Thank you, John. Character consistency is one feature I have been trying to achieve in my stories and your video will really help. I am retired and create personalized flip books for my grandkids. Thanks again and good luck in your endeavors.
I hope the Midjourney part of this video soon becomes obsolete when MJ introduces consistent characters. Very impressed by the GTP output. Thanks for sharing!
Sooooo amazing and THANK YOU! This solution addresses the reason I’ve been reluctant to work in this Ai space! How to keep a character image you create and love going in a series of works. Again “Thank you”! God bless
Brilliant and thanks. Yeah I need to try out the different art styles on the GPT now. But it's a solid approach and hopefully we'll see similar things rolling out elsewhere.
Thanks for tutorial! I tried your way, on Dall-E part, amazing, everything is fine and smooth. But in MidJourney part, it really seems less controllable. First, it stuck on "Thinking" after I set a character with around 10-12 images. Second, when I used this character to create an scene, to be honest, it's really in lack of consistency and hard to be recognized as same person. Third, it's also difficult to lock the art style in MidJourney when making a series of image all prompted with "Pixar Styled 3d animation", it resulted with various understanding over the style. Still not a stable way to support a short video. I do feel for AIGC, next focus should be CONTROL, to support creating meaningful longer experience.
Yeah, I included that part in the video to show the approach but it definitely needs refinement and is no where near as consistent as the GPT part of the video. You can play around with image weighting though, plus add extra scene references and go round in the endless loop of generations and refinements to try and get the scene your after. The '/pre option set' approach just adds a bit of time saving really to 'try' and get better consistency in MidJourney. The GPT using Dall-E is certaining impressive though.
yous should use scenario ai to train a model based on the images you created and then it will give you consistent characters, we'll see if it's better or not thsn midjourney's image training
Really useful. Out of interest - why do you move over to Midjourney if the characters are less consistent than with Dall-E in ChatGPT? I’m finding Dall-E tends to create better 3D or illustrated characters than Midjourney anyway but maybe I’m missing something?
Yes, the Midjourney part seemed a bit of a dud in this workflow. But super impressed with the Dalle/Chatgpt combo. That is insane. Need to see many other styles supported. I've seen way too much Pixar stuff already.
@@j0shj0shj0sh yeah, I don't love when there's essentially a pulldown menu of options for 'looks'. better when the range is wide open. otherwise, Dall-e consistency was impressive
Yeah, I wasn't sure about including the Midjourney part. But thought teaching people about the 'prefer option set' would be useful for some. I'd recorded a bit saying how the results aren't great here yet, but you can play with image weights and prompt weights. To get better results. But it didn't make the edit and I had to rush the export and go do some non AI life stuff 😂🤷🏼♂️... So didn't have time to try and show some more midjounrey variations. (+It'll be cool once they add the zoom out and in-painting (vary region) functions to the v6 Alpha... (but there are ways around that via Photoshop or Midjourney V5). Glad the GPT part is popular though 👍
Hi, thk for this tutorial. I have a question please. I already have a character picture and i would like to change his position with AI method to give him many differents position . what do you advise? Thanks a lot
@@LazyJudewhat’s the best way to make a consistent character then? I thought I’d cracked it but I’m not having issues, I sse ppl have don’t it though, I need directing
You could and maybe should. But there's a lot of things MidJourney can do really very well, (albeit not consistent characters very well.. yet). But 100% you could stay in Dalle-3.
I hope soon it will be possible to pose multiple consistent characters together, add a consistent background, and have more control over the scene we create.
does it work with wonly anime images or images that are more life like (such as photos and the more human like images dalle3/bing can create)? It seems bing recognizes certain images (photos or images close to photos) as real and therefore struggles to use those images to make any changes or consistency with those types. Would love to hear from you on this
I am having a difficult time getting it to generate a full image of my character. It will only give me headshots/close-ups. I am trying to create a baby eagle and I need a whole body image of it flying.
Yeah, that will be cool. You could take the image of a t-pose and make it a 3D model (with ok-ish quality) via something like CSM using AI. Then rig it via Mixamo pretty quickly... (if you're lucky) and animate it using Mootion's AI generated file. (Or do any all those steps manually for better quality and control).
@@blairzettl3933 . Hey, that's interesting I might have to revisit it. Out of interest did you go with the Pixar style character or one of the other options. I trie the photoreal option and it was less consistent, whilst still hitting all the requested key attributes there was definitely more variety in the output.
Yes, just keeps going. Runs with GPT 3.5. Ran out of time but keen to explore the other art styles and see how much the background can/can't be changed as well. But being able to then take the characters into Photoshop and future work is cooll.
Not using this tool. If you had enough sketches of a character, you could explore training up a LoRa, and using that with a model i.e. Flux/Stable Diffusion to genenerate new images. A much more involved process though, but getting simpler.
Due to the popularity of custom GPTs you need to be more specific in your search query. Using “consistent character gpt” should get you different results and the GPT used in this video is now over 200k downloads There wasn’t a prefilled starter question anymore, and it didn’t ask scenery details. The GPT description says to just start by clicking “CLICK HERE” but it’s not showing up for me. I asked it, “click where?” and it started walking me through the questions. Was asked … male/female - characters name - then details and then it generated. I was expecting more questions On my second pose I asked it for eyes closed and the result had shorter hair and the facial complexion was way different.
Can you tell me if you know of an app or program that allows you to make it appear that your dog is talking in a video? Not from a photograph but actually from a video? Any help you can provide will help. Thank you. Phillip
I've not seen any 'easy' software that would achieve this yet. Unless your dog character is particulalry stylised (near human character) then it might work with something like Lalamu lalamu.studio/demo/ Studio or SyncLabs. synclabs.so . Otherwise I think it will involve a more traditional digitial animation style workflow.
i am wondering, how git this that consistent. I tried it for several times, even with your prompts and i can't achieve consistancy at all, could you repeat that strategy?
There's a few somethings coming out. Take a look at CSM or Meshigy for some 'basic' image to 3D models, Mixamo can help with quick character rigging of a character rig. Or 'Mootion' with two o's, can animate a character (currently from an exciting selection of character) and lets you download the 3D fbx character rig.
Thanks for your video! Consistently find that making characters with the /POS works for training the MJ, but the --ar 16:9 will, almost without fail, crop the image right above the forehead. Changing the --ar to a more portrait ratio can help, but then you're stuck trying to accommodate multi-aspect ratio images into a timeline. Even adding descriptions of "objects" above the "headline" is hit or miss. Does anyone run into this?
Yeah, have much the same output results. Potentially using more zoomed out images (adding extra blank space around the images) uploaded to midjourney 'might help'. Otherwise if you have Photoshop the expand canvas function works really very well and does a decent job and maintaining the art style in most instances. Based on the Midjourney office hours, Give it 3-5 weeks and hopefully we should see zoom out/pan and vary region abilities rolled out to V6. (Which work in V5).
Thanks, But I still got confuse how can I make this character generation keep the same look in next time while I open ChatGPT again? Could you answer me? or anyone have related idea for my question? Thanks a lot again!
great vid. I think it would have helped to add --iw 2 at the end of your prefer set prompt so that it would try to follow the images more closely, even though we all know it's still not great. 😢
Yeah that or '::4' at the end of the image prompts helps a bit. Looking forward to the future character consistency hinted at in the latest Midjourney office hours.
Tried using the Consistent Character feature with FREE ChatGPT 4.0. But not able to generate. It asked for few prompts (initial 2/3 only matched with your sequence) and then gave an error. Do we need PAID version of ChatGPT
Looks like you have to have the paid GPT4 for this? Ive tried Bing Image Creator and its good, but i cannot make characters to use over and over, it changes their faces after each prompt. Trying to also do this without PC and on android only. Any ideas how i can make a keep using my own characters thst way, preferably without a subscription?
@AIAnimationStudio Assuming I don't know the first thing about 3d rendering but I can manage to download a 3d obj or use an app to make one with a physical object scanner, I'd like to be able to feed it to a "3d obj to img" ai tool, and give a simple prompt of the scene I would like to see it in. I also thought something like that might help for consistent characters down the road, but it looks like pretty good solutions are developing. Could still be useful, idk?
@@BlitzDjingga . Oh 100% agree. If it wasn't for this channel and wanting to try out all the tools, to see what can be done, I'd probably just focus on 1 or 2. On the plus side, most if not all offer limited free trial, then you can pay a monthly subscription and cancel. You could then pay for Runway, Pika, Leonardo... and try and figure out the best for each. Or use Stable diffusion and Comfy UI for things and run it locally... which whilst free, comes with a different learning curve and investment in some semi-decent computer system. (I do like ComfyUI a lot though).
It runs on Chat GPT 3.5 for me ...and uses Dalle-3... so.... I think that means it's currently free.? ..... or is that paid. I pay the lowest tier as use Chat GPT for work quite a bit so not 100% sure.
Not impressed at all with the chat gpt consistent character function. It would not do anything but give me saccharin happy images for the pastel animation style and could not reference my uploaded image at all to get a unique look. At this point it should have much more nuance than that. Midjourney is still leaps and bounds better.
Feels like stealing though to me. What data are they using to train the software? AI has real benefits in certain fields, but this type of use case feels like robbing from peoples hard work and talent.
Absolutely useless app. Irritating, frustrating, don't follow instructions. It's a waste of time. Don't suggest useless apps to get viewership. You will lose credibility
I appreciated the Midjourney part because I didn’t know about the prefer option piece. Thanks!
Thank you, John.
Character consistency is one feature I have been trying to achieve in my stories and your video will really help. I am retired and create personalized flip books for my grandkids.
Thanks again and good luck in your endeavors.
Thansk Bill, glad you found it useful. Good luck with your projects for the grandkids. 👍
I hope the Midjourney part of this video soon becomes obsolete when MJ introduces consistent characters. Very impressed by the GTP output. Thanks for sharing!
i didnt know any of this. Everything is moving so fast. lol thanks
This is great! Thank you so much for showing us this character GPT.
Sooooo amazing and THANK YOU! This solution addresses the reason I’ve been reluctant to work in this Ai space! How to keep a character image you create and love going in a series of works. Again “Thank you”! God bless
Brilliant and thanks. Yeah I need to try out the different art styles on the GPT now. But it's a solid approach and hopefully we'll see similar things rolling out elsewhere.
ruclips.net/user/shortsIYl3bL89QFQ?si=E8qfooX6PWQz4YhA
damn, this is awesome, Thank you so much!!
Well done. Very clear.
Thanks for tutorial! I tried your way, on Dall-E part, amazing, everything is fine and smooth. But in MidJourney part, it really seems less controllable. First, it stuck on "Thinking" after I set a character with around 10-12 images. Second, when I used this character to create an scene, to be honest, it's really in lack of consistency and hard to be recognized as same person. Third, it's also difficult to lock the art style in MidJourney when making a series of image all prompted with "Pixar Styled 3d animation", it resulted with various understanding over the style. Still not a stable way to support a short video. I do feel for AIGC, next focus should be CONTROL, to support creating meaningful longer experience.
Yeah, I included that part in the video to show the approach but it definitely needs refinement and is no where near as consistent as the GPT part of the video. You can play around with image weighting though, plus add extra scene references and go round in the endless loop of generations and refinements to try and get the scene your after. The '/pre option set' approach just adds a bit of time saving really to 'try' and get better consistency in MidJourney. The GPT using Dall-E is certaining impressive though.
Wow great find, thank you for sharing!
This was very helpful, thanks!
Thanks for this, does it require subscription
This is so good, thank you!
yous should use scenario ai to train a model based on the images you created and then it will give you consistent characters, we'll see if it's better or not thsn midjourney's image training
Fantastic results! Congratulations!
ruclips.net/user/shortsIYl3bL89QFQ?si=E8qfooX6PWQz4YhA
great workflow - thank you sharing:)
Thanks ;)
Really useful. Out of interest - why do you move over to Midjourney if the characters are less consistent than with Dall-E in ChatGPT? I’m finding Dall-E tends to create better 3D or illustrated characters than Midjourney anyway but maybe I’m missing something?
Yes, the Midjourney part seemed a bit of a dud in this workflow. But super impressed with the Dalle/Chatgpt combo. That is insane. Need to see many other styles supported. I've seen way too much Pixar stuff already.
@@j0shj0shj0sh yeah, I don't love when there's essentially a pulldown menu of options for 'looks'. better when the range is wide open. otherwise, Dall-e consistency was impressive
Yeah, I wasn't sure about including the Midjourney part. But thought teaching people about the 'prefer option set' would be useful for some.
I'd recorded a bit saying how the results aren't great here yet, but you can play with image weights and prompt weights. To get better results.
But it didn't make the edit and I had to rush the export and go do some non AI life stuff 😂🤷🏼♂️... So didn't have time to try and show some more midjounrey variations. (+It'll be cool once they add the zoom out and in-painting (vary region) functions to the v6 Alpha... (but there are ways around that via Photoshop or Midjourney V5).
Glad the GPT part is popular though 👍
@@AIAnimationStudio any kind of info on consistency within AI image making is always welcome! that's one of the big challenges now
The Reel Robot has a video about consistency that is worth checking out. His workflow was Chatgpt/Dalle + Midjourney + EverArt.@@AIAnimationStudio
Thank you for the short but very informational tutorial! I also love your audio effects when you were speeding up in the Midjourney.
Thanks... yeah I'm a sucker for the spend up typing noises too, like to leave a few in the videos here and there.
Great tool
Next....LEVEL!!!!!
What i found is, in the toy world and animation world, Bob will never let you down.
Oh yeah, no doubt Bob is a dependable chap 😂
Thank you for your vedio.
Hi, thk for this tutorial. I have a question please. I already have a character picture and i would like to change his position with AI method to give him many differents position . what do you advise? Thanks a lot
it doesn't generate consistent image. Don't waste your time.
The GPT does an impressive job though? (But.. I've not had chance to try out the other art styles, i.e. photography is that less consistent? 🤔)
i tried it out for 2 days. different characters every time. don’t waste your time.
Thank you for the time saved
@@LazyJudewhat’s the best way to make a consistent character then? I thought I’d cracked it but I’m not having issues, I sse ppl have don’t it though, I need directing
One Love ✊🏿
But why not keep using Dall-E for image generations
You could and maybe should. But there's a lot of things MidJourney can do really very well, (albeit not consistent characters very well.. yet). But 100% you could stay in Dalle-3.
I hope soon it will be possible to pose multiple consistent characters together, add a consistent background, and have more control over the scene we create.
ruclips.net/user/shortsIYl3bL89QFQ?si=E8qfooX6PWQz4YhA
yes, its possible ruclips.net/video/C2DPGWVii_Y/видео.html
Amazing new follower!
Thanks, and I really appreciate the sub. 👍.. I need to start planning the next deep dive into another AI tool.
does it work with wonly anime images or images that are more life like (such as photos and the more human like images dalle3/bing can create)? It seems bing recognizes certain images (photos or images close to photos) as real and therefore struggles to use those images to make any changes or consistency with those types. Would love to hear from you on this
I have given up on trying to do consistent characters mine are never consistent and I had a great idea for a kids book
Thanks for the video! For me this Consistent character GPT failed to show character from the back. Did anyone faced this issue?
I am having a difficult time getting it to generate a full image of my character. It will only give me headshots/close-ups. I am trying to create a baby eagle and I need a whole body image of it flying.
Thank so much,👍
can you please provide all promots for chat gpt like if we want same charter in one story for this what is the correct prompt for this ?
Need upgrade to gpt PLUS or TEAM?
amazing thank you
Crazy. Now if would only export as FBX rigged, skinned, and with textures!
Yeah, that will be cool. You could take the image of a t-pose and make it a 3D model (with ok-ish quality) via something like CSM using AI. Then rig it via Mixamo pretty quickly... (if you're lucky) and animate it using Mootion's AI generated file. (Or do any all those steps manually for better quality and control).
@@AIAnimationStudioI tried this out and I did not get the consistency that you achieved, however I was blown away by the quality.
@@blairzettl3933 . Hey, that's interesting I might have to revisit it. Out of interest did you go with the Pixar style character or one of the other options. I trie the photoreal option and it was less consistent, whilst still hitting all the requested key attributes there was definitely more variety in the output.
so can you do more images with the same caracter in that chat?
Yes, just keeps going. Runs with GPT 3.5. Ran out of time but keen to explore the other art styles and see how much the background can/can't be changed as well. But being able to then take the characters into Photoshop and future work is cooll.
What if you already designed a sketch of the character could you upload it and create consistent artwork from that
Not using this tool. If you had enough sketches of a character, you could explore training up a LoRa, and using that with a model i.e. Flux/Stable Diffusion to genenerate new images. A much more involved process though, but getting simpler.
This is bang on.
why can't i find that particular GPT? i can find others tho but not the one mentioned in the video by Sachin Kamath
Not sure. It comes up as the first result for me when searching 'Character'.
Due to the popularity of custom GPTs you need to be more specific in your search query. Using “consistent character gpt” should get you different results and the GPT used in this video is now over 200k downloads
There wasn’t a prefilled starter question anymore, and it didn’t ask scenery details. The GPT description says to just start by clicking “CLICK HERE” but it’s not showing up for me. I asked it, “click where?” and it started walking me through the questions.
Was asked … male/female - characters name - then details and then it generated. I was expecting more questions
On my second pose I asked it for eyes closed and the result had shorter hair and the facial complexion was way different.
Can you tell me if you know of an app or program that allows you to make it appear that your dog is talking in a video? Not from a photograph but actually from a video? Any help you can provide will help. Thank you. Phillip
I've not seen any 'easy' software that would achieve this yet. Unless your dog character is particulalry stylised (near human character) then it might work with something like Lalamu lalamu.studio/demo/ Studio or SyncLabs. synclabs.so . Otherwise I think it will involve a more traditional digitial animation style workflow.
Awesome!!
Is there a way to upload a vector animal character face and get it to lip sync? I tried Heygen but it doesn’t detect animal faces only human vectors
thankyou so much 😍
i am wondering, how git this that consistent. I tried it for several times, even with your prompts and i can't achieve consistancy at all, could you repeat that strategy?
Great, Ty!
But this is just a photo of the character.
How do I make a 3D model completely with a bone skeleton?
There's a few somethings coming out. Take a look at CSM or Meshigy for some 'basic' image to 3D models, Mixamo can help with quick character rigging of a character rig. Or 'Mootion' with two o's, can animate a character (currently from an exciting selection of character) and lets you download the 3D fbx character rig.
유용한 정보 잘보고 가요~~~
Could we animate this characters ?
Yeah, via some other tools and techniques. Got some things in the works for that for a future video.
Thanks for your video!
Consistently find that making characters with the /POS works for training the MJ, but the --ar 16:9 will, almost without fail, crop the image right above the forehead.
Changing the --ar to a more portrait ratio can help, but then you're stuck trying to accommodate multi-aspect ratio images into a timeline. Even adding descriptions of "objects" above the "headline" is hit or miss.
Does anyone run into this?
Yeah, have much the same output results. Potentially using more zoomed out images (adding extra blank space around the images) uploaded to midjourney 'might help'. Otherwise if you have Photoshop the expand canvas function works really very well and does a decent job and maintaining the art style in most instances. Based on the Midjourney office hours, Give it 3-5 weeks and hopefully we should see zoom out/pan and vary region abilities rolled out to V6. (Which work in V5).
Great information, thank you, new follower
Thanks, very much appreciated.
thank you. is the Consistent GPT accessible via a free ChatGPT account?
Nope it’s not. Just tried this
@@AfricantalesbyJudy Thanks
Great content, subscribed :)
Cheers 🏆
Thanks, But I still got confuse how can I make this character generation keep the same look in next time while I open ChatGPT again? Could you answer me? or anyone have related idea for my question? Thanks a lot again!
Just use the same chat...
Oh right! Just need to keep the conversation stick on the left and mark it. Thanks again@@tonon_AI
great vid. I think it would have helped to add --iw 2 at the end of your prefer set prompt so that it would try to follow the images more closely, even though we all know it's still not great. 😢
Yeah that or '::4' at the end of the image prompts helps a bit. Looking forward to the future character consistency hinted at in the latest Midjourney office hours.
awesome!
yesssssssssssssss!
Tried using the Consistent Character feature with FREE ChatGPT 4.0. But not able to generate. It asked for few prompts (initial 2/3 only matched with your sequence) and then gave an error.
Do we need PAID version of ChatGPT
Is it available for gpt 3 also?
No :/ I'm searching how use it for free
Looks like you have to have the paid GPT4 for this? Ive tried Bing Image Creator and its good, but i cannot make characters to use over and over, it changes their faces after each prompt. Trying to also do this without PC and on android only. Any ideas how i can make a keep using my own characters thst way, preferably without a subscription?
Hmm. Not sure.
It worked for GPT 3.5 but that was after I'd started using the character got via GPT4.
Are there any 3d obj to img GPTs?
Not sure. Like a simple renderer of a 3D file you have?
@AIAnimationStudio Assuming I don't know the first thing about 3d rendering but I can manage to download a 3d obj or use an app to make one with a physical object scanner, I'd like to be able to feed it to a "3d obj to img" ai tool, and give a simple prompt of the scene I would like to see it in. I also thought something like that might help for consistent characters down the road, but it looks like pretty good solutions are developing. Could still be useful, idk?
This is great, thanks sharing
Tried it. It did not output anything like what you have here.
❤❤❤
So you subscriptions for both chatggpt and midjourney ?
I do both for things I do for this channel, plus client work.
I'm not sure but dalle-3 and chat gpt 3.5 might still be free.
@@AIAnimationStudio not sure about dall-e. It just feel adding up and worse is like paying double.. chatgpt, co-pilot, dall-e.. and then the othersss…
@@BlitzDjingga . Oh 100% agree. If it wasn't for this channel and wanting to try out all the tools, to see what can be done, I'd probably just focus on 1 or 2.
On the plus side, most if not all offer limited free trial, then you can pay a monthly subscription and cancel.
You could then pay for Runway, Pika, Leonardo... and try and figure out the best for each. Or use Stable diffusion and Comfy UI for things and run it locally... which whilst free, comes with a different learning curve and investment in some semi-decent computer system. (I do like ComfyUI a lot though).
@@AIAnimationStudio thanks mate!
Thanks! Useful information.👍👍👍
Si chatgpt crea los personajes con consistencia, que necesidad hay de llevarlos a Midjourney?
Cualidad
Interestingly, the Consistent character only worked once. Now it only takes the first image and then it always gives an error.
Can I use this method for free?
am I a creative? How to make free account for my case? tnx
I think (may be wrong) that you can use a free account for Chat GPT? (Mid Journey is paid)
is it free to use consistancy character?
It runs on Chat GPT 3.5 for me ...and uses Dalle-3... so.... I think that means it's currently free.? ..... or is that paid. I pay the lowest tier as use Chat GPT for work quite a bit so not 100% sure.
@@AIAnimationStudio ok thanks bro
This GPT is not creating character consistently and I tried very much. Instead one can use DALL-E itself directly.
Mine is not consistent at all.
Character GPT was impressive but the MidJourney result was a little disappointing.
🍎🍎🍎🍎🍎
Not impressed at all with the chat gpt consistent character function. It would not do anything but give me saccharin happy images for the pastel animation style and could not reference my uploaded image at all to get a unique look. At this point it should have much more nuance than that. Midjourney is still leaps and bounds better.
A Nigerian woman just copied your video exactly but changed with shoes to Nike's , and made it a black kid. 😂😂😂😂😂😂
Wow!.. just saw the video you were referring too.... just WOW!.... and they've got more views than me..
uh seed locking
Would be nice if it was free.
I never found any helpful tips from your videos all i found making a simple thing complex in the videos.
Arr well... Must try harder I guess. Did the consistent character approach with a really easy to use new GPT app not seem useful?
Feels like stealing though to me. What data are they using to train the software? AI has real benefits in certain fields, but this type of use case feels like robbing from peoples hard work and talent.
Chat GPT does so much better job. This is paid version?. I did 2 picture books with MJ. Not so good :)
Absolutely useless app. Irritating, frustrating, don't follow instructions. It's a waste of time. Don't suggest useless apps to get viewership. You will lose credibility
It used to work. I can only assume something has changed since they first published the GPT... but good to know, I won't bother revisiting it.
Midjourney is HORRIBLE