Which tool has the best accuracy for extracting hardsubs from video?

[Shaorin]

Can someone tell me the best free tool with highest accuracy that can extract hard subtitles from video and convert to srt for windows?

[ProWo]

You could give this a try:
https://www.videohelp.com/software/VideoSubFinder

or newer:
https://github.com/timminator/VideOCR

The easier way is to find a subtitle online, g.e.
https://www.opensubtitles.org/en/search/subs

[Moralez]

Can someone tell me the best free tool with highest accuracy that can extract hard subtitles from video and convert to srt for windows?

I would use faster-whisper-xxl. It does a pretty good job, depending on the language.

[VoodooFX]

Can someone tell me the best free tool with highest accuracy that can extract hard subtitles from video and convert to srt for windows?

InpaintDelogo

[aheatherly]

Try the tool: https://github.com/SWHL/RapidVideOCR

[ilektryon]

Extract the images with the subtitles with InpaintDelogo.
Do not use VSF or any tool that uses VSF. It is just inferior compared to InpaintDelogo.
InpaintDelogo is easy to use if you organize the needed files and just open the avs file with AVSMeter by using a cmd.
I wish there was a GUI for this InpaintDelogo proccess.
Then open the images in the latest beta of SubtitleEdit. It supports Google Lens OCR in the latest beta.
Do the OCR in SubtitleEdit with Google Lens.
Google Lens OCR is amazing, I am shocked with it, its OCR is so accurate.

[VoodooFX]

InpaintDelogo is easy to use if you organize the needed files and just open the avs file with AVSMeter by using a cmd.
I wish there was a GUI for this InpaintDelogo proccess.

How do you preview the dynamic masks with AVSMeter?

For example, one of the "GUI"s you can use is AvsPmod GPo.

Do not use VSF or any tool that uses VSF. It is just inferior compared to InpaintDelogo

That's why I implemented the subs extraction in InpaintDelogo, VSF wasn't good enough for me.

[ilektryon]

How do you preview the dynamic masks with AVSMeter?

For example, one of the "GUI"s you can use is AvsPmod GPo.

I don't even calculate anything.
I always have the subtitles to the bottom and all videos are 1920x1080 in good quality and crop to 50% width.
I always use the following settings and I always get perfect results.

LWLibavVideoSource("InpaintDelogo.mkv", cache=false)
Crop(0,540,0,0)
InpaintDelogo(Loc="0,0,0,0", Show=4, DynMask=4, DynTune=210, DynMask4H=120)
SubsMask2Img(ImgInflate=1, ImgDir="output")

It always works perfectly and I am not missing any subtitle.
I always check for missing subtitles.
It's perfect

[VoodooFX]

I don't even calculate anything.

It always works perfectly and I am not missing any subtitle.
I always check for missing subtitles.
It's perfect

Then, you are are lucky that your videos fits that one setting.

LWLibavVideoSource("InpaintDelogo.mkv", cache=false)
Crop(0,540,0,0)
InpaintDelogo(Loc="0,0,0,0", Show=4, DynMask=4, DynTune=210, DynMask4H=120)
SubsMask2Img(ImgInflate=1, ImgDir="output")

You wrote that you use AI for OCR, in that case you want to pass raw images to OCR (not masks).

a = LWLibavVideoSource("InpaintDelogo.mkv", cache=false)
rawclip = a.crop(0,540,0,0)

a.InpaintDelogo(Loc="0,540,0,0", Show=4, DynMask=4, DynTune=210, DynMask4H=120)
SubsMask2Img(ImgInflate=1, ImgDir="output", Raw=rawclip, RawAvr=1)

Few notes:
RawAvr=1 is default
For "Raw" to work "_raw" subfolder should exist in your "output" folder.
Subs extraction can be speedup by ~2x automatically, if you use extraction directly from InpaintDelogo (no SubsMask2Img in your script), that's negative "Extract" option.

[ilektryon]

I don't even calculate anything.

It always works perfectly and I am not missing any subtitle.
I always check for missing subtitles.
It's perfect

Then, you are are lucky that your videos fits that one setting.

LWLibavVideoSource("InpaintDelogo.mkv", cache=false)
Crop(0,540,0,0)
InpaintDelogo(Loc="0,0,0,0", Show=4, DynMask=4, DynTune=210, DynMask4H=120)
SubsMask2Img(ImgInflate=1, ImgDir="output")

You wrote that you use AI for OCR, in that case you want to pass raw images to OCR (not masks).

a = LWLibavVideoSource("InpaintDelogo.mkv", cache=false)
rawclip = a.crop(0,540,0,0)

a.InpaintDelogo(Loc="0,540,0,0", Show=4, DynMask=4, DynTune=210, DynMask4H=120)
SubsMask2Img(ImgInflate=1, ImgDir="output", Raw=rawclip, RawAvr=1)

Few notes:
RawAvr=1 is default
For "Raw" to work "_raw" subfolder should exist in your "output" folder.
Subs extraction can be speedup by ~2x automatically, if you use extraction directly from InpaintDelogo (no SubsMask2Img in your script), that's negative "Extract" option.

Yes, I use AI, Google Lens uses AI.
Thank you, I will try your suggestions ASAP.
Of course you know better since you are the developer of this awesome tool