I am completely blown away sir! I have watched a zillion videos trying to get to grips with animation using Comfyui. Your video is the first one that actually makes sense and lays each step out clearly. More power to you sir!! THANKS A TON !!! :)
It gives very good results, I like it a lot. Be aware the different interpolation methods give different results and also have different computer requirements. If you push for too many frames you may have memory errors
That is an interesting case...I do not know, it depends if the recognition model tracks the blurred face all the time. Good thing with the gradio demo is you can check frame by frame and correct where needed.
Check the videos with Animatediff and IPadapter. With different clips will be difficult and tricky. I think training a Lora would be the way to go, but I have not yet tried to do such a thing with comfy
do you have any idea why i keep getting this error everytime i try and run "TrackAnything" Traceback (most recent call last): File "/home/sairam/Track-Anything/app.py", line 506, in with gr.Row(scale=0.4): File "/home/sairam/tany/lib/python3.10/site-packages/gradio/component_meta.py", line 152, in wrapper return fn(self, **kwargs) TypeError: Row.__init__() got an unexpected keyword argument 'scale'
For TrackAnything you have to do it outside of SD WedUI, I am afraid. Unless someone creates and extension...the masking process I guess can be done via img2img batch, but I am not much familiar with Automatic1111
Hi. Can we put 3D clothes instead of just changing the color of the clothes? I mean, you still have to make a mask. So why can't you add, for example, "astronaut suit" instead of blue? And second question: only works if you remove the background beforehand? One more question. I watched the video on git hub. And there is a video where when you use the mask, the person is removed and the background is left.Does it have to be done in an editor or can it be done with your program without a video editor? It would be cool if with the help of your program it would be possible to delete both the character and the background without any editor.Your masks are good, a full size person and at a small distance works great.Congratulations!
Hi tanks! Several things: - I just changed the colors and try to keep the rest. I was not successful to control the color of the clothes as I wanted, when doing img2img. Sometimes was mixing up the colors of the different parts, etc. Segmentation in ComfyUI is possible, but for animations I found out TrackAnything was more efficient. I image that, with better control of the prompt you may still have what you want. Changing the type of clothes may be difficult. I have not tried, to be honest. I think with use of Loras could be possible. - You can also do it without removing the background. I however remove it before to 'help' SD to focus and do not get confused by the surroundings. The sampling speed also increases when there is no background. - I guess the video is from the github repo? In TrackAnything it seems you can inpaint and remove the subject. I have not tried it myself though. I prefer to use the editor from Runway, as it seems easier at this moment.
This is awesome !! My question is: Would this workflow allow you to change an existing video (anime) and change only the head on the character to something new? I have a clip that I want to change a head from having hair to being bald and with a beard
Great TUT,however when installing on win10,last step got this error:Traceback (most recent call last): File "E:\SDiff\Track-Anything\app.py", line 506, in with gr.Row(scale=0.4): File "E:\SDiff\Track-Anything\tany\lib\site-packages\gradio\component_meta.py", line 146, in wrapper return fn(self, **kwargs) TypeError: Row.__init__() got an unexpected keyword argument 'scale'
Hi thanks for the feedback! Yep, this is a little bit too complicated. I could have made it simpler by maybe changing one object only. I wanted to show that there are many possibilities working with segments, but I understand that makes the workflow more difficult
I am completely blown away sir! I have watched a zillion videos trying to get to grips with animation using Comfyui. Your video is the first one that actually makes sense and lays each step out clearly. More power to you sir!! THANKS A TON !!! :)
Frame intepolation looks very useful. I will definitely need to try it.
It gives very good results, I like it a lot. Be aware the different interpolation methods give different results and also have different computer requirements. If you push for too many frames you may have memory errors
Great tut, methodical and clear explanation of the process. Thanks !
Thanks!
Is it possible to use the model to segment Stable Difusion / Midjourney deformed pictures? (multiple fingers, blurry flaces, etc)
That is an interesting case...I do not know, it depends if the recognition model tracks the blurred face all the time. Good thing with the gradio demo is you can check frame by frame and correct where needed.
TrackAnything failed to install, it was installed all afternoon without success, can I have an episode of installation lesson
There has been changes in the gradio package. I have updated the requirements.txt file and it looks it works again
Thank you!
Is it possible to get consistent output without any glitches, of the same character? With different clips.
Check the videos with Animatediff and IPadapter. With different clips will be difficult and tricky. I think training a Lora would be the way to go, but I have not yet tried to do such a thing with comfy
do you have any idea why i keep getting this error everytime i try and run "TrackAnything"
Traceback (most recent call last):
File "/home/sairam/Track-Anything/app.py", line 506, in
with gr.Row(scale=0.4):
File "/home/sairam/tany/lib/python3.10/site-packages/gradio/component_meta.py", line 152, in wrapper
return fn(self, **kwargs)
TypeError: Row.__init__() got an unexpected keyword argument 'scale'
I do not know...maybe to reinstall gradio?
@@koalanation works just needed to uninstall installed gradio (v4.3.0) and install gradio version 3.48.0
Thanks. The original repo has not been maintained and I just did some tweaks to make it work. Good to know
@@koalanationsadly my 4gb vram was not enough to run it always getting running out of memory error
How do i do this on stable diffusion
For TrackAnything you have to do it outside of SD WedUI, I am afraid. Unless someone creates and extension...the masking process I guess can be done via img2img batch, but I am not much familiar with Automatic1111
Hi. Can we put 3D clothes instead of just changing the color of the clothes? I mean, you still have to make a mask. So why can't you add, for example, "astronaut suit" instead of blue? And second question: only works if you remove the background beforehand? One more question. I watched the video on git hub. And there is a video where when you use the mask, the person is removed and the background is left.Does it have to be done in an editor or can it be done with your program without a video editor? It would be cool if with the help of your program it would be possible to delete both the character and the background without any editor.Your masks are good, a full size person and at a small distance works great.Congratulations!
Hi tanks! Several things:
- I just changed the colors and try to keep the rest. I was not successful to control the color of the clothes as I wanted, when doing img2img. Sometimes was mixing up the colors of the different parts, etc. Segmentation in ComfyUI is possible, but for animations I found out TrackAnything was more efficient. I image that, with better control of the prompt you may still have what you want. Changing the type of clothes may be difficult. I have not tried, to be honest. I think with use of Loras could be possible.
- You can also do it without removing the background. I however remove it before to 'help' SD to focus and do not get confused by the surroundings. The sampling speed also increases when there is no background.
- I guess the video is from the github repo? In TrackAnything it seems you can inpaint and remove the subject. I have not tried it myself though. I prefer to use the editor from Runway, as it seems easier at this moment.
Your video is very good, but can you tell me more details, a lot of them don't understand ah
I tried to put as much detail as possible (maybe too much...). Can you tell me where you want more details?
This is awesome !!
My question is: Would this workflow allow you to change an existing video (anime) and change only the head on the character to something new?
I have a clip that I want to change a head from having hair to being bald and with a beard
These are awesome ideas! I think you need to do some inpainting. Try to do it directly in TrackAnything.
Very usefull thanks
Thanks to you!
Amazing 🤩
Thanks 🤗
Great TUT,however when installing on win10,last step got this error:Traceback (most recent call last):
File "E:\SDiff\Track-Anything\app.py", line 506, in
with gr.Row(scale=0.4):
File "E:\SDiff\Track-Anything\tany\lib\site-packages\gradio\component_meta.py", line 146, in wrapper
return fn(self, **kwargs)
TypeError: Row.__init__() got an unexpected keyword argument 'scale'
I don't know if it's a gradio version problem, haha, I'm not good at coding
Try to install the last version of gradio using pip install gradio from the command line
You are right, I tried serveral versions,gradio 3.48.0 resolve this problem,haha @@koalanation
remove installed gradio using "pip uninstall gradio" then install gradio-version-3.48.0 using "pip install gradio==3.48.0"
@@sairampv1 I changed the requirements.txt file with the gradio version you say. It seems to work again
i like your videos but can you explain it step by step
there are may steps that me as a beginner cant do...
Hi thanks for the feedback! Yep, this is a little bit too complicated. I could have made it simpler by maybe changing one object only. I wanted to show that there are many possibilities working with segments, but I understand that makes the workflow more difficult
I commented a issue ymthag I have on an other video that you made I need really some help
brain can't compute
🤣🤣🤣