/uckGT-_rmcw
where do I get a reference link?
You can paste in a URL to any image you want to use as a --cref. If you want to use a character you've created with Midjourney (recommended), open the image on Midjourney's website and click the 3 lines in the upper right. A menu will pop up and you can click Copy, then click Image URL. Then paste that image URL after the --cref parameter in your prompt. Make sure there's a space after --cref. Hope that helps!
@@woollyferncreative thank you!
0:26 - Why would I like a video... I have not seen yet...?
Midjourney Is not free