This really made me reconsider using Midjourney for depicting scenes from my books and look for alternatives. This one seems quite viable. Thanks for sharing.
Just watching this, you did excellent work. In the AI world, still one of the most difficult and still almost impossible things is to get multiple consistent characters in one scene. Two characters is one thing, but I'm working on 4-5 consistent characters in 1 scene to enable AI family portrait features. But as said; this tutorial motivates me to take another look at Leonardo
Thanks! The more characters the more difficult it will be. 4-5 might require some photoshop to get the all in the same image. Ai is not quite there yet imo
Hi Tao, this is a great and very helpful video, thanks! Is there a way to configure the prompting such that Leonardo AI is able to keep the clothing/attire the same as the character reference too? For fantasy characters, their attire is representative and does not really change across scenes/activities. Not sure if you have experimented in this regard, and have any tips to share please. Because I keep getting variations in the attire. Thanks!
It is hard to get the clothing to stay exactly the same. You could try being more specific in your prompts about exactly what the clothing should look like. "a light blue denim jacket with a white t shirt underneath" But sometimes it just takes a bunch of retries to get it to look the way you want.
Very grateful for your tutorial! I found that the characters’ fingers or toes were sometimes distorted even when I turned the motion strength to 1 for animation. Grateful if you can come up with a tutorial to help me fix this problem.
You do, but it's easy to upscale again. The free Upscayl app does a great job of this.... remember to change dpi from 72dpi to 300dpi before you use Upscayl though as it doesn't change the dpi.
There is some documentation of how to use their api for Image Guidance (and character reference) here: docs.leonardo.ai/docs/generate-images-using-image-to-image-guidance
I'm not too sure about your workflow, though. There are actually a few crucial steps that we wouldn't want to miss out on if using this method of separating and then combining two different images into one. I would add style references as well so the result would follow the original look. Leonardo is complex, but if used properly with the right workflow, it could produce great results
This maybe a dumb question but I am new to this. Could you have changed the background image( the trees the park) to grayscale. thereby the man and the women be full color? They would be the only color in the image? Thanks
In Leonardo we can also generate images with two characters in a single frame? Are these options only available in paid version, means on upgraded version. I am trying many such generation websites and checking for such. I could only find that in your video. Also, taking a reference photo and adding the face we need in maintaining the consistency. This really matters for me. Thank you for the video.
I think character reference is only for paid plans. This feature works best if you use images generated with Leonardo, it won't work as well with your own photos.
For user friendly interfaces, I've only tried insight face and photoshop Ai. If you're more familiar with programming & machine learning you could try InstantID or similar tech: github.com/InstantID/InstantID
Hello, Tao . Great content as always! I learned a lot from your videos. I currently have a Midjourney subscription and I’m pretty happy with the results I get there , but I struggle with getting the exact same poses as in the reference images I use . In Leonardo this seems to be pretty easy job with the Pose reference. Could you please give me an advise ?
Hey, that's one of the downsides of Midjourney. You don't really get fine-grained pose control the same way that you can do it in Leonardo Ai. Maybe they will add a feature for it in the future. In general, pose reference is done with ControlNet. Ai's with that can map poses for you.
Pika Labs has a function to do video inpainting, so you can swap out clothing I cover that in this video: th-cam.com/video/F4vZTKOCeds/w-d-xo.html at the 17:59 timestamp. It does only work on 3 second videos though.
Can you help me maybe with a prompt I have? I want to have a hiphop style white t-shirt and white dungarees and have the dungarees with the straps not tight but loose. How can I accomplish this?
I haven't tested that out, but if you check Leonardo's official announcement, they show some examples of cartoon & illustration characters: leonardo.ai/news/character-consistency-with-leonardo-character-reference-6-examples/
Main issue with leonardo is, its really just Stable Diffusion and SD really has a hard time making characters interact and NOT face the camera. Thats because Nearly all SD Checkpoints have been heavily trained on Selfies or similar posing shots. Midjourney and Ideogram are much better at prompt adherence and cinematic shots but Ideo doesnt have cref
It does lack the photorealism that Midjourney has, but you do get some more flexibility in the poses. You could try adding in a style reference, or elements to see if that helps.
![[The NO Prompt Method] MULTIPLE Consistent Characters with Custom GPT & DALL-E](/img/n.gif)
Hooray! You did it, Tao. Thank you. I know this took time and work. Thank you.
Thanks Suzanne, I'm excited to see what else Leonardo comes out with. They've been releasing so many good features recently.
This really made me reconsider using Midjourney for depicting scenes from my books and look for alternatives. This one seems quite viable. Thanks for sharing.
Midjourney creates the best looking images, Leonardo gives you more control over the final result 👍
Hello Tao, you are great. Thank you very much for your gentle, clear and details teaching. Keep it up!!!
Thanks! I'll make more guides for Leonardo if they have any big updates 👍
Very helpful, Tao. I learned a lot from this single tutorial. Thanks for sharing.
That's great 👍. Leonardo is a really good tool for animating characters, especially with the pose control
This is exactly the video I needed! Fantastically helpful. Thanks very much! Subscribed.
Thanks for the sub! And great to hear this was helpful 👍
Esto era justo lo que estaba buscando. Quería validar si Leonardo me servía para lo que necesitaba y gracias a tus tutoriales ahora sé que sí.
Es genial saber que esto ayudó. Leonardo es una herramienta única y tiene muchas funciones útiles.
Very practical guide. No BS at all. Thank you!
Thanks! Glad you found this helpful, Leonardo has improved a lot recently
THANKS. This video is very useful. When will Leonardo simplify the procedure to get two persistent characters in the same image?
It was great, thank you for your practical training, my friend
Glad this was helpful man 👍
Excellent tutorial, thank you for sharing your knowledge!
Glad to know this helps, Leonardo is a pretty good alternative to midjourney 👍
Just watching this, you did excellent work. In the AI world, still one of the most difficult and still almost impossible things is to get multiple consistent characters in one scene. Two characters is one thing, but I'm working on 4-5 consistent characters in 1 scene to enable AI family portrait features. But as said; this tutorial motivates me to take another look at Leonardo
Thanks!
The more characters the more difficult it will be. 4-5 might require some photoshop to get the all in the same image. Ai is not quite there yet imo
Amazing, thanks Tao.
You deserve subscription man,this is what i need.❤
Thanks man, Leonardo can do some cool stuff and they're always releasing new features
everything I needed is in this video, thanks a lot bro!
For sure man, glad to hear this video helped!
Good, your explanation is so great. I want to learn more from you.
Thanks, I appreciate that, I've got a lot more guides planned soon!
Fascinating, fantastic break down explanation vid Tao bro, I’ve gotta play/use Leonardo now and not MidJourney 100
Thanks man, It seems like Leonardo is more willing to try out experimental features even if they are not perfect.
Great tutorial. Nice a clear. Thanks and keep it up!
Can I mix grey scale and color images in dataset training for creating character models in Leonardo? Thank you.
Man you have 1 video for every need I have, I'm loving it.. great job !
Thanks man, I try to put out as much useful content as possible
This is cool 😎 thx for sharing
Hi Tao, this is a great and very helpful video, thanks! Is there a way to configure the prompting such that Leonardo AI is able to keep the clothing/attire the same as the character reference too? For fantasy characters, their attire is representative and does not really change across scenes/activities. Not sure if you have experimented in this regard, and have any tips to share please. Because I keep getting variations in the attire. Thanks!
It is hard to get the clothing to stay exactly the same. You could try being more specific in your prompts about exactly what the clothing should look like.
"a light blue denim jacket with a white t shirt underneath"
But sometimes it just takes a bunch of retries to get it to look the way you want.
Very grateful for your tutorial! I found that the characters’ fingers or toes were sometimes distorted even when I turned the motion strength to 1 for animation. Grateful if you can come up with a tutorial to help me fix this problem.
Thanks for the suggestion, I will checkout Leonardo's motion generator soon. I haven't tried it yet though.
amazing thank you so much, how can i add weapons or get weapons on my characters
I would specifically ask for that in the prompt before combining the characters. ie "the man is holding a sword"
great, thanks
Great Tutorial 👍
Thanks Denis 🙏
Damn.. leonardo got big Ws with this.
Leonardo seems to be always adding in new features and updates 👍
Great guide! Is there any way to ensure that your clothes always stay the same?
Not really in Leonardo, the best way would probably to train a custom Flux model like in this tutorial: th-cam.com/video/5Z8fwEeWfRg/w-d-xo.html
@@taoprompts Thank you
Can you make it in vertical format?
Yes, you can use the same techniques on a vertical image.
Can you use the combined image to go back into the image generator, to generate a new image combined?
Yeah definitely, the image generator can work on any existing image
when you resize in the Canvas editor, don't you lower the final resolution (quality)?
You do, but it's easy to upscale again. The free Upscayl app does a great job of this.... remember to change dpi from 72dpi to 300dpi before you use Upscayl though as it doesn't change the dpi.
@@dekhog dpi matters only for printing, as I know
@@ivan_12345 Yes, true, but I print a lot - if it's only for screen lower dpi is fine.....
Great video! Do you know if it is possible to achieve this consistent characters through their API as well?
There is some documentation of how to use their api for Image Guidance (and character reference) here: docs.leonardo.ai/docs/generate-images-using-image-to-image-guidance
THIS is what I needed! Appreciate this.
💯The character reference feature was a great update.
I'm not too sure about your workflow, though. There are actually a few crucial steps that we wouldn't want to miss out on if using this method of separating and then combining two different images into one. I would add style references as well so the result would follow the original look. Leonardo is complex, but if used properly with the right workflow, it could produce great results
Using style reference would help with consistency. Content reference also helps with more consistent backgrounds.
nice!
very helpful video. is it possible to upload real face as character reference? thx
You can try, but it won't work as well as using an image generator from Leonardo Ai itself.
This maybe a dumb question but I am new to this. Could you have changed the background image( the trees the park) to grayscale. thereby the man and the women be full color? They would be the only color in the image? Thanks
You'd probably need photoshop to select the background and set the color saturation to grayscale.
@@taoprompts I have already created images on Leonardo in grayscale. You don't know that's ok I'll figure it out.
Which card support Leonardo AI to buy
Leonardo runs in the browser (cloud) so no need to buy any specific card
In Leonardo we can also generate images with two characters in a single frame?
Are these options only available in paid version, means on upgraded version.
I am trying many such generation websites and checking for such. I could only find that in your video. Also, taking a reference photo and adding the face we need in maintaining the consistency. This really matters for me. Thank you for the video.
I think character reference is only for paid plans. This feature works best if you use images generated with Leonardo, it won't work as well with your own photos.
awesome, thx
Hey Tao! Is there any good face swap AI tool for photos? Anything better than insight face swap?
For user friendly interfaces, I've only tried insight face and photoshop Ai. If you're more familiar with programming & machine learning you could try InstantID or similar tech: github.com/InstantID/InstantID
Hello, is it possible to create scenes with multiple static characters in Leonardo Ai for free?
I don't think so
Hello, Tao . Great content as always! I learned a lot from your videos. I currently have a Midjourney subscription and I’m pretty happy with the results I get there , but I struggle with getting the exact same poses as in the reference images I use . In Leonardo this seems to be pretty easy job with the Pose reference. Could you please give me an advise ?
Hey, that's one of the downsides of Midjourney. You don't really get fine-grained pose control the same way that you can do it in Leonardo Ai. Maybe they will add a feature for it in the future.
In general, pose reference is done with ControlNet. Ai's with that can map poses for you.
Thanks! I hope they will sort this out soon.
Is it available on the free plan?
No, it's only for paid subscription
@@taoprompts thank you
Merci pour le tuto
awesome
Thanks, It was great tutorial.
Can you introduce an AI to change clothes in the movies?
Pika Labs has a function to do video inpainting, so you can swap out clothing
I cover that in this video: th-cam.com/video/F4vZTKOCeds/w-d-xo.html
at the 17:59 timestamp. It does only work on 3 second videos though.
@@taoprompts I have tried Pika before but I did not get good results. Do you know of another program that can do this better and more professionally?
thank you.
💯, I hope this was helpful!
Can you help me maybe with a prompt I have? I want to have a hiphop style white t-shirt and white dungarees and have the dungarees with the straps not tight but loose. How can I accomplish this?
Maybe you could use image guidance with content guidance, or edge guidance with an example image reference that matches the look you are going for.
Does this work for anime / cartoon characters?
I haven't tested that out, but if you check Leonardo's official announcement, they show some examples of cartoon & illustration characters:
leonardo.ai/news/character-consistency-with-leonardo-character-reference-6-examples/
@@taoprompts Great, thanks!
This really helped. Any tips on how to animate two characters on the same scene?
I will have to test that out more
@@taoprompts thanks for the great tutorial!
esta faltando conceção
Main issue with leonardo is, its really just Stable Diffusion and SD really has a hard time making characters interact and NOT face the camera. Thats because Nearly all SD Checkpoints have been heavily trained on Selfies or similar posing shots. Midjourney and Ideogram are much better at prompt adherence and cinematic shots but Ideo doesnt have cref
That's a good point about the selfie data. Maybe someone will finetune a model for multiple character interactions at some point.
@@taoprompts At this point its worth waiting for SD3 to see if finally a solution has been made starting with the base model.
I face the same problem whenever I need one character to look back at another, or show back views of their poses. All came out frontals.
Leonardo Ai image generations always looks off, distorted or look cartoony.
It does lack the photorealism that Midjourney has, but you do get some more flexibility in the poses. You could try adding in a style reference, or elements to see if that helps.