Patreon is shitting itself again so I have to paste the link here instead of having it embed above the post: https://icedrive.net/0/deKvZMQyxf
This update mostly consists of QoL improvements. The encoding settings can now be directly accessed from the main tab, and they now have a color format option (for 10-bit or 4:4:4 encodes - but not HDR!), the preview tab can be disabled, and there's a button to reset all ...
This update fixes some slowdown with RIFE 3.1 and adds a new feature that will pause the interpolation automatically if auto-encode cannot keep up. This way you can avoid filling up your disk space if you don't have much left.
Full changelog on Discord and in Flowframes.
2021-07-01 20:16:32 +0000 UTC View PostThis update adds the latest RIFE 3.8 model (the previous ones after 3.1 were skipped because they were not as good, or broken). I'm not sure about the quality yet, which is why 3.1 is still the default.
This new model natively supports transparency, now you can interpolate transparent GIFs with better quality and without running RIFE twice.
Full changelog on Discord and in Flowframes.
2021-06-23 18:45:42 +0000 UTC View PostThis update brings a nice little feature: Backups during interpolation. If enabled (in Interpolation Settings), the interpolated video will be copied to the output folder even before it's 100% done, so if your program/PC crashes, you will have an incomplete video instead of nothing.
Full changelog on Discord and in Flowframes.
2021-06-10 09:37:03 +0000 UTC View PostMostly a bugfix update, but with one new feature: Image Sequence inputs can now be used directly, without any importing or copying, if they are compatible.
Full changelog on Discord and in Flowframes.
2021-06-01 18:23:54 +0000 UTC View PostThis one's pretty big again - RIFE v3 is out, and Flowframes now includes it (both CUDA and NCNN implementations support it, with the new 3.0 and 3.1 models). For all video encoding enthusiasts, AV1 encoding is now up to 35% faster. Additionally there's quite a bunch of smaller changes which you can see in the changelog.
Full changelog on Discord and in Flowframes.
2021-05-24 14:36:59 +0000 UTC View PostSadly 1.26.0 was not without bugs, so this is mostly just a bugfix update.
It fixes problems with image sequences, looping, the batch queue, and video encoding.
Full changelog on Discord and in Flowframes.
Note: I am looking into a way to update Flowframes without having to re-download all components. Until then, if you don't want to redownload the Pytorch files, you could install the new version to a different location, don't enable Pytorch in the...
2021-05-15 13:57:20 +0000 UTC View PostThis was originally meant to be 1.25.2, but I ended up adding so many features that it became 1.26 :P
This update adds a nice "Debug Tools" window which has a Log Viewer and Config Editor. Image Sequence inputs no longer need the "importing" step if they are already compatible, more video metadata will be preserved (display aspect ratio and color space), the GUI no longer freezes during processing, and a lot more.
Full changelog on Discord and in Flowframes.
2021-05-11 10:34:42 +0000 UTC View PostInitial Beta release of my brand new GUI tool: ReStyler!
It's a tool for applying neural style transfer. It's based on this implementation.
Only runs on Nvidia GPUs! GTX 750 Ti or newer. 8 GB VRAM recommended.
Just download the 7z, extract it with 7zip, and run Restyler.exe.
2021-05-08 14:12:27 +0000 UTC View PostThis update adds a button to pause a running interpolation, there's AV1 encoding support, and image sequences can now optionally use JPEG or WEBP instead of PNG. Additionally, accurate deduplication not working with JPEG frames was fixed.
Full changelog on Discord and in Flowframes.
2021-05-01 14:58:30 +0000 UTC View PostQuite a big update once again - and for the first time on Patreon, bundled in the brand new Flowframes Installer!
This update adds another AVI codec, FPS limiting now also works for image sequence outputs (will export fewer frames so you can manually encode), there's an option to import video frames as JPEG instead of PNG (after some research, I concluded that this is usually worth it as there is almost no quality loss, but the files are much smaller). Then there's a whole bunch of UI/U...
2021-04-27 19:10:44 +0000 UTC View PostSo, Flowframes has been getting so much attention recently that my traffic quota of 4 TB per month was exceed yesterday, so after just 13 days.
As a temporary solution, I have rehosted it on someone's personal cloud.
There will be a permanent solution soon as I'm working on the Flowframes Installer, which is a web-based installer.
The 1.24.6 post has been updated with a backup link.
2021-04-14 07:16:37 +0000 UTC View PostJust when I thought I got rid of all the bugs, I discovered a few more.
This one should be the most stable Flowframes build ever :P
It fixes some rare bugs like audio transfer not working with trimming, MKV muxing failing if the input has no attached metadata, and CUDA implementations not starting if running on CPU or a GPU that wasn't detected ...
2021-04-11 15:48:22 +0000 UTC View PostThis is most likely the final 1.24 version.
Fixes a critical bug (manual FPS input for image sequences not working) and adds a bunch of useful little features, like an option to preserve MKV metadata (embedded fonts etc) and lossless NVENC encoding (CRF 0).
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-04-06 13:30:35 +0000 UTC View PostSo scene detection was somehow still broken in 1.24.3, but now it really works.
Also, I fixed a video encoding bottleneck - it's now up to 370% faster with NVENC and up to 170% faster with CPU codecs!
Also, paths with special characters no longer break Pytorch-based AI implementations.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-04-04 18:25:03 +0000 UTC View PostOnce again some fixes - it should finally be stable now lol.
Most importantly, scene change transitions now work perfectly again, even with deduplicated or VFR content. Also, deduplicating (or picking scene changes) manually in step-by-step mode now works again - in 1.24.2, it would get ignored. And last but not least, CUDA implementations can be used without a recognized Nvidia GPU again, in case you want to run it with CPU or an undetected (Quadro/Tesla) card.
Full changelog on ...
2021-04-03 13:38:43 +0000 UTC View PostThis update mostly has some fixes (scene detection with image sequence inputs, VFR inputs, ...) and minor convenience improvements.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-04-02 19:28:15 +0000 UTC View PostTurns out there was one critical bug I missed for 1.24.0: Subtitles were always disabled.
This hotfix gets rid of that problem, and subtitle transfer should work fine again.
If you don't use the subtitle feature, there's no need to update, 1.24.0 is fine apart from that one bug.
2021-03-25 18:17:44 +0000 UTC View PostSo, here it is, the final 1.24.0 release!
It mostly fixes audio problems and adds more codecs (mp3 and wma) than can be transferred losslessly from the input file when using MKV output.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-03-24 20:28:43 +0000 UTC View PostSo, Pre-Release 2 was kinda broken, so here's a new update that should work better. Might be the final release if there are no bugs I missed.
Scene change transitions work perfectly again, and some problems in step-by-step mode were fixed.
Full changelog on Discord and in Flowframes.
The in-app updater will still say that 1.23.4 is the latest version as this is an incomplete Pre-Release.
Password can be viewed by going to your membership tab on my creator page.
2021-03-23 19:20:50 +0000 UTC View Post1.24.0 is still not complete, but I didn't want to let you guys wait too long, so here's a new build.
There's a new major (more or less) feature: Blended scene changes! This means that instead of duplicating the last frame of the previous scene, Flowframes will now fade the previous scene into the new one which looks a lot cleaner than trying to interpolate it.
This is disabled by default for now (as it needs some extra processing time, roughly 500ms per scene change) but can be e...
2021-03-21 19:15:28 +0000 UTC View Post
To avoid conversion fees (as I am in the EU) from USD, I've switched my Patreon currency to EUR.
This might slightly change the tier prices depending on your currency, however, it should mostly be the same as before.
2021-03-21 15:51:56 +0000 UTC View PostEarly release of upcoming 1.24.0, including FLAVR!
Might contains some bugs as I made it a priority to get a FLAVR build out before anything else, but those will be fixed in the coming days.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-03-14 19:52:32 +0000 UTC View PostYet another minor update to extend 1.23. It adds a disk space check that runs in the background, enables scene detection for frame sequence inputs, and some bugs have been fixed, like MKV outputs having audio stutter.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-03-10 20:50:41 +0000 UTC View PostThis update adds the option to customize the output filename pattern and uses a different way of merging audio and subtitles from the original video, no longer requiring extracting them first, saving time and disk space.
In addition, multiple bugs have been fixed.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-03-03 22:09:52 +0000 UTC View PostThis update adds the RIFE 2.4 model to both RIFE-CUDA and NCNN.
Do note that in my personal experience, 2.3 is a little bit better, so if you use CUDA, you should probably stick to 2.3.
The update also fixes image sequence inputs not working due to the locked input FPS being 0.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-02-28 15:44:58 +0000 UTC View PostThis update adds the RIFE 2.3 model which is better than ever for real-world content (for Anime, 1.8 is probably still the best choice). It also fixes a bunch of bugs.
Full changelog on Discord and in Flowframes.
Password can be viewed by going to your membership tab on my creator page.
2021-02-26 11:21:53 +0000 UTC View PostThis one took a bit longer, but it's also quite a bit bigger than originally planned!
As mentioned on Discord, the resuming feature was postponed (again lol) but there's tons of other great stuff.
Fast Mode makes RIFE at least 20% faster on RTX cards, video inputs can be trimmed, multiple audio tracks can be transferred from the input, and the quick settings tab provides faster access to frequently used options like deduplication and looping.
Full changelog on Discord...
2021-02-24 21:47:44 +0000 UTC View Post
This is a little side project - Check the screenshot and you might know what kind of AI this uses :)
Will release an early version for Patrons soon.
2021-02-13 01:01:27 +0000 UTC View PostI thought I'd get to 1.23, but RIFE 2.1 and 2.2 came out, plus I found another bunch of bugs that were fixed, so here's yet another 1.22 update.
It also includes the 2.0 model for RIFE-NCNN, which sees a ~40% speed boost with it.
Some of the fixed bugs include a double frame on looped interpolations and extremely high RAM usage for accurate deduplication.
Check DMs for the download password.
2021-02-12 15:04:40 +0000 UTC View Post