Flowframes Early Access Download: 1.37.2
Added 2022-06-01 17:14:29 +0000 UTCFractional interpolation now works fine with RIFE-NCNN, both scene detection and deduplication have been fixed.
Also, the RIFE 4.1 model has been added to RIFE CUDA and NCNN however the NCNN version seems to be broken and causes flickering. Stick to 4.0 for now.
Full changelog on Discord and in Flowframes.
Comments
before python was installed with installer up to 3.6? how to get that back in this version
Celestial Creator
2022-07-09 02:51:26 +0000 UTCWell, it turns out that unchecking the "Make Output Match Input Duration" option was a solution to make past generating the frame order step. Not sure why that option is an issue in this version, when in 1.35 did not pose an obstacle... In any case, now I just hope that audio will not mismatch in final mux, but it's a one frame of difference, how bad can it be...
Milica Zivanovic
2022-07-03 17:33:52 +0000 UTCI've just tried x5 interpolation with Rife 4.1 in this version (CUDA), and the error appeared in the step of generating frame order: "Error generating frame order information: Sequence contains no elements". Did any of you get this? How to solve. I am trying now to do the same with Rife 4.0, but expect the same will happen... I use the same settings as with free version that I had no issues with. Not sure what is going on?! Edit: Yeah, the same happened also with 4.0 and NCNN, but I think that version has nothing to do with it, as the issue seems to be before rife event starts...
Milica Zivanovic
2022-07-03 12:35:53 +0000 UTCIn general, does rife-ncnn use less gpu than rife-cuda? Over time in rife-ncnn the gpu load rate drops from 95% to less than 40%. The ai model is rife-4.0, and the gpu is Rtx 3070. The operating system is Windows 10. Does it any solution?
jo
2022-07-03 11:13:17 +0000 UTCI am getting this error too
JoΓ£o Paulo PE
2022-06-28 19:16:28 +0000 UTCI am getting also the error that i should install Python/Pytorch. I am using RIFE (NCNN) 4.0 Tried 1.37.1 with no errors.
Michele Meschke
2022-06-19 15:33:28 +0000 UTCDid you select "RIFE (NCNN)" in the AI dropdown?
LTSM
2022-06-18 15:39:22 +0000 UTCDid you select "RIFE (NCNN)" in the AI dropdown?
LTSM
2022-06-18 15:39:19 +0000 UTCalso get error, darkmoon what is the solutoiopn?
Kenneth Flournoy
2022-06-18 14:02:23 +0000 UTCI am getting the same error message as Nika does with the latest version. The free version works just fine but as soon as I install the patreon build both versions give me this error.
Lukas
2022-06-15 09:30:37 +0000 UTCI am getting this error message "Neither the Flowframess Python Runtime nor System Python Installation could be found! Either redownload Flowframes with the embedded Python runtime enabled or install Python/Pytorch yourself." Im not quiet sure what is missing as my previous installation of Frameflow worked just fine.
Nika
2022-06-14 10:56:57 +0000 UTCGotcha, thanks for the info. I tested both and either one or the other is faster depending on video quality, but either way they're both fast. Look forward to future updates!
Jake
2022-06-02 11:20:36 +0000 UTCQuality should be the same, nowadays I recommend NCNN because it's faster with 4.x models.
LTSM
2022-06-02 10:49:52 +0000 UTCis rife NCNN or CUDA the higher quality one? And thanks for the help
Jake
2022-06-02 10:34:50 +0000 UTCOnly with RIFE-NCNN, and only when using fractional factors, yes
LTSM
2022-06-02 10:23:17 +0000 UTCJust as reference, deduplication and scene detection was only broken on NCNN prior correct?
Jake
2022-06-02 09:34:28 +0000 UTCPing me on the Discord in the support channel. Also there's no real reason to use CUDA because NCNN is now faster, unless you want to test the 4.1 model.
LTSM
2022-06-01 22:30:02 +0000 UTCPing me on Discord and I can take a closer look
LTSM
2022-06-01 22:29:32 +0000 UTCgetting a no encoding error, reset settings to default, still getting the error...every version prior to this worked fine, any suggestions? on rtx 3070, 64gb ram, tried mp4, mkv, and mov...attempt to 4.1RIDECUDA
Kenneth Flournoy
2022-06-01 19:14:39 +0000 UTCNot sure if that was directed at me. The file import slowness occurs before you even select a custom model - simply opening the video file is where the issue is hit.
g90ak
2022-06-01 18:21:20 +0000 UTCcan you please explain a little on how to you use custom models?
Kenneth Flournoy
2022-06-01 18:17:30 +0000 UTCdeh wuhtid ih cuhh
Kenneth Flournoy
2022-06-01 18:14:00 +0000 UTCBummer. Files seem to be compatible, they've imported immediately for all versions up to and including 1.35. Few others have reported this on Discord as well. Uncompressed video is a non-option for me past 1.35.
g90ak
2022-06-01 18:10:38 +0000 UTCIt's not slow if the files are compatible. Nothing's broken right now from what I know.
LTSM
2022-06-01 18:06:34 +0000 UTCHey thanks alot!!!!! Gonna try this.
EvilMadness
2022-06-01 17:26:30 +0000 UTCnice
Andrew
2022-06-01 17:24:51 +0000 UTCthank you!
Diego Chavez
2022-06-01 17:23:17 +0000 UTCAny plans to address the super slow import process for large files that was broken on versions after 1.35?
g90ak
2022-06-01 17:16:28 +0000 UTC
