Thanks Nolan! I can see this being very useful when putting together a brand identity for a company. Being able to keep colors, etc. consistent with the brand.
Another great video. Like your positive energy, Nolan. One thing I discovered when trying out --sref is that when you prompt you can't put a comma after the subject, and before sref--or you get an error message. So portrait --sref, not portrait, --sref
The style ref option is CRAZY GOOOOOD I've spent the entire day yesterday using it and it was stellar !!! Especially when you use the same image as both the inspiration and then as the reference image with sref / s / iw. It's awesomeeeeee and SUPER CONSISTENT with every new prompt you create with the sref image in mind. Now I just need character consistency and I will be forever happy with midjourney :)
I’ve watch a hundred videos trying to figure out how to do this. This was the only one that worked and made it simple and straight forward. Thank you!!
I tried using the same image for both in the same prompt and I was very happy with the results. Bringing in random styles from the users feed is very funny anyway 😊
SREF is proably the thing that i asked the most for MJ to implement it and its freaking amazing! ive been inspired like it was the first time using midjourney! I cant wait for the CREF / character reference, that should make MJ a lot better! Would love to be able to use SREF and CREF in the same prompt!
dude i literally had this question, like what is the point of prompting anymore with --sref? this is the answer. love the 'sandwich vs milkshake' analogy. thanks man
@@FutureTechPilot so this is something im confused about--can you use sref and an image prompt at the same time? the announcement made it seem like no, but in my testing i think it makes a difference. would love to see a deep dive on using them both together, if indeed it makes a difference
I just used these tools today to regenerate a portrait. I noticed that you can also control the weight of image prompting by typing -iw and a value between 0.1 and 3👍🏼
Very good information! Simple and straight forward. What I'm trying to figure out though, is how this style reference is different from creating a prefer option set, which can sort of do the same thing. Might be a rookie question, but...?
I'm not entirely sure I understand the question, so let me know if my answer isn't what you were thinking of - but I would say that it's just another branch of consistency. You could use words to keep a style similar, you could remix the image to keep the style similar, you could image prompt for consistency, or you could include a style reference. They all result in different-looking generations, but they all accomplish the same idea. Hope that makes sense
I used a image with Asian buildings as my style but it didn't add asian buildings to the target prompt only the style of the image. Image prompt however adds the Asian buildings.
we're so close! now we just need a --dref (depth ref) feature and we're golden for composition. well.... for at least the next 9 months anyway until I get greedy again. ;)
Hello, what is the absolute best image you've generated with mj in your lifetime yet? Would be cool if you could show your top 3 of all time favorites (should be your original prompts/generations) in the next video, maybe at the end as an extra.
You explained this well but Im surprised you didn't see what would happen if you used both a style reference AND an image prompt. I havent tried it, maybe it wont let you
1:25 PRO TIP: The reason why is because you chose the wrong aspect ratio. You'd very likey have gotten full red jacket and shoes in all images with a portrait orientation image 👍
Yah very good point! That's probably the first thing to suggest when Midjourney isn't making what you want - change up the orientation of the generation!
😂 I didn't want to confuse people with that combination - because yes it's possible, and it's a little overwhelming to monitor all those possibilities. Let me know if you find any cool tricks!
Do you offer paid support? I'm looking for help with a specific midjourney project, which is to produce images in a given style using a reference image.
I have a couple of hours each week dedicated to consulting - check out my calendar and see if my schedule works for you! calendly.com/futuretechpilot/1-on-1-session - use that link if you'd like to pay with PayPal calendly.com/futuretechpilot/1-on-1-stripe - use that link if you'd like to pay with Stripe
Hi sir and thank you for the wonderfull information you always share I have a question how can we get a fully painted scene from our own rough sketch in midjourney v6 would you please help me with that and thank you
hi i find your videos worldclass but now I have a problem with sref: I always or very often get this Could not fetch image. Received status code 403 what can I do
Sounds frustrating. I'm really not sure of the solution. You should definitely ask around on one of the support channels on Discord! Sorry I'm not more help
Please tell me how to use only the style of an image or the colours of an image but get the exact prompt i put. İ used the sref but it gave me a result too closed to the reference image without taking into account my prompt. Also for example i want to make a wall art and I want to keep a specific painting style. But when i add the sref it doesn't give me the new art but the reference image art with some addition. For instance i want to create a collection of art that will looks like painted by the same person and the same colour scheme 😊
hi! im looking to change up a background of my own non ai created photo but can't find a solution for it. i could only utilise 'vary region' type on ai created photos for now or is there another way to do things? please help!
They're considering letting us edit non-midjourney photos, but you can't do that right now. You definitely should look into Photoshop - that program has amazing AI Generative Fill
@@FutureTechPilot Thanks for the reply. I've never understood Midjourney's mission if artists can't have the foundational influence over their art rather than take what they're given on a random. It feels like you're stuck playing with these things but not accomplishing anything. Some people have generated more than 5000 images, but nothing to show for it, no viral artist has come from the world of AI other than you guys that teach how to use it. All because realistically it's not useable if there's no control to consistency.
@@haljordan1575 Their mission is definitely just to create something that is fun to 'play' with. Their goal is to advance artistic expression - they're not really motivated by creating professional tools (even though they know the audience wants it)
Very good demonstration of these two features. But did you know that for Style Reference there are 4 versions you can use together as a parameter? This video explains the versions if you like to watch it => ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-stFGM7XdHqo.htmlsi=dg9TU8HArv9a8atu
I'm not sure I understand the question, but I'd say it's similar, not the same - and you can't use --sref and style tuning together right now (style tuning is only available in 5.2 and sref is only in v 6)
@@FutureTechPilot yeah. It’s a thing in stable diffusion where the ai tries to copy the “style” of the input image. It has different types of focus. Some ipadapters focus on faces. Some colors. Some lines. Some general image patterns (this is the most common one). The one thing to note. Ultimately it works on a square 1:1 image. Whatever you give it. It’s gonna crop it to 1:1. Usually in the center. Midjourney MAY be doing some complex logic to find where the image focus is and then crop there. (In comfyui and automatic1111 we do this manually)
lol, the guy doesn't understand it himself, so probably subconsciously trying to understand by giving us many "explanations" hoping as he said "one of them click" assuming we don't understand it neither. The answer is in the name of both features and doesn't even require explanation, at least not bigger than this: image prompt is recreating the image with additional content added in the prompt. Style reference only grab the style of the image, doesn't care about the content, otherwise there would be a girl on the couch but no, only the style of the image, colors, lights were taken. So sref is surely not a blend nor a smoothie (unless you believe the girls body parts are hidden in the couch), it is rather taking a dish, throwing away it's main element (in our image it's the content) and only take the seasoning, topping, fresh herbs to put on the top, any other decoractions, etc. Besides that you can use both at the same time even without text, if there is a photo of a girl in the prompt image and there is a dog in a watercolor style as sref, you will get a girl in watercolor style, no dog. If u use the watercolor dog as image prompt and add sref the photo of a girl, you'll get a photo of a dog, no girl, no watercolor anymore. So it's surely not a blend. Blend would be humanoid female dog.
So you start with a mean-girl takedown, then body-slam yourself into a bowl of word salad. I think most people would prefer Nolan's well thought out approach and appreciate his generosity over your verbosity.
@@morpheus2573 I don't create the video, so I can only use words. I know it's too much for you to follow and visual representation seems easier to understand. The problem is, his explanation is wrong. I only used 2 sentences to explaing it all, the rest was relating to what he said in the video. So my "verbosity" is only referring to examples your "generous" author of the video gave, to point out faultiness of his "well thought" understanding of sref image. The simplest explanation is: the sref only takes style from the image. As you already haven't understood once what is the style of the image, I'll describe it to you: it grabs the overall composition, read if it's a photo, anime, animation, 3d render, oil paint, watercolor, etc or combination of these styles. It also reads the colors, the lights, occasionally some decorative elements. Most importantly, it skips the content of the image. It doesn't care if there is a woman or the dog in it, it completely skips it. The explanation in the video this guy gave us, he says it is a blend of the sref image with a text prompt is totally wrong. A blend wouldn't skip the content of the sref image. Read my response to the video author for more details.
@@FutureTechPilot The problem is not in the "many explanations" but that each of your explanations of sref is wrong. A first example you show an image of a pattern. To explain how STYLE reference image works, using a pattern image, that is literally made of style and decorative elements (considered by ai as style) only, is completely illogical and is simply invalid example. You say "style is transferred through and subject get's blended together" - this is the closest you got to understanding it, and it's only becouse the sref image is a pattern, but still not close enough. Becouse they are not even blended together, in the image prompt example of man wearing a red jacket and gold shoes the image and text prompt are blended together more, since the pink parts of the image become red matching the jacket or trousers, and yellow parts of the image become gold, matching the shoes. While sref image + 'man wearing red jacket' only transfer the style to it, not blending together, not taking anything from it. Second example, the dragon + sref of lego image is grabbing only the colors, overall compostion and the size of the dragon (or rather size of the scene). The size of the scene make it a figurine, doesn't make it a toy and neither of those is a toy figurine, just a figurine. "Blend come toghether a little more", yeah, when you prompted it has to be 'dragon made of lego' xD, lol. Blend come only after putting it in the prompt, there is no slightest piece of lego in it until you clarify it in the prompt, only style is trasnferred, there is zero blend before that. 3rd example the couch is absolutely not a blend. There is no girl at all after using sref image (unless you believe she's blended into the couch which we can't see). The couch prompt only grabs the style, so there is no girl, there is strong sunlight grabbed, strong contrast level grabbed, the shadows, product photography, unsplash, everything that is related to style besides the actual objects in this picture, that is no girl, no coat, only the lubricity of the coat. The last example clearly shows the image prompt with text prompt create much more of a blend than sref with text prompt. Ninja turtle and batman are literally blended toghether, the character consists of both, either it looks like a turtle but have a batman cape, or it looks more like batman with the mask, but you can clearly see the turtles face behind it and greenish body. It couldn't be more obvious. Image prompt + text prompt is literally like the text prompt that created that image + text u added, which is way more of a blend between these two, like normal text prompting is usually the blend of all of the things u put in there. While sref only grabs the style of the image, so there is absolutely zero batman in these results, only the style of the image, and turtle literally swap the batman from it's place, grabbing only the style of the image. Now as I clearly described you what is sref - style reference skipping all the content in it, you better rerecord the video with the proper explanation, not misleading your viewers. "Sref doesn't care who was in the original image" "only black and white manga inspired visuals" - finally at the end you said something that makes sense. Unfortunetly all of the multuplication, blending and smoothie nonsense throughout the entire video is all wrong.