I love watching people use ai plus their talent to make something more than just an ai Generated picture. These videos show that ai doesn't replace people, just enables them to creates much more than they ever could alone. Much support friend.
Droppin back through here agin to say man this is still the best! Keep goin! We haven't heard from you in awhile but know you are inspiring people across the world friend.
Hey man, this turned out beautifully! Any way for you to do a tutorial that goes into more detail on how you did this, or if there is a place you can point us to that could allow for us to replicate this in our own projects? I got a school project coming up and would love to be able to do something creatively like this! Appreciate it!
Love it. This is probably the best workflow. Did you train a model specifically on your character? Or just apply a certain checkpoint on the footage and controllnetted the hell out of it? Or no checkpoint at all? Looking forward to future stuff you create!
Not really specifically train model. i just dreambooth myself with custom checkpoints from civitai and process controlnet IMG2IMG with some ebsynth. Then final touch with some compositing
Hello friend, congratulations on your evolution, it is amazing. Could you explain to me how you managed to do that rotoscoping with multicontrolnet? Just a little more detail please.
This is cool! Liked and followed! I noticed that you used img2img in combination with multiple control nets to generate multiple frames of images. However, the light and shadow on each face and the whole body seem to be constantly changing and differing from the previous one. But after using Ebsynth, the light and shadow on the face become stable instantly. Is this due to Ebsynth's keyframe extraction? But I'm not sure how to choose the keyframe extraction to achieve a stable solution for both light and shadow and flicker problems. I hope the author can give some guidance, thank you very much!
Then I noticed that the male character initially followed the 'input-img2img-Ebsynth' process, while the female character went through 'input with blender hair-img2img-Ebsynth'. Are these two different methods used to compare the effects, or was the male character done first, then the female character was created with the aid of Blender's hair simulation to achieve the best hair restoration effect? I'm a bit confused here because I'm not very familiar with Blender. Both characters look good after using Ebsynth, so does the Blender hair simulation help with the image animation in some way?
like i list in description workflow. Both character are done with mutiple ebsynth process. And mask it out only the good part to combine each other. i pick only the frame that have similar light and shadow to avoid as much inconsistant as possible. Oh yeah there are few trick to help sell the illusion like (Deflicker, limit framerate to 12 *to reduce uncanny feeling*)
@@JJ-yw3ug technically both male and female are using the same method. blender is use only to add wig. u can just go wear the real wig to save ur time from running simulation and composite it.
Wow Amazing! I just have two question, how far can you push the CFG Scale and and reduse Controlnet and still get the video coherent (push it to a different style like Pixar 3D characters)? secondly when you understand the workflow, how long dose 10 Seconds take you to get to this level? I made my calculations 2 months ago and got to about 2 weeks for 30 Seconds final production without the audio. Is it correct or am I being optimistic.
In this video CFG is around 7-8 with denoise strength around 0.5-0.6 + multi controlnet maintain the same pose and not too much away from original input. i don't understand ur 2nd question but the more frame / second u put in the longer it take. i use 3080 which is pretty fine with this but still take half a day to generate all these character frame. also all the process in this video took me a whole week.
[Urgent question]. The image storage function in Stable Diffusion(pc ver) is normal. It is impossible to work with the image because the image keeps getting deleted in the viewer. As soon as the image is created, the image disappears from the screen. How can I get the final image to remain intact on the screen???
@@BunnetSong The images are created normally. But in the screen viewer, the image disappears right after 100% and becomes blank. Sending as "image to image" is not possible.😭😭😭😭😭
@@MAKARUTV look like u tick something wrong the GUI setting. idk much all those stuff but i believe it is there. And If something wierd keep happen just reinstall it ( it dont take long )
Sadly no, im just not good at all for tutor thing. But with all these keyword u can go around searching more on it. Alot of other channel are doing tutor already
I love watching people use ai plus their talent to make something more than just an ai Generated picture. These videos show that ai doesn't replace people, just enables them to creates much more than they ever could alone. Much support friend.
estas hablando a corto plazo. pero a largo plazo si van a reemplazar a las personas. en fin xd
You are always pushing this workflow forward, amazing result.
Droppin back through here agin to say man this is still the best! Keep goin! We haven't heard from you in awhile but know you are inspiring people across the world friend.
Awesome!
Very impressive!! Congrats!
Thank you always! What should I do to change the costumes and hairstyles of the characters in the video?
Next level 💯
Amazing results!
Hey man, this turned out beautifully! Any way for you to do a tutorial that goes into more detail on how you did this, or if there is a place you can point us to that could allow for us to replicate this in our own projects? I got a school project coming up and would love to be able to do something creatively like this! Appreciate it!
Love it. This is probably the best workflow. Did you train a model specifically on your character? Or just apply a certain checkpoint on the footage and controllnetted the hell out of it? Or no checkpoint at all? Looking forward to future stuff you create!
Not really specifically train model. i just dreambooth myself with custom checkpoints from civitai and process controlnet IMG2IMG with some ebsynth. Then final touch with some compositing
Amazing work... I couldn't get a good result with EBsynth at all. I just get very distorted image, No matter what I do.
Hello friend, congratulations on your evolution, it is amazing. Could you explain to me how you managed to do that rotoscoping with multicontrolnet? Just a little more detail please.
Rotoscoping was done with third party software *AE rotobrush. Then i put it out with green background to easy mask out later after doing batch img2img
wow, looking good, but how do you remove the background from SD? (so much fickering)
wow This is really amazing.
Hey Bunnet, great content. how can we connec?
This is cool! Liked and followed! I noticed that you used img2img in combination with multiple control nets to generate multiple frames of images. However, the light and shadow on each face and the whole body seem to be constantly changing and differing from the previous one. But after using Ebsynth, the light and shadow on the face become stable instantly. Is this due to Ebsynth's keyframe extraction? But I'm not sure how to choose the keyframe extraction to achieve a stable solution for both light and shadow and flicker problems. I hope the author can give some guidance, thank you very much!
Then I noticed that the male character initially followed the 'input-img2img-Ebsynth' process, while the female character went through 'input with blender hair-img2img-Ebsynth'. Are these two different methods used to compare the effects, or was the male character done first, then the female character was created with the aid of Blender's hair simulation to achieve the best hair restoration effect? I'm a bit confused here because I'm not very familiar with Blender. Both characters look good after using Ebsynth, so does the Blender hair simulation help with the image animation in some way?
like i list in description workflow.
Both character are done with mutiple ebsynth process. And mask it out only the good part to combine each other. i pick only the frame that have similar light and shadow to avoid as much inconsistant as possible. Oh yeah there are few trick to help sell the illusion like (Deflicker, limit framerate to 12 *to reduce uncanny feeling*)
@@JJ-yw3ug technically both male and female are using the same method. blender is use only to add wig. u can just go wear the real wig to save ur time from running simulation and composite it.
bro pls reply are you traing your models( FINE TUNING ) or lora for this stuff
nice work keep it up😍🤩🤩🤩🤩
Wow Amazing!
I just have two question, how far can you push the CFG Scale and and reduse Controlnet and still get the video coherent (push it to a different style like Pixar 3D characters)?
secondly when you understand the workflow, how long dose 10 Seconds take you to get to this level?
I made my calculations 2 months ago and got to about 2 weeks for 30 Seconds final production without the audio. Is it correct or am I being optimistic.
In this video CFG is around 7-8 with denoise strength around 0.5-0.6 + multi controlnet maintain the same pose and not too much away from original input.
i don't understand ur 2nd question but the more frame / second u put in the longer it take. i use 3080 which is pretty fine with this but still take half a day to generate all these character frame. also all the process in this video took me a whole week.
Hey, which stable diffusion model did you use?
Bunnet. how are you?. like,you created fantastic ~ ✨
Wow great content I’m amazed
Hi Bunnet, can I check what do you use to do the rotoscoping and why do you need to rotoscope? Thanks 🙆🏻♂️
Rotoscope is to use as alpha layer later (mask out people from background) after doing batch img2img
Good god man... genius..
can you test us how to do a random 3d geometry camera projection
howd you get the 3d geometry to match the ai image?
great work!
Omg blowing mind
Fantastic work man.👍
Good Stuff my guy!!
How you do this, teach plz
[Urgent question]. The image storage function in Stable Diffusion(pc ver) is normal. It is impossible to work with the image because the image keeps getting deleted in the viewer. As soon as the image is created, the image disappears from the screen. How can I get the final image to remain intact on the screen???
It not go anywhere. Check the Automatic1111 directory there is a folder called output. All generated images store there
@@BunnetSong The images are created normally. But in the screen viewer, the image disappears right after 100% and becomes blank. Sending as "image to image" is not possible.😭😭😭😭😭
@@MAKARUTV look like u tick something wrong the GUI setting. idk much all those stuff but i believe it is there. And If something wierd keep happen just reinstall it ( it dont take long )
@@BunnetSong Thank you for your sincere reply.😁😁😁😁
Amazing! Keep it up.
Can u made a tutorial for the camera projection? i dont get it how u did it
im pretty bad at teaching. but this video might help u understand it
th-cam.com/video/ua8CrGf6wP8/w-d-xo.html
holy sh**
Lmao that's crazy good job bro
nice video
Can you do a full tutorial about this ?
Sadly no, im just not good at all for tutor thing. But with all these keyword u can go around searching more on it. Alot of other channel are doing tutor already
when you look closer to it, it's uncanny, the eyes are really strange and the shadows aren't realistic
Ye try making vfx of such quality at home by urself
Ai alone is uncanny. It is what is, for now
Shaky camera aint cool.
Hi Bunnet! Do you have a email or instagram where I can shoot you a message?
sorry i only use FB. u can add me there : t.ly/H3sma