[The NO Prompt Method] MULTIPLE Consistent Characters with Custom GPT & DALL-E
HTML-код
- Опубликовано: 21 дек 2023
- I challenged myself to create multiple consistent characters with this custom Story Illustrator Bot.🤖
In this video, I'll guide you through the steps of building this custom GPT with ChatGPT, from setting up character designs and art styles to configuring the bot for optimal performance.🌟
📝 Notes for this video: mia-meow.notion.site/Story-Il...
◽️◽️◽️◽️◽️◽️
🔔 Subscribe for more tips like this🔔
/ @iammiameow
💌 Join my email list: miamew-newsletter.beehiiv.com...
◽️◽️◽️◽️◽️◽️
⏳Time Travel
0:40 The Process
1:32 Character Design
3:30 Art Style
4:18 Bot Configuration
11:45 Testing
13:50 Bonus Hack
◽️◽️◽️◽️◽️◽️
👇 TOOLS I RECOMMEND (Affiliate Links) ✨
Best RUclips Analytics Tool ➡️ VidIQ vidiq.com/digimeow
The Ultimate Image Control ➡️ Stylar AI www.stylar.ai/?via=mia
Create AI voice ➡️ Elevenlabs - elevenlabs.io/?from=partnerhodge151
Join the best newsletter platform ➡️ Beehiiv www.beehiiv.com/?via=mia-meow
Create Your Own AI Talking Head ➡️ HeyGen app.heygen.com/guest/template...
Create Your KDP Book Content ➡️ BookBolt - bookbolt.io/5488.html
◽️◽️◽️◽️◽️◽️
🐱About Meow🐱
My name is Mia and I share my journey with digital products, Kindle Direct Publishing (KDP), self-publishing together with the help of AI tools. With over a decade in digital marketing, and a vision of a world where entrepreneurship is accessible to all, I am on a mission of making it easy for everyone to start their online businesses with minimal resources. If you're interested in tips and tricks in working from home, making money online with AI, or building a one-person online business, subscribe and join me on my journey!
The information on this RUclips channel and the resources available are for educational and informational purposes only.
◽️◽️◽️◽️◽️◽️
If you like my videos, you can also buy me a coffee☕️ as a small treat for me❤️
www.buymeacoffee.com/miameow 
Развлечения
Initially, I believed using the seed number or gen_ID from a generated image would help maintain character consistency, as it seemed to work previously. However, it appears there have been changes in ChatGPT, and it no longer accepts seed numbers or gen_IDs for image generation. But I still thought this GPT can be useful for some people so I still wanted to share it.
In the case of Pixar style characters, I think they look similar to each other enough that you can rely on multiple rerolls, Photoshop or faceswap tools to get a consistent look. That's why I put more focus on outfits for character identification. The use case with this GPT might be to first generate multiple characters in a shared setting, then refine the details using other tools to ensure consistency.
Hope this answers some of the questions you may have 🙂
This is absolutely fantastic, Mia! Thank you so much. Been face-palming with results but this looks like it can make all the difference.
Thank you @jabrowneauthor! Glad it helps and hope it works! :)
This is great, and I really like how you approach it from the angle of using this technology to enhance your ability to tell a story. Thank you. I look forward to more videos.
Thanks for your feedback!
This is great. I’m so glad I found your channel. Great work and subscribed! Cute characters. 🎉
Thanks for subbing! 😊
WOw! Amazing tricks, excellent video. Thankyou so much for sharing
Thank you for your kind words! Glad that it helped 🙂
I used your instructions as a guide and made changes based on my character. I was very pleased with the outcome. As you noted, the bot is not perfect but it did create amazing images of the characters as I imagined them in my head. Much more consistent. Thank you for teaching this newbie a new technique.
Glad it helps! You're very welcome 😊
Great information and thank you! Happy New Year!
Thank you and happy new year to you! 😃
Thank you so much for all your hard work and inspiration.
I subscribed and look forward to seeing more great content.
I've just started using chat gpt, dall-e 3 and other AI tools for storytelling. Your tutorial are absolutely brilliant!!
and PERFECT for growing, learning and understanding. 😊
Wow, this made my day!! Your kind words and encouragement mean a lot to me! Thank you so much for watching/subscribing and I’m glad that it helps! Happy holidays! 😆
Thank you 😊- I have been needing your knowledge base to take me to the next level
You are welcome :)
Exceptional video! Thank you for sharing!
Thank you! 😊
Thank you so much for the tutorial!😍
My pleasure! Thanks for watching! 😊
you are absolutely incredible with your instructions. You made this so much easier to understand, thankyou!
Glad it was helpful! Thank you for watching!
Amazing content. Can't believe you only have 2.5k followers! Thanks so much for this :)
Thank you for your kind words ☺️
This is amazing, thank you! I'm going to try it later!!!
That’s great. Thanks for watching! 😄
Awesome tutorial! I appreciate how much work you must put into this. Thank you,.
Glad it was helpful! :)
Very interesting - I will try.
An excellent video indeed, full of very useful stuff. Great!
Thank you!
Fantastic video. Very much appreciated. Many thanks
Glad it was helpful! Thanks for watching!
Obrigado Mia, por compartilhar isso! Vai me ajudar muito 🙏🏾
Glad it helps! 😀
This is great. Thank you for this.
You’re welcome :)
Thank you! Your videos are excellent and very helpful 👍
Thank you! ☺️
Pretty awesome, great job :)
Thank you! 😄
This is a great video. Thanks
Thank you!!
Amazing! Wow!
Thank you! 😄
Great Job!!!
Thank you!!
Thank you!
You’re welcome ☺️
Obrigado por compartilhar, muito bom seu trabalho.
Thank you! 😀
Adorei seu vídeo!!!! Você é ótima !!!!
Obrigada!
So that was very helpful, because I illustrate /design my own characters. I also like creating a prompt so the Bot learns my style and choices for all my videos.
Great, glad it helps! 😄
amazing tutorial
Thank you 😊
Finally! I have been searching for instructions like this for weeks. Thank you so much for your detailed notes and work! I owe you a cup of coffee ;)
Glad that I can help! Thank you for the coffee! 😆☕️
thank you... so useful
Glad you find it useful ☺️
Thank you
You got one more subscriber !!! ❤❤
Thanks for subscribing!!
❤@@iammiameow
Thank you.
You're welcome!
Great video thanks
Thanks for watching!
This is excellent information and Training. Thank you. I learner and easier way to create a format for building my gpt. I appreciate you.
You’re welcome! Glad it helped :)
thank you , I will try ,.
You’re welcome!
I like this. Thank you. God bless.
Thank you!
💥💥💥thanks!!!!!
You’re welcome :)
Thanks
Any better way to improve this bot?
What would you use this for?
Also, sorry for the audio variation. Will do better next time 😅
Hi Mia, I just rewatched your video and saw your comment about the seed number. Please don't remove this video, you did such a great job. Thank you for your consideration♥
I won’t remove it. Thank you for your kind words❤️
This tutorial had so much value! Thanks for all your hard work Mia, Can all my stories use the same bot, as long as I change out the characters? Thanks so very much!♥
I’m so glad you find it helpful! You can use the same bot but I would add a new bot for a new story as when you feed it too much information, sometimes it would get confused and mix up the characters. You are welcome to experiment as I haven’t really tried changing characters myself 😀
Tysm Mia!Do we own the characters we create? Is there some way to protect our intellectual property, and does the bot we create belong to us? Thanks for your time in answering my questions Mia@@iammiameow
The ownership of a GPT is a complicated subject. I would not claim to be knowledgeable in laws of intellectual property rights of either character creation nor the GPT bot.
You may consider asking ChatGPT for a general understanding. It will give you a general idea. BUT, for definitive advice and to ensure you're getting accurate information, I think it's safer to consult with a legal expert who specializes in intellectual property.
I don't want to provide you with false information here for something I am unsure. Thanks for the question though! :)
I have illustrated children's books before - the old fashioned way - and it is a lot of work, but fun and rewarding. This is amazing and also a little scary lol.
You are one of the lucky one who have the skills to illustrate children books the old fashion way! I imagine in the future all artists would train their own AI models in their own styles so they can visualize their ideas a lot quicker and AI tools would just accelerate the creative process and at the same time allows more people to have access to content creation. That being said, it's moving so fast so I wouldn't pretend that I know what would happen in the future. We will see! 😁
Yes, I agree with artists training models in their own style. It could make them insanely productive. @@iammiameow
True that!
Thank you for your video. Is it better to go this way or use the consistent character app in 4.0
You're welcome 🙂 Whether it's better depends on what you are trying to achieve. With this way, you get to have multiple characters in the same scene seamlessly but they may be less consistent. With the Consistent Character GPT at the GPT store, you get more consistent characters but need to combine multiple characters together with other tools. You can watch my other video for more details about this: ruclips.net/video/N5L7fNdZDJw/видео.html
I love your process of preparation
Thank you! 😀
Thank you, this helped and I realized that I am not the only one suffering from the "weird" results that GPT generates, I am working on my own prompt. My approach is slightly different than yours but I will share it with you cause it mught help you as well, and hopefully you share with us the results. So basically I do the role or persona, then the task, then instructions, then what to avoid, and then example or examples (if available it's better). The thing that I learned from you video is the create section. But here is a problem and I would like to know what your approach when you face it. So whenever you update the configuration of the prompt and continue the chat, the chat is not updated instead it stays configured on the previous prompt, so I am wondering what do you do, do you start a new conversation over and over again, because that's what I do😅.. or you fix the full prompt keep testing and then you start working?
Thanks for your feedback! Yeah, this is a weird problem when building the GPT. It seems to understand your instructions when you are in the chat but when you see in instruction sometimes it doesn’t update what you think it should do. What I did is I keep the instruction on a separate Google doc and when I chat with the GPT and it updates, I would look into the instruction and see if it did it correctly. If not, I would just add the new update I think is correct to my Google doc and adjust it as I see fit. Then I copy paste the final instruction I kept and revised on the Google doc to the instruction section. And asked the GPT builder to read through it (to make sure it actually get that) then use it as the final instruction to test. Yeah, it’s not straightforward and requires a lot of trials and errors. Hope that I understand your question correctly. 🙂
May God bless your soul ❤
Thank you 🙏
You can also check out an existing GPT named 'Consistent Character GPT'. It has been trained to produce very consistent characters like that. But it's also beneficial to have one yourself so you can have control of your outputs.
Thanks for letting me know! Yeah, the results are pretty consistent! The only downside is it only generates plain backgrounds and is meant for character design. I may make a video about this one soon :)
no way you are so much better than that gpt! i tried many many ways for this goal but it was like torture me, until i saw your video, i hope you give an update for it very soon! @iammiameow
Thank you. The video is now up 🙂
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
you said you want to make a film with this characters. may i know what app do you use to animate those pictures? thanks in advance
I made a video here about the tools I used to animate the images. Check it out😺 ruclips.net/video/pMKR44zYyIU/видео.html
GREAT VIDEO! how do i get the prompts to quit changing clothes, heights, and hair types in the new images?
Thanks! The more details you add to the prompt, the more likely it doesn’t give you everything you ask it to. This is still the limitation of the current DALL•E and I haven’t found a way to solve it other than just trying more times ☹️
i built the custom GPT in your video, described my characters like you did, and now its spitting out all sorts of characters. maybe i should rebuild it? Thanks for the quick reply@@iammiameow
another issue is im trying you make images for youtube shorts in 9:16 AR, but it keeps giving me 16:9 images turned sideways.
Yeah, sometimes it’s trials and errors. The more details you ask it to keep in mind, the more likely it gets the wrong thing ☹️ for aspect ratio, if you have it as instruction in your GPT as 16:9 and then ask it to generate 9:16 in the chat, it may get confused. Not sure how you set it up but I’m guessing it could be the issue. Also, I do think that DALL•E doesn’t have a good understanding of aspect ratio in general compared to midjourney 😫
it also produces real people instead of the pixar models, even tho i have 10 reference images.... any fix for that?@@iammiameow
I want to create a nail art design on theindex finger. But it changes the position of the brush and finger every time. How can i fix it?
Could you please elaborate? Are you trying to build the design at ChatGPT? AI generators now are not so good at generating hands for sure.
Just a quick question??
Why when you got to the character Marcus, you fast forward and didn’t give his description at all, you spoke more about the other 2 characters..
I simply cut that part off because the content is redundant, just wanted to give an example of how the character design is like. I include the dog because it’s an animal but Marcus and Yoko’s content is pretty similar.
Great video!
You picked a corgi! You get a comment and like =)
Thank you!! Corgi is da cutest 🐾
Wowwww i learned canva. Can you make a video on canvas super powers
Thanks for the suggestion. I will take a note :)
@@iammiameow you welcome 🥰
will it allow you to upload a photo to base it on?
At 13:15 there’re something similar to what you are looking for 🙂
This is just shooting in the dark. Optimizing chaos.
You are right. It is not perfect.
so if chat gpt doesn't accept seed numbers or gen ids, then how do you create a storybook? I've been trying for a few days now but nothing i do is working. i've watched videos, tried from as many angels as i can each time i learn something new, nothing i do is working. it's like chat gpt will *not* look at seed numbers or gen ids.
please send help as i throw my hands up in the air 🤷♀
Yeah, the technology is limited here. You can check out my other video about consistent character GPT, or you can check out midjourney’s character reference function as you may get consistent character better there. If you can use stable diffusion, you may be able to do that for free as well. But to my knowledge, nothing is 100% consistent yet so you may need to photoshop after getting something like 80-90% consistency.
i wonder if we could use multiple gpt builders for each character ?
I don’t quite understand the reason you want to do that? Could you please explain a bit more?
What program did you use to turn your images into a movie?
I animated the images using runway and pika labs. 🙂
So.. shall I pay for GPT4 or midjourney?? does anyone have feedback on this or preferences?
I think it depends on what you’re trying to do. ChatGPT has a lot more than just image generation but midjourney is probably stronger in image creation :)
Mia, we appreciate your creativity; however, the consistency of your output requires further attention. How can i fix that?
Thanks, you can try asking chatgpt for the Gen ID of the image you prefer and use that to generate the subsequent images. So the following images would look more like that image. Also, my next video may be addressing some of the consistency issues :)
Interesting video, but how do you achieve true character consistency?
You don’t. the tech simply isn’t there yet
Great question. Initially, I believed using the seed number or gen_ID from a generated image would help maintain character consistency, as it seemed to work previously. However, it appears there have been changes in ChatGPT, and it no longer accepts seed numbers or gen_IDs for image generation. But I still thought this GPT can be useful for some people so I still wanted to share it.
In the case of Pixar style characters, they look similar to each other enough that you can rely on multiple rerolls, Photoshop or faceswap tools to get a consistent look. That's why I put more focus on outfits for character identification. The use case with this GPT might be to first generate multiple characters in a shared setting, then refine the details using other tools to ensure consistency.
@@iammiameowIt is very helpful. Yes, there are corrections and regenerations but the consistency is waaaaaaaaay better. Thanks, Mia!
where you create the characters for knowledge?
Thanks for watching! After you set up the instruction, you can go back to create and chat with the GPT builder to give you some image samples or you can simply save the bot to start asking it to give you the images, and then come back to GPT backend to upload the images you get. I briefly mentioned it around 8:21. I hope this answer your question if I understand correctly. 🙂
can this process be used for bing image generator?
You can try to use the same kind of prompt structure, but you cannot really specify that much within bing image generator. With the bing image generator basically you have to write the prompt each time you are generating an image and this GPT bot is designed to help you with that part of the task.
tnx a lot@@iammiameow
Interesting method. Is this only possible with GPT-4, the paid method?
I can create images using Bing Chat, but i have a feeling this method is irrelevant there.
Yeah, the purpose is to ask the GPT to write the proper prompt for you so you don’t have to. And unfortunately this may only apply to the paid version. Using Bing Chat you can write the prompts yourself just like regularly how you would do it, but you can have a saved template that includes your character details and just change the action and environment I think. Should have mentioned that this is a paid version method in the beginning. Will do next time. Thanks for bringing this up.
@@iammiameow no no. Thank YOU.
What i tried to do is define characters to Bing chat and describe them, and then define a style. Then, whenever i said their names, chat GPT remembered their details.
It didnt come out as well as what you have shown, but it was kinda close.
Still, i learned something new from your video, so thanks a lot :)
@@adarwinterdror7245 I haven't tried Bing Chat so I am not sure, but glad that you have learned something new at least :)
I have a problem the format of image not be consistent like background and lightening and this style
It might be because there's too much information for DALL-E to process. You may want to make your base prompt more concise. Or you might need to try a few more times. It's not perfect, but hopefully we can get to the results after a few tries.
The bot instructions in the video are not in the notes. There is only a very small snippet of the file you used in the video. Are you able to provide the whole file?
The instruction should be in the notes. For some of them, you have to toggle it open (click the triangle on the left to expand it) and you should be able to see it :)
is Dalle 3 only available thru chatgpt
You can use bing.com/create or bing chat but not in the same way tho…
hey there is nothing in the description did you removed it ?
It's in the first link under the description - notes for this video.
no there wasnt anything that i could copy and use can you recheck the link kindly@@iammiameow
and soo much thnx for replying
@@iammiameow
Oh, yes, it’s a toggled list. Try clicking the triangles to expand the points and see if it works.
Can't I create an image using chatgpt3.5?
Sorry, forgot to mention, only the paid version would give you images for now.
Do you need a premium account for this? Or is it now free with GPT-4o?
You do need a paid account for this but you can create images with Dalle for free on bing.com/create
Do you need a subscription to GPT Pro to access the GPT Builder? When I asked GPT it said it wasn't familiar with anything called a GPT Builder!
Yes, you do need a paid subscription for this. The GPT builder is just the chat interface when you are constructing the GPT.
Does this work for iPad?
I haven’t tried but it should work.
is this only for the paid version
Yeah, sorry, it’s only in the paid version.
@@iammiameow no problem just wanted to make sure
Really nice
Do you feel frustrated sometimes? 😂
Do you mean when DALL•E doesn’t follow your prompts? I’ve learned to accept what is and let go of what was and take a deep breath, 😂
@@iammiameow I remember exactly a year ago, when we only had GPT-3.5, which provided just hallucinations, and Midjourney was super slow, requiring a lot of 'prompts' to get 'nice images.' I felt so frustrated that I just wanted to punch my laptop.😂
I feel your pain... it is also scary how fast the technology is moving now. Soon they will prompt themselves and we will not be needed 😅
Sadly something has happened to Dalle 3 and it seems impossible to have consistency on characters, rendering unusable for projects that go further than a single image request. Hopefully they will bring back SEEDS and make Dalle3 available again: now chatgpt plus subscription hasn't Dalle3 available anymore but just DallE 2
True that. It used to work but no more. OpenAI seems to change its mind every other week. I didn't know about that ChatGPT plus subscription only has DALL-E 2. That is surprising!
At now it seems that the only way to get consistency is with midjourney creating characters sheets, or with leonardo training a model (still using character sheets). Even if DallE inside ChatGPT with the illustration model you show us has the best capabilities for layout more than one characters and make them doing specific actions and understanding the logic behind the shot.
True, I agree. Hopefully there will be new updates soon to make it easier for people.
Oh she’s very cute
Thank you.
Gpt 4.?
Yes, it has to be the paid version.
are u paid the GPT 4 or Free ?
Sorry that I forgot to mention in the video. This method is for paid version ChatGPT.
no problem @@iammiameow
It seems like some of the generation failures might come from the rules they put on the dalle. Like the article 8 speaking of diversity might interfere with the race of the character etc. Too much wokeness feels annoying sometimes lmao. If not for all the rules it might be pretty straightforward to create a bot for children to first create characters and then stories with generations based on the previously created characters, would need some backend code and api but many children would love to play with that.
Lol, there’s indeed too much wokeness in it. Yeah, the censorship could be one of the reason that the bot doesn’t follow simple instruction, I haven’t thought of that. And I also thought of a similar idea where children and parents can co-create their customized stories - i think it’s an fantastic idea! It can probably live off of ChatGPT system on an app or something. Thanks for your insights!
GPT 4 its Free or you paid for that ?
Sorry, forgot to mention. It’s paid.
I thought it was free, btw thanks for your video, keep it up @@iammiameow
Thank you!! 😄
the content policy is a pain in the asss for me and it makes dall e tool completely useless. I want to create romance vibe with 20yo characters but i can't describe anything that is not content policy OK. For example i want a skinny unconfident guy it's not ok. I want a gorgeous woman it's not ok. I want an antagonist muscular masculine look guy it's not okay like wtf... it make it completely useless sadly
Wow, I didn’t know it was so strict. These seem like reasonable requests. Then probably midjourney or stable diffusion would be more useful for you. OpenAI censorship is really their weakness for sure.
I feel you bro...i can't even create an angry face expression
I feel you bro...i can't even create an angry face expression
I feel you bro...i can't even create an angry face expression
Yeh Ikr😢
This process is done with chat GPT 4, it is not free
Yeah, it is not free. Thanks for mentioning it.
how much does it cost to subscribed to ChatGpt4
to promote a relationship between 2 young characters : one is black one is asian is wrong, especially the black character is a male....
Thank you for your comment. I want to clarify that the content only depicts a friendship between the boy and girl, with no romantic undertones. Their interactions are intended to be purely platonic, and the idea of them being friends is central to the story.
Are you a professional illustrator? Can you really draw well with a pencil?
I am more of a marketer with digital design experience for corporates. I learned drawing since I was a kid but I guess I don't really have the talent to draw well professionally.