Patreon Community (free & paid tiers) ➡www.patreon.com/Woollyfern Midjourney Guides ➡ woollyfern.gumroad.com many are free and paid ones are included with some Patreon tiers ⚡What do you think of the new Character Reference (--cref) feature? Any specific topics you'd like to see in my upcoming Cref deep dive video?
I would like to watch a video of how to bring our older jobs without any changes, from versions 4-5 to be able to work and extend them with the version 6 New features. Also, is there any online tool that can fix bad images from Midjourney such as missing heads on animals or many-fingered hands for people and strange incomplete faces?
I'm finding it working better when the reference image features multiple "looks" of the character, like a model's comp card. Without it my --cref images look like my initial character has had some bad cosmetic surgery that distorted their features. They're recognizable...but not great looking. Could you consider a deeper --cref guide video? And also thanks for giving /describe some love, i think it's a great feature!
I use characters in stories for online training. The previous methods for obtaining consistent characters, making character sheets, were imperfect and required a lot of photoshop work and upscaling in Topaz. Now I am able to make beautiful detailed consistent characters in various positions to tell the stories, although I still do some photoshop work. The difference is night and day and has speeded up my workflow as well as adding a new level of quality in character detail.
This is super helpful. I had no idea the /describe prompt existed. ANd yes, I would love more on the Character Reference. It's working for me, although it's very a iterative process to get to what I want. One problem I have is no matter how I prompt it, it struggles to give me a full body shot.
Yeah, /describe is not well-known! I've noticed with Character Reference that it can help to start with a full-body reference if you intend to create a full-body image. However, even without a Character Ref, full-body images can be a challenge. I'll try to include this in my upcoming video.
Hi and thanks in advance for any assistance. I'm in the process of illustrating/writing a graphic novel. Have spent ages creating a portrait of the main character that I'm now happy with. The issues I'm having are two-fold. 1) Attempting to generate a series of images of that character from the back and profile and 2) Creating either scenes with them in or not that have exactly the same aesthetic. As an example of what's going wrong: If I write a prompt that describes a scene, and then include the --cref of the character, the results are more photographic with certain characteristics included (facial scars as an example) but the sketched aspect of the image has gone. I've even tried using the same URL as as both a --cref and --sref to try and keep all intact.
Using your cref as both a cref and an sref is a great approach to try to solve the problem. Midjourney often defaults to a more photographic look when a style is not specified in the prompt text nor as an sref. Some sref's need a little boost in the prompt text to bring them to full strength, so it might be helfpul to add a couple words that describe the style you are going for. You can also try increasing the strength of the style reference with the --sw parameter. If you haven't yet, I recommend checking out my video on camera positioning prompts (ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-4vLU6KM9KdQ.htmlsi=se6UQVnE5prvuUZu) and my cref deep dive (ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BOgqKKRT7IE.htmlsi=xI5Ypdotw_A3r5p0). Hope that helps a bit!
Many thanks for your suggestions and I will check out the videos mentioned. I'm getting there (slowly) but a little frustrated in crafting a character, and getting it spot on, only for the rest of the process to slowly but surely pull it apart (but don't want to sound ungrateful when the software is so powerful).
4 месяца назад
I wasn't happy with - -cref because i didn't know - -cw !! This will help a lot, thanks a bunch!
Here's my latest video on cref ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BOgqKKRT7IE.html. Let me know if there's something specific you're looking for that I didn't cover in it. Thanks!
One weird thing which i found with cref is that,it works ok and perfect for faces belonging to a certain ethinicity. In other cases,it produces very different images with no match to the cref image. Also it seems to work better on certain aspect ratios, for me 16:9 works worst.
I wish I did more than faces. I need consistent objects, or designs in general as well. Like if you make an image with a book and a cedar chest, I want to make images with the same exact book and chest.
Yes, I've noticed this to some degree as well. Specifically, I notice that Midjourney will often make the faces slightly more rounded/fuller compared to the --cref that I provide.
Yeah, you shouldn't need to seed anymore to try to get consistent characters. I share a lot of examples in this video ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BOgqKKRT7IE.html
Fantastic highlight video, walk through and examples! Luv your vid n narration YT style And SREF and CREF are priceless best MJ prompts - and I was just playing with -sv {1,2,3,4}
YOU ARE SO GREAT! Question, on the website, I'm having trouble creating the URL from an image I've already created in MJ to that I can use cref. HELP, please.
To get the URL of an image you want to use as a cref, open up the image in Midjourney. Then, click the 3 lines in the upper right and a menu will pop up. Click Copy. Then Click Image URL. Now you can paste that image URL after the --cref parameter in your prompt. Make sure there's a space between --cref and your URL. Hope that helps!
When I tried to use --cref with an externally produced portrait, the function destroyed the face shape and even the ethnic origin of the model. Now I plan to coax Midjourney to produce an exact likeness of the original with a truckload of different prompts, in the hope that I can have a Midjourney-made portrait true to the origin, and use it as a model for further --cref experiments.
Yes, --cref is not meant to work with non-Midjourney images. I made my own attempt at uploading a selfie and the results looked nothing like me 🙃. Your plan sounds like a good one, good luck!
Thanks! Here's the prompt for the end screen woolly mammoth :-) "adorable woolly mammoth standing in the forest. to the right is a vintage tv with a glowing screen. --ar 16:9 --sref s.mj.run/1sYPm3Gt96k --sw 20"
Thanks! Here you go 🙂: adorable woolly mammoth standing in the forest. to the right is a vintage tv with a glowing screen. --ar 16:9 --sref s.mj.run/1sYPm3Gt96k --sw 20
It's pretty good at keeping a highly similar character across images, but not identical. With a little extra effort, it might sufficient for your needs. I recommend watching my cref deep dive video where I go through a bunch of examples ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BOgqKKRT7IE.htmlsi=vq6ix8nUcXvq2YWP
Wow! These are great updates News. Thank you for sharing these with us. You have a new subscriber to your channel. By the way I think that you have confused Midjourney here @ 7:05 by adding the Stylize 1000, and too many parameters.
Thanks! You might be surprised to learn that a higher stylize value with --cref can sometimes help improve coherence and make body poses more natural. I learned this during a recent Midjourney Prompt Craft event. With the original prompt that I was using @ 07:05, I think my text was conflicting too much with the type of image I wanted. Using --stylize 1000 was part of my workflow in trying to discover a way to fix my images. Alas, it was not the definitive solution.
Hello! I am your subscriber and I like all your videos 😍. Very useful, thank you! But my question is off topic. I'm just starting to run my channel and I have some questions about the design. Please tell me how you add display of your Featured playlist in the description under this video 😊? If possible, share this secret, I will be very grateful, it is very important to me 🙏!
Have been hours and hours using the CREF, but its a nightmare in terms of detail, if you want consistent clothing etc in terms of colour, style and patterns. Can't create a multiple pose character sheet for a children's book and get enough original images to use it gives me 3 or 4 on zoom out and changes styling of the clothing even if you use --seed, --cref -- cw, I've tried all variations and have had to specify certain colours exactly and edited out the patterns in canva. Bit frustrating but it seems even with a midjourney created image it does not keep everything the same and just change the pose which is what I need. Is there a trick to getting a bigger number of multiples in a character sheet? Thanks
@sharonkamelonline Yeah, I have been running into this as well as I'm putting my --cref deep dive video together. --cref is not meant to be a character duplication tool, yet, so it won't get all of the details right. Cref can, however, be used to make characters with a similar likeness to the reference image. When the details matter, they need to be included in the prompt text, but I realize even that's not foolproof. I don't have a lot of experience making character sheets, but I would try using different aspect ratios to force Midjourney to utilize the space differently. You could also try including something like "character turnaround sheet with 10 different poses" in the prompt.
For prompt inspiration you can use /describe on an image that is in the style you are trying to convey. Alternatively, search "caricature" on Midjourney's website to see how others structure their prompts. Then, start experimenting. Try using the character as a reference with --cref. Try using an image that shows caricature style as a style reference with --sref.
--cref works best for characters. It's possible Midjourney will have an "object reference" parameter in the future, but I don't think they are focused on that right now.
You're correct that Character Ref was released a couple weeks ago. This video aims to provide an overview of multiple new/updated Midjourney features that have been rolled out recently, the latest being the March 21st updates to Style Reference. Thanks for the feedback!
I don't get it. ALL of these examples with character references look fantastic. ALL of mine, even if I copy the prompt, NEVER Look like the reference, EVEN IF I isolate that person and it's a really good photo of them.
Are you using a reference image that was made by Midjourney? It does not work well with non-Midjourney images. I've also noticed that it works better with some reference characters and not others (different face shapes, hair, ethnicities). Even in the examples I share in the video, I notice the imperfections. Not all of my results turn out good every time. Sometimes it takes a few re-rolls. I made a deep dive vide on --cref that you might find helpful ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BOgqKKRT7IE.html.
Please check out my Character Reference Deep Dive video for information on multiple characters ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BOgqKKRT7IE.htmlsi=QuNGmVkLsfTGMffC
