This video delivers such valuable insights with a clear explanation. It’s engaging and very well-structured. Thank you for sharing this amazing content!
Inside the colours of the glitch, with the mask at 5:35, there are indications of the how the model is internally functioning ? AI glitches = unrecognised artform ? 😃
Do I always need to train new models in order to generate consistent characters together? What if I already have a ton of loras created of my characters? Is there a specific replicate repo I can go to that will allow me to just upload the image I'm trying to inpaint on, upload the mask, and then use my character lora to input my trained character lora?
Beautiful tutorial, very clear and then you are very good at explaining all the steps. thank you I really admire your generosity in giving all this precious information. I wish you immense luck
Hello Tao. I wanted to ask - Would you now consider FLUX over MIDJOURNEY ? Or MIDJOURNEY still is good according to u? and also, does Flux have an editor like MJ ?
yeah Midjourney is still good. The thing with FLux is the ability to train your own characters into it. Here are some new editing features in FLUX: blackforestlabs.ai/flux-1-tools/
Awesome job! I really what to try this. Could you please give a Replicate link that where there is a Mask ... for a person that already has a model trained?
Your tutorials are very good. Hopefully, soon AI will develop so that we’ll get easier, quicker and more natural-looking results when working with consistent characters & scenes.
yeah: take the photo you want to face swap, add a mask for the face, and inpaint using the trained model. You can use the same method shown in this video
That was awesome. Thanks Toa, you created female images in Midjourney, is it possible to use another AI like Freepik for create different images of person for training?
Love your videos!!! Question on this one: After you trained your model in replicate it and its done, it give you the option to Run the Trained Model. If I click on that, what Engine does it run? Is it still going to use Flux? What if I want to use another version of Flux - there are like three now! LOL Or can you only use what ever the button takes you to?
Thank you! What the training does is add a small customized component (for your dataset) to the existing Flux model. With the trained models, they are only meant to work with the specific version of Flux they are trained with.
I see all of the training images are images of the character’s upper body. I guess AI still cannot maintain consistency when it comes to the character’s body type & entire outfit. It seems like it can’t do full-body shots or shots that involve part of the lower body. Also, in your training images, I noticed that you didn’t include any images of the back of your head, so I wonder if the AI would struggle with generating images that show the back of the head. Lastly, your training images seem to be shot from the same camera angle, so I wonder how the AI would do at generating the same characters from different camera angles.
You can do the entire body, however the smaller the face is in the image, the harder it will be for the Ai to learn the details of the face. It does a decent job of generating different camera angles, including the back of the head. Although it may not look like the exact person since it hasn't seen that data
You really are one of the best ai producers with clear workflows. Keep up the good work and wish you an exponential growth 🙏🏾🔥
Thank you 🙏. I'll keep you updated on the latest tech!
This video delivers such valuable insights with a clear explanation. It’s engaging and very well-structured. Thank you for sharing this amazing content!
Thanks, I'm glad you found this helpful!
Wow! Tao you’re the best! Hands down. Thanks for the research and the help.👍🏾
Thanks! I was using this method recently and thought I'd share
Inside the colours of the glitch, with the mask at 5:35, there are indications of the how the model is internally functioning ?
AI glitches = unrecognised artform ?
😃
Yeah probably, it looks like the diffusion process isn't being done at all in those cases
Do I always need to train new models in order to generate consistent characters together? What if I already have a ton of loras created of my characters? Is there a specific replicate repo I can go to that will allow me to just upload the image I'm trying to inpaint on, upload the mask, and then use my character lora to input my trained character lora?
It depends on if the LoRA is compatible with the specific setup they have. I don't know how you'd use a Lora from somewhere else in replicate
@@taoprompts i was actually able to figure it out. Thanks for the reply!
Beautiful tutorial, very clear and then you are very good at explaining all the steps. thank you I really admire your generosity in giving all this precious information. I wish you immense luck
thanks! I thought a guide like this needed to be made since people kept asking about multiple characters 👍
Man, we need more videos of you exploring enchanted forests, encountering mythical creatures, and rescuing beautiful princesses 😊❤
I may need to slay some dragons also 🔥
Can you do full body with Leonardo ai?
you can, but the smaller the face is the less accurate the details will be
Hello Tao. I wanted to ask - Would you now consider FLUX over MIDJOURNEY ? Or MIDJOURNEY still is good according to u? and also, does Flux have an editor like MJ ?
yeah Midjourney is still good. The thing with FLux is the ability to train your own characters into it. Here are some new editing features in FLUX: blackforestlabs.ai/flux-1-tools/
SDXL lightning
Excellent.
Awesome job! I really what to try this. Could you please give a Replicate link that where there is a Mask ... for a person that already has a model trained?
Question what are some of your clients ask for mostly you to do so I can start developing my skills Thanks
mostly Ai video generation. Consistent characters, lip sync, etc.
I created a AI, now i love her❤❤❤❤
True Love
Your tutorials are very good. Hopefully, soon AI will develop so that we’ll get easier, quicker and more natural-looking results when working with consistent characters & scenes.
Woooah,dream come true!
This was a highly requested video! I think it's pretty amazing to be able to combine yourself with Ai characters like this
This is great, thank you! Do you know if it works for anime style images?
Thanks! You can use this on any types of characters, although anime won't be as accurate as real people
@taoprompts Awesome, thank you!
Amazing! Is it possible to only do a faceswap with this method? If yes would you be opposed to make a demonstration please 🙏?
yeah: take the photo you want to face swap, add a mask for the face, and inpaint using the trained model.
You can use the same method shown in this video
@taoprompts perfect, thanks for your response ♥️
will these flux dev loras work in flux Schnell?
I don't think so
Very informative
lovely
There must be a faster way to get it done. An inpainting workflow in ComfyUI with LoRA loader or so
thank you bro for the info, can you suggest a good tuto ?
@@pandamanou I haven't find something like that so I started modifying an inpainting workflow yesterday, it looks promising
yeah that would be possible. Although you'll still need the same steps of training multiple character Lora's and using masking/inpainting
That was awesome. Thanks
Toa, you created female images in Midjourney, is it possible to use another AI like Freepik for create different images of person for training?
yeah definitely if Freepik has a character consistency feature. i'm pretty sure Free pik also trains a LoRA on a flux model though
Love your videos!!! Question on this one: After you trained your model in replicate it and its done, it give you the option to Run the Trained Model. If I click on that, what Engine does it run? Is it still going to use Flux? What if I want to use another version of Flux - there are like three now! LOL Or can you only use what ever the button takes you to?
Thank you!
What the training does is add a small customized component (for your dataset) to the existing Flux model.
With the trained models, they are only meant to work with the specific version of Flux they are trained with.
@@taoprompts Cab you choose which model you want it to train with?
@@lineardrift right now on replicate you can choose between flux-dev or flux-schnell. On other platforms you maybe able to use a newer model
@@taoprompts Got it! Thanks!
Am I finally getting a girlfriend ?
you might be!
Please let me how lipsync one frame if multiple characters
you need to use runway multi character lip sync
Thanks
I see all of the training images are images of the character’s upper body. I guess AI still cannot maintain consistency when it comes to the character’s body type & entire outfit. It seems like it can’t do full-body shots or shots that involve part of the lower body. Also, in your training images, I noticed that you didn’t include any images of the back of your head, so I wonder if the AI would struggle with generating images that show the back of the head. Lastly, your training images seem to be shot from the same camera angle, so I wonder how the AI would do at generating the same characters from different camera angles.
You can do the entire body, however the smaller the face is in the image, the harder it will be for the Ai to learn the details of the face.
It does a decent job of generating different camera angles, including the back of the head. Although it may not look like the exact person since it hasn't seen that data
am i tripping or you've never showed which tool you used in rerplicate nor link it
I said in the beginning, replicate.com/ostris/flux-dev-lora-trainer/train
Do you have a free one ? This one is paid
there's a $10 coupon in the description
@@taoprompts Hi, the coupon in the description doesn't work anymore, do you have another one ?
@@gaspardmon I didn't realize it was expired, I don't have another one
@@taoprompts Ok, thanks
What is the best way to make some money with AI that actually works ?
thanks, for a music video would be a perfect workflow
definitely if you need multiple characters, something like this makes sense to use
@@taoprompts 2 characters, I upload fiew music videos, but they are all text to video, so not much good as what you have now..
@@gumvue.studio I'd definitely recommend using image-to-video for the best results
She needs a brush