Using Stable Diffision models for Colorization

[didris]

Hi didris,

  In the case it could be useful I added a GUI to DiTServerRPC see post: #1.425

Dan

Hi, Dan

The installation was not easy, but I managed to get it working in the end. I think the problem was related to

tkinter_embed

in

requirements.txt

, which was causing issues in my environment. After removing it, everything installed correctly.

Overall, the software works now, and I successfully encoded five videos with durations of 10–15 minutes. Speed is 5 sec/frame with RTX5090. During testing, I noticed three issues:

1. It seems to work reliably only with MP4 and MKV files. I tested a MOV file, and it only started processing after I renamed it to MP4. Even then, it encoded only about 2 minutes and 30 seconds out of a 14-minute video.
   
2. The final output videos have no audio track.
   
3. In many videos, the reference frames were not colorized consistently, especially in the second half of the footage. In one particular video, almost none of the reference frames were colorized. I am not sure what this depends on or what type of source material works best. Perhaps a different Qwen model would help.
   
   

Despite these issues, this is a wonderful project and has a lot of potential. A few fixes would make it close to perfect.
I would also like to mention that with the first version I never had problems with frame colorization itself, apart from occasional inconsistencies where the same objects received different colors.

Congratulations on the great work, and thank you for making this available.