ComfyUI InvSR is the savior of blurry old pictures - actual test

The node used is from this project (unofficial implementation, the official one cannot run)
github.com/han...
Installation
1. Enter the ComfyUI/custom_nodes directory
2. Copy the project
git clone github.com/yuv...
3. Install the extension (I didn't perform this step and it worked, maybe I already had the dependency package)
pip install -r requirements.txt
How to use
Use the workflow provided in the project, drag it to ComfyUI, select the image and run it.
The first execution will automatically download the required models
InvSRSampler basically uses the default parameters, unless the graphics card memory is very large, you can adjust it appropriately
If the size is less than 16GB, the default value is the best parameter.
In the actual test, InvSR can indeed turn blurry images into high-definition ones, but not 100%. Some images have good original quality, but will be distorted after enlarging. The original images may have been compressed by some software, which may also affect the quality.
Graphics card uses RTX3060 12GB
#comfyui
#HD Zoom
#Blurred old picture

[@kana-ci1qr]

不能自动下载，手动下载sd-turbo和noise_predictor_sd_turbo_v5.pth这两个文件之后应该放在哪个路径下面呢？

[@aidecspace]

手動下載的話
sd-turbo 放在 ComfyUI/models/diffusers 資料夾名稱可能要改成叫---> models--stabilityai--sd-turbo
noise_predictor_sd_turbo_v5.pth 放在 ComfyUI/models/invsr 這個目錄 (沒有invsr的話自己新增一個)

[@aidecspace]

我分享了一份模型放在夸克，還在上傳中
链接：pan.quark.cn/s/46e23c7aa55a
提取码：J1Aw

[@kana-ci1qr]

@@aidecspace非常感谢你的迅速答复，不过我放在下面这两个路径中都报错，说找不到，你能看看问题出在哪吗？麻烦了。
ComfyUI_windows_portable\ComfyUI\models\diffusers\models--stabilityai--sd-turbo
ComfyUI_windows_portable\ComfyUI\models\diffusers\models\stabilityai\sd-turbo

[@aidecspace]

@@kana-ci1qr 你的裡面是長 blobs , refs , snapshots 嗎？ 它的是長這樣的目錄 跟 huggingface 上的不太一樣
等等試試看我分享在夸克的試試看

[@aidecspace]

@@kana-ci1qr 因為不能上傳截圖，我錄了一個它裡面的目錄結構 ruclips.net/video/noWP5zPwjEU/видео.html