🎯 Key Takeaways for quick navigation: 00:00 🎨 Identify the art style you want for your AI art, and for this demonstration, the creator is inspired by the art style of Digimon: The Movie from their childhood. 01:09 🚀 Use Midjourney's new feature, "/tune," to fine-tune the style with reference images. Drag and drop image links into the prompt, ensuring a space between each link. Include relevant keywords like "anime aesthetic." 01:52 🔍 Create a tuning test with 16 style directions. Experiment with different options until you find a style you like. Copy the entire prompt, including image links, and run it with various stylized values (e.g., S400 and S40). 02:49 🔄 Enable "remix mode" and "high variation mode" in the settings to avoid an average of all image prompts. Delete image prompts and style code from the chosen style. Lower stylize value for closer adherence to your prompt. 03:44 🔄 Continue the remix process with different characters or scenes, maintaining consistency by using the same base style. Input new subjects, such as Wonder Woman, and generate AI art with a consistent look. 04:54 🖼️ Include descriptive words like "full body," "wide angle lens," and "action pose" to refine the generated art. Exclude specific words if needed to avoid unwanted elements. Adjusting parameters helps achieve the desired consistent style. 06:29 🔄 Recap of the workflow: Tune style with image prompts, select a style, generate the entire style, enable remix mode, use the variation button, delete image prompts and style code, use lower stylize values, include new subjects, and continue the remixing process for a consistent look. Made with HARPA AI
This is going to change the game when it comes to consistency , too often it's a massive gamble to just type an artist name or anime film directors name and get good results. Either the person is way too niche or the AI consumes generic names and it misses with the results, having the ability to train the AI IRL with the exact taste you want and have remix guide the slot machine is really a positive game changer especially for those of us who want to master this machine. Thank you so much for bringing awareness
An excellent followup on /tune. Tuner is a difficult feature of Midjourney to make useful, and this approach is another good one to drop into the toolbox.
You lost me at 2:07 when running the prompt in different stylized values. Do we simply add "--s 400" at the end of the prompt like your video shows and hit enter? Doesn't work for me. Can you please tell me what I'm missing here?
@@FutureTechPilot When I add " --s 400" after the prompt and hit enter, it simply reposts my two original images with my prompt on top. It does not generate any new images based on the styles I chose from the mid journey tuner.
@@Ara-qn8sl I gotta say, I'm super confused. If you want to take some screenshots, you can send them to me - futuretechpilot@gmail.com - and I'll take a look
Thanks. Looking forward to trying this technique. But I'm a little confused about not including the style code in the prompt. And yes, it is counterintuitive. I get that you don't want it to act as a "force multiplier," but I thought the whole point of style tuning was to create a bunch of style codes you can create then add to a prompt to overlay it with the style of your choice. This method is a work around of that approach. If this is really necessary, you're essentially saying that the entire style tune concept doesn't work as intended. If so, MJ needs to provide clarity around their best use practice.
hmmm. maybe I could have been more clear. So, the style code definitely maintains a particular look, but in my testing, it *seemed* to interfere with the 'remix' feature ... like, once we have a style we want, we try to 'remix' to a different character, but the style code was also acting on the remix. Basically, after testing, it seemed like removing the style code helped the remix feature maintain the look I wanted. Hope that explains it a little better
@@FutureTechPilot I hear ya. There's no issue with your clarity, and your work around makes great sense. The issue is that MJ have released a "pushmepullyou."
Ok, that was awesome! And I love the fact that you did that quick recap at the end. Very helpful. 💯 And I agree about going over 16 variations. You can burn through those hours _really quickly_ . Thank you for this. 👍
@@FutureTechPilot This workflow has been sooo useful! I've been feeding my own gens as references to keep tuning/generating non-stop and I'm getting marvelous and sooo many eye-catching weird results! Like you say:'goosebumps!'🤘🏽
Awesome tutorial. Very insightful Keep up the great work. /Tune is such an amazing feature. Listened to Tunein Tuesday Midjourneys Discord. Wow! crazy amounts of info disclosed.
I've never heard of that happening! I suggest you mention it in the 'member support' channel on Discord. Maybe someone from the team can help you out. Honestly I would say just run the tuner again
@@ianalvord3903 look near the top on the left side - you should see 'home' '3 event's and 'channels and roles' - click channels and roles and then click 'browse channels' - you should find member support there
@@FutureTechPilot Thanks for he help. Turns out Midjourney was not a fan of one of the pictures I imported and refused the whole batch because of it. Took a while to find the one it didn't like.
U are the man! Fantastic vídeo. Glad I waited couple weeks before diving into custom styles. Question: If you want a lot of the same style established why are using -s 40? Why not more of the style added in?
So this is all based off of testing, and any stylize value higher than 100 really started to ignore the subject of the prompt. Remixing helped keep the style, while --s 40 helped make sure Midjourney followed my prompt, in this case, created the character I asked for
This process is exactly what I have been searching for! Thank you so much for posting this. Now that we have V6, has anything changed or improved? Will this procedure work the same in 6? On a side note, the 2x and 4x upscalers are gone in 6 - re[placed by the upscale subtle and creative - are they 2x or 4x?
Upscalers 2x and 4x didn't make it to V6, but might come later. Right now Subtle and Creative work slightly different but get it up to probably 2x. Unfortunately style tuning isn't available in Version 6 yet but this same basic idea might work (image prompt and then remix)
Hmmmm, I used tune/ in 6. It did run but the results were terrible. None of the 32 options were acceptable. I’ll try again with a new selection of reference images.
You can bring back the upscale by vary region'ing a small bit of the image, but changing the mode to 5.2 in the prompt. that converts it to 5.2 and brings back the upscaler
Thank You for your help! I keep getting an error when trying to use style or s value "Unrecognized parameter(s): style, s", not sure what doing wrong. I get up to the part where I select from the grid of 16, then copy that code and put it in with "--style (the custom code)" and "--s 40". Thanks for your help
I've seen this problem before but I cant figure out how I fixed it .. I'm thinking, try putting the '--s 40' before the '--style code' ... it's something like that
Thank you for sharing this! I'm trying to create a comic with 8 characters. I need to replicate them in different scenes and positions. Do you have any content that could help me with this? I'm struggling with creating consistency between the characters. The character "a" can't be replicated in different positions or scenes, for example. Thank you very much!
There are some things you can try, but I don't want to give you false hope. In my opinion, there isn't a way to get consistent characters like that - nothing you'll be satisfied with.
How about using the generated "tuned" style codes as sort of "presets" that can be applied to new prompts??? Is that just the same as using this remix method?
That's slightly different than remixing - I mean, you can try both and see the difference yourself - because it's hard to explain. But you can't use those style codes in Version 6, just a reminder!
But then why using the tuner at all ? Wouldn’t you have the same result by using your five images without the tuner style and then do variations like you did?
You're not entirely wrong .. but the /tune helps with the style because it shows you all the different options you actually have for that prompt - whereas image prompting alone will only leave you with 1 outcome
Is there any way to reference a scene instead of a character? --sref just copies the style. What if i want variations of the same scenery? Like my house for example, same house but maybe it looks older.
Any idea what 0/0 Questions Answered means. I've create two separate tuners twice and both times when I go to the URL the screen flickers and then stops with nothing but 0/0 Questions Answered displayed on screen.
I could be wrong about this ... but I think it means the prompt you created was inappropriate? In the sense that, the pictures that it made were flagged NSFW. Again, I could be wrong about that. You should ask on the 'Member Support' channel
@@FutureTechPilot Thanks for the reply. The next day I did it a third time and it worked to the extent that it generated style options without the errors I was receiving. What I was working on trying to do was to create a style based on a character that I created in DALL-E 3 in ChatGPT. The character in question is an anthropomorphic canine female and all of the images I uploaded into MJ were that same character in the same style just posed differently. Looking at the viewer, turned to the left, turned to the right, facing away from the viewer, etc. The third attempt I tried was I took just one image - not multiples, just one - and used tuner to try and generate a style based on the way the character looks. MJ tuner created a batch of 16 images but, unfortunately, none of them even remotely resemble what I was trying to create. Having dealt with A.I. art tools for close to a year now I'm sure the character I was trying to upload and use was not NWFW in any way. Again, an anthropomorphic canine fully dressed, to the point of a turtleneck shirt on the character. No skin "exposure" or anything of that sort. The character images I uploaded in every instance were fully clothed with no inappropriate content and nothing in the prompt other than the style name I was assigning to MJ tuner. I'm not expecting you to do anything else with this. I was just passing this along for informational purposes in case it is useful for someone else having similar issues. As a side question, has anyone else in the last 24 hours after Sam Altman was given the boot, ChatGPT has been running squirrely?
Well ... that's not easy to do that the moment ... - you can take a look at this video - th-cam.com/video/ICh9ose7tGs/w-d-xo.html - and try the techniques in there, but there is no quick solution built into Midjourney right now
when I do /imagine it gives me an error code and I followed all the steps! I'm super confused if I did something wrong or maybe it is my images I put down?
"DIGIMON THE MOVIE!" "As an AI language model, yes I am capable of re-rendering Cowboy Bebop and Akira in the style of Digimon the movie, but I beg you sir. They are classic anime icons. It feels like... heresy, like a crime against..." "D I G I M O N ! ! T H E ! ! M O V I E E E E ! ! ! ! ! !"
After tuning, do you copy the entire prompt into discord just as it is, or so you type in /imagine first and then paste the entire prompt?? It seems like you must have typed in /imagine first but you don't mention that in the video. Thanks
Couldn't you do the same thing just by asking MJ to render a "cartoon scene like Digimon." And then, when you find a frame you like, just remix based on that one.
Yeah I guess, but Digimon here was just an example. If you wanted a different style that was difficult to put into words, your best bet would be to tune it with image prompts
🎯 Key Takeaways for quick navigation:
00:00 🎨 Identify the art style you want for your AI art, and for this demonstration, the creator is inspired by the art style of Digimon: The Movie from their childhood.
01:09 🚀 Use Midjourney's new feature, "/tune," to fine-tune the style with reference images. Drag and drop image links into the prompt, ensuring a space between each link. Include relevant keywords like "anime aesthetic."
01:52 🔍 Create a tuning test with 16 style directions. Experiment with different options until you find a style you like. Copy the entire prompt, including image links, and run it with various stylized values (e.g., S400 and S40).
02:49 🔄 Enable "remix mode" and "high variation mode" in the settings to avoid an average of all image prompts. Delete image prompts and style code from the chosen style. Lower stylize value for closer adherence to your prompt.
03:44 🔄 Continue the remix process with different characters or scenes, maintaining consistency by using the same base style. Input new subjects, such as Wonder Woman, and generate AI art with a consistent look.
04:54 🖼️ Include descriptive words like "full body," "wide angle lens," and "action pose" to refine the generated art. Exclude specific words if needed to avoid unwanted elements. Adjusting parameters helps achieve the desired consistent style.
06:29 🔄 Recap of the workflow: Tune style with image prompts, select a style, generate the entire style, enable remix mode, use the variation button, delete image prompts and style code, use lower stylize values, include new subjects, and continue the remixing process for a consistent look.
Made with HARPA AI
This is going to change the game when it comes to consistency , too often it's a massive gamble to just type an artist name or anime film directors name and get good results. Either the person is way too niche or the AI consumes generic names and it misses with the results, having the ability to train the AI IRL with the exact taste you want and have remix guide the slot machine is really a positive game changer especially for those of us who want to master this machine. Thank you so much for bringing awareness
'remix guide slot machine' - exactly. You definitely won't always win, but you should be able to win eventually lol
An excellent followup on /tune. Tuner is a difficult feature of Midjourney to make useful, and this approach is another good one to drop into the toolbox.
Cheers pal!!
Pure Gold! Thank you!!
Cheers pal!
Thanks for the great video. Going to go try it now…..🤞
Yes I hope it works!
You lost me at 2:07 when running the prompt in different stylized values. Do we simply add "--s 400" at the end of the prompt like your video shows and hit enter? Doesn't work for me. Can you please tell me what I'm missing here?
Yeah put --s and then any value between 0 - 1000 ... what does it say when it's not working for you?
@@FutureTechPilot When I add " --s 400" after the prompt and hit enter, it simply reposts my two original images with my prompt on top. It does not generate any new images based on the styles I chose from the mid journey tuner.
@@Ara-qn8sl I gotta say, I'm super confused. If you want to take some screenshots, you can send them to me - futuretechpilot@gmail.com - and I'll take a look
@@FutureTechPilot I just sent you an email with the subject "Midjourney Style Troubleshoot". Thanks for your help!
Legendary human you make life better you are appreciated very much
And thank you for the kind words!
Thanks for posting this, i'm using openart at the moment but will look into midjourney in future because of your video.
I hope it helps!
You blew my mind with this one.
I'm glad you think it's as cool as I do!
Thank you so much. This is Gold. You've gained a well deserved subscriber. So happy I found your channel.
And I really appreciate the support!
@@FutureTechPilot you're welcome 😊
Now that you can create tunes on relax mode wouldn't 128 be the best option?
Maybe! But it can be kinda overwhelming to choose between that many images
@@FutureTechPilotOh yeah that for sure! And that's why I love it haha. One single grid with infinite codes!
Thanks. Looking forward to trying this technique. But I'm a little confused about not including the style code in the prompt. And yes, it is counterintuitive.
I get that you don't want it to act as a "force multiplier," but I thought the whole point of style tuning was to create a bunch of style codes you can create then add to a prompt to overlay it with the style of your choice. This method is a work around of that approach. If this is really necessary, you're essentially saying that the entire style tune concept doesn't work as intended. If so, MJ needs to provide clarity around their best use practice.
hmmm. maybe I could have been more clear. So, the style code definitely maintains a particular look, but in my testing, it *seemed* to interfere with the 'remix' feature ... like, once we have a style we want, we try to 'remix' to a different character, but the style code was also acting on the remix.
Basically, after testing, it seemed like removing the style code helped the remix feature maintain the look I wanted.
Hope that explains it a little better
@@FutureTechPilot I hear ya. There's no issue with your clarity, and your work around makes great sense. The issue is that MJ have released a "pushmepullyou."
Hey, you're a lifesaver! You just spared me hours of hassle. Always worth tuning in to your channel
Happy I could help, pal. Thanks for the comment!
Another brilliant insightful process tutorial. Thank you. Jim
I appreciate the comment!
tune is AWESOME!
Yeah too many rabbit holes makes my head hurt though lol
This is a super fun & very insightful experiment. ❤ thanks for sharing this info.
Cheers pal, I appreciate the comment!
Ok, that was awesome! And I love the fact that you did that quick recap at the end. Very helpful. 💯
And I agree about going over 16 variations. You can burn through those hours _really quickly_ . Thank you for this. 👍
haha honestly, I thought this little workflow wasn't the easiest to follow, so the recap became a necessity
Thank you!
And thank you for the support!
/tune is certainly life changer
the possibilities are mind boggling
Great tutorial as usual!🤘🏽
I appreciate the kind words!
@@FutureTechPilot This workflow has been sooo useful! I've been feeding my own gens as references to keep tuning/generating non-stop and I'm getting marvelous and sooo many eye-catching weird results! Like you say:'goosebumps!'🤘🏽
@@Juan_Teppa wow, that's a cool idea!!
@@FutureTechPilot 👍🏽🔥
Awesome tutorial. Very insightful Keep up the great work. /Tune is such an amazing feature.
Listened to Tunein Tuesday Midjourneys Discord. Wow! crazy amounts of info disclosed.
Hey drop some cool knowledge bombs if you have any!
Super helpful!
I'm glad to hear that!
you are a saviour!!!
I'm glad I could help!
So awesome Bro -- I have been thinking how much more powerful /tune would be if you could use images, and you went and did it
haha and I think it worked out well enough!
1:48 something has gone wrong for me. My page has no pictures. It's just a message saying "0/0 questions answered Try to answer a few more questions!"
I've never heard of that happening! I suggest you mention it in the 'member support' channel on Discord. Maybe someone from the team can help you out. Honestly I would say just run the tuner again
@@FutureTechPilot I tried it again with the same error. I'm contacting support.
@@ianalvord3903 look near the top on the left side - you should see 'home' '3 event's and 'channels and roles' - click channels and roles and then click 'browse channels' - you should find member support there
@@FutureTechPilot Thanks for he help. Turns out Midjourney was not a fan of one of the pictures I imported and refused the whole batch because of it. Took a while to find the one it didn't like.
@@ianalvord3903 oh haha okay. well, at least we found the problem!
U are the man! Fantastic vídeo. Glad I waited couple weeks before diving into custom styles. Question: If you want a lot of the same style established why are using -s 40? Why not more of the style added in?
So this is all based off of testing, and any stylize value higher than 100 really started to ignore the subject of the prompt.
Remixing helped keep the style, while --s 40 helped make sure Midjourney followed my prompt, in this case, created the character I asked for
Once we create the styles, can we save those styles in Midjourney in order to pull them up whenever we need to?
Yeah check out this video here - th-cam.com/video/Xf8UOwTX-FQ/w-d-xo.html - one of the steps is saving a style as a shortcut
Genius. I really need to go back to a better subscription to try all of this !
haha enjoy!
Thank you
And thanks for the comment!
Thank you 🙏🏽
🤝 cheers pal
This process is exactly what I have been searching for! Thank you so much for posting this. Now that we have V6, has anything changed or improved? Will this procedure work the same in 6? On a side note, the 2x and 4x upscalers are gone in 6 - re[placed by the upscale subtle and creative - are they 2x or 4x?
Upscalers 2x and 4x didn't make it to V6, but might come later. Right now Subtle and Creative work slightly different but get it up to probably 2x.
Unfortunately style tuning isn't available in Version 6 yet but this same basic idea might work (image prompt and then remix)
Hmmmm, I used tune/ in 6. It did run but the results were terrible. None of the 32 options were acceptable. I’ll try again with a new selection of reference images.
@@dangilbert493 tune still uses 5.2
You can bring back the upscale by vary region'ing a small bit of the image, but changing the mode to 5.2 in the prompt. that converts it to 5.2 and brings back the upscaler
really awesome, thanks bro.
Cheers pal, thanks for the comment
THANK YOU!
Cheers pal!
Thank You for your help! I keep getting an error when trying to use style or s value "Unrecognized parameter(s): style, s", not sure what doing wrong. I get up to the part where I select from the grid of 16, then copy that code and put it in with "--style (the custom code)" and "--s 40". Thanks for your help
oh and I looked up the problem and switched to medium variation and MJ ver3 as suggested on the 'wealthquint' website, but that didn't help
I've seen this problem before but I cant figure out how I fixed it .. I'm thinking, try putting the '--s 40' before the '--style code' ... it's something like that
@@FutureTechPilot Thanks so much! Perhaps I need to use /imagine instead of /tune after selecting from the variations page?
@@julienlandau1072 oh. yah lol that's definitely it. /tune is just for getting styles - /imagine is to make new images
Having same issue here. Were you able to resolve the issues? Pls help 🙏
Thank you for sharing this! I'm trying to create a comic with 8 characters. I need to replicate them in different scenes and positions. Do you have any content that could help me with this? I'm struggling with creating consistency between the characters. The character "a" can't be replicated in different positions or scenes, for example. Thank you very much!
There are some things you can try, but I don't want to give you false hope. In my opinion, there isn't a way to get consistent characters like that - nothing you'll be satisfied with.
How about using the generated "tuned" style codes as sort of "presets" that can be applied to new prompts??? Is that just the same as using this remix method?
That's slightly different than remixing - I mean, you can try both and see the difference yourself - because it's hard to explain. But you can't use those style codes in Version 6, just a reminder!
Nice❤. Thanks!
🤝 happy to share
awesome video, does anyone know how to get consistent composition??
I think we will have to wait a few more months for them to develop that tool!
very nice as always
Thanks!
now I want to watch digimon
hahah you and me both
But then why using the tuner at all ? Wouldn’t you have the same result by using your five images without the tuner style and then do variations like you did?
You're not entirely wrong .. but the /tune helps with the style because it shows you all the different options you actually have for that prompt - whereas image prompting alone will only leave you with 1 outcome
Is there any way to reference a scene instead of a character? --sref just copies the style. What if i want variations of the same scenery? Like my house for example, same house but maybe it looks older.
I wish! maybe one day
Any idea what 0/0 Questions Answered means. I've create two separate tuners twice and both times when I go to the URL the screen flickers and then stops with nothing but 0/0 Questions Answered displayed on screen.
I could be wrong about this ... but I think it means the prompt you created was inappropriate? In the sense that, the pictures that it made were flagged NSFW.
Again, I could be wrong about that. You should ask on the 'Member Support' channel
@@FutureTechPilot Thanks for the reply. The next day I did it a third time and it worked to the extent that it generated style options without the errors I was receiving.
What I was working on trying to do was to create a style based on a character that I created in DALL-E 3 in ChatGPT. The character in question is an anthropomorphic canine female and all of the images I uploaded into MJ were that same character in the same style just posed differently. Looking at the viewer, turned to the left, turned to the right, facing away from the viewer, etc. The third attempt I tried was I took just one image - not multiples, just one - and used tuner to try and generate a style based on the way the character looks. MJ tuner created a batch of 16 images but, unfortunately, none of them even remotely resemble what I was trying to create. Having dealt with A.I. art tools for close to a year now I'm sure the character I was trying to upload and use was not NWFW in any way. Again, an anthropomorphic canine fully dressed, to the point of a turtleneck shirt on the character. No skin "exposure" or anything of that sort. The character images I uploaded in every instance were fully clothed with no inappropriate content and nothing in the prompt other than the style name I was assigning to MJ tuner. I'm not expecting you to do anything else with this. I was just passing this along for informational purposes in case it is useful for someone else having similar issues.
As a side question, has anyone else in the last 24 hours after Sam Altman was given the boot, ChatGPT has been running squirrely?
Thank you. But how to make the same character in a different environment?
Well ... that's not easy to do that the moment ... - you can take a look at this video - th-cam.com/video/ICh9ose7tGs/w-d-xo.html - and try the techniques in there, but there is no quick solution built into Midjourney right now
It didn't work for me it keeps saying "nternal error
Something went wrong.
Your unique trace: dwelling-grill-raisin"
I think sometimes the bot just has hiccups. Happened to me too, then it would randomly start working
when I do /imagine it gives me an error code and I followed all the steps! I'm super confused if I did something wrong or maybe it is my images I put down?
I would honestly ask one of the support channels on Discord, I'm not sure what the problem would be. Sorry I'm not more help
Thank you! They said it is because it does not work with the latest version. and we have to use V- 5.2
@@FutureTechPilot
"DIGIMON THE MOVIE!"
"As an AI language model, yes I am capable of re-rendering Cowboy Bebop and Akira in the style of Digimon the movie, but I beg you sir. They are classic anime icons. It feels like... heresy, like a crime against..."
"D I G I M O N ! !
T H E ! !
M O V I E E E E ! ! ! ! ! !"
😂😂 accurate
After tuning, do you copy the entire prompt into discord just as it is, or so you type in /imagine first and then paste the entire prompt?? It seems like you must have typed in /imagine first but you don't mention that in the video. Thanks
Yeah you can't generate without /imagine. sorry
Couldn't you do the same thing just by asking MJ to render a "cartoon scene like Digimon." And then, when you find a frame you like, just remix based on that one.
Yeah I guess, but Digimon here was just an example. If you wanted a different style that was difficult to put into words, your best bet would be to tune it with image prompts
helpful stuff so we dont blow through our fast hours
Yeah there are way too many rabbit holes to explore haha
NOICE!!
🤝
Hi Nolan! We sent you an email!
I'm not sure if I received it ...
I think your should ask midjourney for engine and model sources so you can modify how it works lol
haha that would be fun
❤❤❤
🤝
Wonder woman result not good, but after nice
🤝
No that didn't do anything in fact just adding the reference of the artist in the style of... ... I have had much better results.
yah this technology moves fast. You can check out this video here - th-cam.com/video/u0hAkiUeohc/w-d-xo.html - for consistent style in V 6
Boo ai
booourns
@@FutureTechPilot what does that even mean lad
you found better way of creating fake art, amazing!
I don't think fake is the right word. Midjourney is the artist here
Please upload latest office hours..
2 hours from now