RIFE stopped working and I have no idea how to fix it

[wonkey_monkey]

I had RIFE working a few weeks ago, but I've just gone to use it again and it's stopped working - no errors, it just inserts black frames where there should be interpolated ones.

I recently updated my video card drivers (GeForce GT 1030, driver version 32.0.15.7270/572.70) but I can't remember version I had before. Is there some specific version or type it needs instead?

[Z2697]

Which RIFE implementation did you use?

By the way, about the previous driver version, assuming you are using Windows, there's "driver store" that stores previous drivers.
Try https://github.com/lostindark/DriverStoreExplorer

[wonkey_monkey]

I just downloaded the latest release from here: https://github.com/Asd-g/AviSynthPlus-RIFE

but it's still not working (using a script that worked previously).

[Sharc]

- Remove Prefetch, if exists
- Deinterlaced footage with wrong field order? (produced black frames here for certain models)

[StainlessS]

Geeks3D main page:- https://www.geeks3d.com/

Geeks3D, Download Zone:- https://www.geeks3d.com/dlz/

Geeks3D GPU Caps Viewer shows various info and tests for Vulkan, OpenGL, OpenCL, CUDA, PhysX.
GPU Caps Viewer 1.64 (is a 32 bit application):- https://geeks3d.com/20250313/gpu-caps-viewer-1-64-released/

EDIT:
GPU Caps Viewer 1.63:- https://www.geeks3d.com/20240212/gpu-caps-viewer-1-63-released/
Seems that v1.64 is currently only available as portable, ie no setup exe. v1.63 has also setup.exe.

[Emulgator]

What works best here on my backup system (Internet access forbidden)
RIFE 1.2.2 under the combo Win10 19045.3570 i9-11900K @ 31.0.101.5534 / RTX3080 @ nVidia 561.09

What works poorly here on my main system (from then on I avoid upgrading like the plague)
RIFE 1.2.2 under the combo Win10 19045.5247 / RTX3080 @ nVidia 565.90
(i9-11900K @ 32.0.101.5972 can run no more iGPU side-by-side with dGPU, BSODs, so I have to hide iGPU)
Since a Win10 update: Daily first bluescreen, then second boot successful.
Will mirror my backup system onto that one, that's for sure.

Z2697, many thanks for pointing me to Driver Store Explorer !
This will help me to draw conclusions who tilted my main system.

[DTL]

Playing with new AVS+ test builds and attempt to test new ideas around denoise with RIFE about mixing of different models output via vsTTempSmooth (median-like mode). Current script to partial denoise of BetacamSP to DV25 capture:

LoadPlugin("vsTTempSmooth.dll")
LoadPlugin("RIFE.dll")

LoadPlugin("ffms2.dll")

Function RIFE_tr1_2m(clip c, int m1, int m2)
{
tr1_m1=RIFE(c, denoise=true, denoise_tr=1, skip=false, sc=true, sc_threshold=0.12, model=m1)
tr1_m2=RIFE(c, denoise=true, denoise_tr=1, skip=false, sc=true, sc_threshold=0.12, model=m2)

Interleave(tr1_m1,c,tr1_m2)
ConvertToYUV444(matrix="PC.601")
vsTTempsmooth(pmode=1, maxr=1, ythresh=255)
return SelectEvery(3, 1)
}

Function RIFE_tr1_2m_ly(clip c, int m1, int m2)
{
tr1_m1=RIFE(c, denoise=true, denoise_tr=1, skip=false, sc=true, sc_threshold=0.12, model=m1)
tr1_m2=RIFE(c, denoise=true, denoise_tr=1, skip=false, sc=true, sc_threshold=0.12, model=m2)

Interleave(tr1_m1,c,tr1_m2)
ConvertToYUV444(matrix="PC.601")
vsTTempsmooth(pmode=1, maxr=1, ythresh=10, ythupd=5, uthupd=10, vthupd=10, ypnew=5, upnew=5, vpnew=5)
return SelectEvery(3, 1)
}

FFmpegSource2("src.avi")

ConvertToPlanarRGB(matrix="PC.601", bits=32)

src_i=SeparateFields()
srceven=SelectEven(src_i)
srcodd=SelectOdd(src_i)

dn1_2m_e=RIFE_tr1_2m(srceven, 69, 73)
dn1_2m_o=RIFE_tr1_2m(srcodd, 69, 73)

dn1_2m_ly_e=RIFE_tr1_2m_ly(srceven, 69, 73)
dn1_2m_ly_o=RIFE_tr1_2m_ly(srcodd, 69, 73)

dn1_2m=interleave(dn1_2m_e,dn1_2m_o)
dn1_2m_ly=interleave(dn1_2m_ly_e,dn1_2m_ly_o)

weave(dn1_2m_ly)
#weave(dn1_2m)

ConvertBits(8)
ConvertToYV12()


2 output versions - dn1_2m_ly with only low deviation areas denoise (keeping fast motion high contrast areas where RIFE fail significantly) and max possible denoise with dn1_2m . Uses only tr=1 for lowest temporal distortions. With tr=2 RIFE fails more at the high motion areas.

Also new idea how to get more quality with a complex scenes with fast complex motion at lower tr with asymmetric frame interpolations requests - https://github.com/Asd-g/AviSynthPlus-RIFE/issues/10
Mixing of different models output helps to denoise at simple low contrast textures but errors from interpolation causes details loss or even bad shifted blends on high contrast sharp objects. For at least some scenes may be better to use single model output.