Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

generated noise as outputs on macbook #18

Open
john2stai opened this issue Nov 23, 2024 · 17 comments
Open

generated noise as outputs on macbook #18

john2stai opened this issue Nov 23, 2024 · 17 comments

Comments

@john2stai
Copy link

tried the example t2v workflow, and all I got was this:

LTXVideo_00001.mp4

No error warning, so might be something wrong with the decoding?

@ivanmea
Copy link

ivanmea commented Nov 23, 2024

I got the exact same result with my Macbook (m3 pro max) with i2v workflow and the latest version of ComfyUI

LTXVideo_00001.mp4

@john2stai
Copy link
Author

I got the exact same result with my Macbook (m3 pro max) with i2v workflow and the latest version of ComfyUI

LTXVideo_00001.mp4

I even downgraded from python 3.11 to python 3.10, but still not change. Did you see any error in you terminal, while LTXVideo was running?

@ivanmea
Copy link

ivanmea commented Nov 23, 2024

No error in the terminal for me.

@luffy889
Copy link

same issue on M4M

@john2stai
Copy link
Author

john2stai commented Nov 24, 2024

I got it working!! Probably a temporary fix, but nonetheless
I downgraded pytorch with the following command:

pip install pytorch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1

Hope someone from the LTXVideo team could help us mac users to have this working with the latest pytorch.

@luffy889
Copy link

I got it working!! Probably a temporary fix, but nonetheless I downgraded pytorch with the following command:

pip install install pytorch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1

Hope someone from the LTXVideo team could help us mac users to have this working with the latest pytorch.

Works indeed! Had to do this:

conda install python=3.8
conda install pytorch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 -c pytorch
pip install -r requirements.txt (from ComfyUI folder)

@ivanmea
Copy link

ivanmea commented Nov 24, 2024

Thank you @john2stai !! It's working now !

@RMarius78
Copy link

I have same issue here on Macbook pro 3 Max.
Running this on a conda environment with python 3.10.13.
If I downgrade to 3.9 LTXVideo will not install since it requires Python 3.10
Models:
checkpoints => ltx-video-2b-v0.9.safetensors
text_encoders => t5xxl_fp16.safetensors
PixArt-XL-2-1024-MS/text_encoders => 2 x ~9GB safe tensors.

PyTorch 2.5.1
FFMPEG 7.1
Lfs 3.6.0

Not sure what else I am missing, would love to get this running on M3M.
Thank you.
image

@john2stai
Copy link
Author

I have same issue here on Macbook pro 3 Max. Running this on a conda environment with python 3.10.13. If I downgrade to 3.9 LTXVideo will not install since it requires Python 3.10 Models: checkpoints => ltx-video-2b-v0.9.safetensors text_encoders => t5xxl_fp16.safetensors PixArt-XL-2-1024-MS/text_encoders => 2 x ~9GB safe tensors.

PyTorch 2.5.1 FFMPEG 7.1 Lfs 3.6.0

Not sure what else I am missing, would love to get this running on M3M. Thank you. image

try
conda install pytorch==2.4.1 torchvision==0.19.1 torchaudio==2.4.1 -c pytorch

downgrade your pytorch down to 2.4.1. At least, that's what works for me!

@RMarius78
Copy link

Thank you so much, it worked. In my case the culprit was torch vision, after downgrading pytorch to 2.4.1 it updated torch visions to 0.15, so I just needed to upgrade torch vision to 0.19.1 by itself.

@roninDday
Copy link

how long time did you guys who user of Mac get the result of LTX in(1216x704|41)? i'm struggling between mac and pc.

@RMarius78
Copy link

On the binned M3 Max (30 core GPU and 14 cores CPU) at 768x512 with 40 steps and 97 length (~4 sec) it takes about 1.5 Minutes per second of video. Quite a long time :(
On the RTX3080 on my PC same exact workflow runs 39 seconds, but it sounds like an airplane is taking off :)
CUDA is definitely faster, but I did not expect such a large gap. The M3 Max is being destroyed here.
My RTX3080 is the founder editions with 10 GB of ram and the most it used was 8GB, I was thinking it would crash since two large models are being used the 9GB LTX Video and the 9 GB t5xxl_fp16 text encoder.

@john2stai
Copy link
Author

how long time did you guys who user of Mac get the result of LTX in(1216x704|41)? i'm struggling between mac and pc.

in terms of speed, mac without cuda support can't compete with pc.

@roninDday
Copy link

768x512 with 40 steps and 97 length, same situation on PC with 4080S spend 30seconds,

on my boss' M1 Max mac studio 64G, it takes about 6minuters,

On the binned M3 Max (30 core GPU and 14 cores CPU) at 768x512 with 40 steps and 97 length (~4 sec) it takes about 1.5 Minutes per second of video. Quite a long time :( On the RTX3080 on my PC same exact workflow runs 39 seconds, but it sounds like an airplane is taking off :) CUDA is definitely faster, but I did not expect such a large gap. The M3 Max is being destroyed here. My RTX3080 is the founder editions with 10 GB of ram and the most it used was 8GB, I was thinking it would crash since two large models are being used the 9GB LTX Video and the 9 GB t5xxl_fp16 text encoder.

@luffy889
Copy link

M4M with 40 cores does 768x512/97/30steps in 168 seconds. Just checked desktop 4090 on same settings... 11 seconds. Ouch

@roninDday
Copy link

M4M with 40 cores does 768x512/97/30steps in 168 seconds. Just checked desktop 4090 on same settings... 11 seconds. Ouch

the best record on mac maybe at present,
i guess i can suffer M4 with 24G memory......

@RamboRogers
Copy link

I spent 3 hours on this..... same issue.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

6 participants