search

NVlabs / Sana

SANA: Efficient High-Resolution Image Synthesis with Linear Diffusion Transformer
https://nvlabs.github.io/Sana
Other
744 stars 26 forks source link

The generated 4K images appear somewhat blurry and out of focus. #14

Open i18nsite opened 6 days ago

i18nsite commented 6 days ago

The generated 4K images appear somewhat blurry and out of focus. I have partially enlarged all the images, including the comparison images, below for easier comparison.

For example, this image: CleanShot 2024-11-16 at 20 25 19

And this one, which has a noticeable blur: CleanShot 2024-11-16 at 20 27 56

For comparison, the effect from redpandaai: https://redpandaai.com/tools/ai-image-generator CleanShot 2024-11-16 at 20 22 03

中文版

生成的4K图像有些模糊和失焦。我已经将所有图片（包括对比对象图片）部分放大，以便更容易比较。

例如，这张图片： CleanShot 2024-11-16 at 20 25 19

还有这张，有明显的模糊： CleanShot 2024-11-16 at 20 27 56

作为对比，redpandaai的效果：https://redpandaai.com/tools/ai-image-generator CleanShot 2024-11-16 at 20 22 03

i18nsite commented 5 days ago

CleanShot 2024-11-17 at 21 30 15 https://www.modelscope.cn/models/iic/cv_nafnet_image-deblur_gopro 可以用这个降低模糊

i18nsite commented 4 days ago

CleanShot 2024-11-17 at 21 30 15 https://www.modelscope.cn/models/iic/cv_nafnet_image-deblur_gopro 可以用这个降低模糊

嗯，这个对是比较艺术虚化的作品效果不好会把造成一些色块。

https://huggingface.co/fal/AuraSR-v2

我发现另外一个细节增强的模型，这个可能比去模糊的效果更好。

lawrence-cj commented 1 day ago

Nice one. Will look into it.