Open JustFrederik opened 1 year ago
Will the function of training your own model and improving the existing one be supported?
Who knows which version of python you can use to https://github.com/qweasdd/manga-colorization and what dependencies?
@My12123 let the errors guide you. The v2 version is an improvement of this repo and will be merged into this repo, so just wait for a while or use clone the fork.
我个人觉得,着色器不应该内置到里面,这个开源,是用来机翻
你可以跑完后,再额外用着色器,而不是内置到里面,这会搞出更多的 bug
@thatDudo
我个人觉得,着色器不应该内置到里面,这个开源,是用来机翻
你可以跑完后,再额外用着色器,而不是内置到里面,这会搞出更多的 bug
它被添加的原因是因为文本在着色后需要进行渲染,以提高质量。无论如何,它在禁用时不会引发错误。这是一个可选功能,提高了可用性和易用性。
我个人觉得,着色器不应该内置到里面,这个开源,是用来机翻 你可以跑完后,再额外用着色器,而不是内置到里面,这会搞出更多的 bug
它被添加的原因是因为文本在着色后需要进行渲染,以提高质量。无论如何,它在禁用时不会引发错误。这是一个可选功能,提高了可用性和易用性。
https://github.com/OValery16/Manga-colorization---cycle-gan
https://github.com/qweasdd/manga-colorization-v2
这2个着色器有什么区别
https://github.com/DwangoMediaVillage/Comicolorization
https://github.com/Dakini/AnimeColorDeOldify
https://github.com/pvnieo/ManGAN
https://github.com/MarkMoHR/Awesome-Image-Colorization/blob/master/README.md
https://github.com/Atharva-Phatak/Manga-Colorizer 这些是使用上 着色器 的项目吗?
我个人觉得,着色器不应该内置到里面,这个开源,是用来机翻 你可以跑完后,再额外用着色器,而不是内置到里面,这会搞出更多的 bug
它被添加的原因是因为文本在着色后需要进行渲染,以提高质量。无论如何,它在禁用时不会引发错误。这是一个可选功能,提高了可用性和易用性。
我跑了下,翻译没有变化,显存占用多了3g
是否能用stable diffusion来上色?
@lumiamilk 这将是困难的,很可能不会,着色的描述不会自动创建或不太准确。 会有很多问题与充分自动化。
It will be difficult and most likely not, the description for coloring is not automatically created or it is not quite accurate. There will be a lot of problems with full automation.
Can you create a model for coloring anime (art, sketch, manga horizontal and vertical)? https://github.com/jantic/DeOldify/issues/475
What would your feature do?
I thought image colorization would be a usefull feature. The problem with colorization is that it has some problems with text and it gets harder to read the text. So the ocr could detect the text before and overlay the text after the image is colorized. it is not quite what this project is trying to achieve, however it would be a very useful feature(like the upscaler). I looked through the colorization projects on github and found these projects. I think deoldify could get a quite good result when a model is trained and manga-colorization is quite good already, but has no train script.
deoldify old architecture