changelog.full.txt

Flowframes 1.39.0 Changelog:
- Added real-time interpolation output mode (with RIFE-NCNN-VS)
- Added RIFE 4.4 and RIFE 4.3 models
- Option "Import HQ JPEG instead of PNG" no longer breaks with RIFE-NCNN-VS
- Fix AV1 export showing an error when exporting >240 FPS that's limited to <=240 FPS


Flowframes 1.38.3 Changelog:
- Slow-motion feature now works with RIFE-NCNN-VS
- Added RIFE 2.3 model to RIFE-NCNN-VS (only 2x supported!)
- Updated RIFE-NCNN to release 20220728
- Updated ffmpeg to include SVT-AV1 1.2.0 (previously 1.0.0 was included)
- Fixed image sequence inputs not working
- Fixed scene change blending not working with integer interpolation factors
- Fixed the function that checks for the max vulkan GPU thread count (was unreliable)


Flowframes 1.38.2 Changelog:
- FPS limiting (downsampling) now works with RIFE-NCNN-VS
- RIFE-NCNN-VS will no longer overwrite existing output files
- Fixed issue where RIFE-NCNN-VS would try to use too many GPU threads
- Fixed issue where a warning would pop up every time Flowframes is started without Pytorch
- Replaced the default Message Box with a custom window, fixes some UI weirdness


Flowframes 1.38.1 Changelog:
- RIFE-NCNN-VS will now use the GPU ID set in the settings (multi-GPU not possible for now)
- Trimming the input video now works with RIFE-NCNN-VS
- RIFE-NCNN-VS now dynamically adjusts threads based on resolution, reducing RAM usage
- Trying to export >240 FPS video in AV1 will now show an error as SVT doesn't support this
- Fixed frame counting sometimes failing
- Fixed CLI -factor argument not being applied


Flowframes 1.38.0 Changelog:
- Implemented VapourSynth-RIFE-ncnn-Vulkan by HolyWu (interpolate completely in RAM)
- Improved the Pytorch availability check, unsupported implementations should no longer show up
- Possibly fixed bug where RIFE-CUDA would run instead of the selected implementation
- Improved AV1 encoding quality by changing bad default parameters


Flowframes 1.37.2 Changelog:
- Added RIFE 4.1 model (RIFE-CUDA)
- Scene detection now works with fractional factors (RIFE-NCNN)
- Frame deduplication now works with fractional factors (RIFE-NCNN)
- Pytorch AIs are no longer shown in the dropdown list if Pytorch is not installed


Flowframes 1.37.1 Changelog:
- Custom interpolation factors are no longer disabled by default (thx @ DrPleaseRespect)
- You can now input a target frame rate instead of a factor with RIFE-NCNN
- Flowframes now prevents Windows from going to sleep while it's running (needs testing)
- FFmpeg log is now shown again when waiting for encoding to finish
- Updated ffmpeg with SVT-AV1 1.0.0 (even faster AV1 export!)
- SVT-AV1 speed/presets have been offset by 1 (slowest to fastest is now 3-9 instead of 2-8)
- Default encoding preset is now "fast" instead of "medium"
- Interpolation factor will now also be validated in step-by-step mode, not just auto mode
- Fixed unreliable input video frame counting
- Fixed some encoding bugs for videos with weird frame rates


Flowframes 1.37.0 Changelog:
- Updated RIFE-NCNN, now supports any interpolation factor
- Fractional factors are now possible with NCNN (currently disables scene detection)
- Updated RIFE-NCNN models (added 2.3 and 4.0)
- Updated ffmpeg, AV1 encoding should now be even faster
- Frame counting should now be more reliable
- UI improvements (backported from Nmkoder)


Flowframes 1.36.0 Changelog:
- Added RIFE 4.0 model which is both faster and higher quality than 3.9
- Fixed issue with GIF exporting where auto-encode would try to mux a h264 stream into GIF
- Fixed custom output folder not working for batch processing
- Fixed issues where the batch queue could get stuck waiting for ffmpeg to finish
- Minor improvements by backporting Nmkoder code


Flowframes 1.35.0 Changelog:
- Updated RIFE - It now supports any interpolation factor (like 3x, 5x, 10x, ...)
- Added RIFE 3.9 model, big speed improvements (54% faster than 3.1, 100% faster than 2.3)
- WEBM/VP9 encoding is now signficantly more efficient by using 2-pass (but a bit slower)
- Updated ffmpeg to 4.4.1 (possible encoding/decoding speed improvements)
- Fixed some specific encoding parameters being set incorrectly (auto-tiling)


Flowframes 1.34.1 Changelog:
- Audio will now be slowed down accordingly when using slowmo
- Optimized encoding: VP9 and AV1 dynamic tiling, h264/h265/VP9/AV1 dynamic keyframe interval
- Fixed unreliable audio channel count detection for audio fallback
- Fixed temp folder deletion dialog interrupting batch processing


Flowframes 1.34.0 Changelog:
- Added resuming functionality (with auto-encode) - Drop a temp folder to resume it
- Resuming an already fully interpolated folder will instead just merge/mux all chunks
- Audio: MP3 tracks will no longer get re-encoded to AAC when using MP4 output
- Audio: Opus in MP4 muxing can now be enabled with config option allowOpusInMp4
- Audio Fallback: Improved AAC fallback quality
- Audio Fallback: Use same channel count as input file, no longer force Stereo
- Fixed problems with weird frame rates where Flowframes would claim it's invalid


Flowframes 1.33.0 Changelog:
- Added option to set quality when using image sequence output (fixes bad WEBP quality)
- Added server selection option (Automatic/EU/US) to allow manual selection of server
- The last selected output mode is now remembered after closing Flowframes
- Fixed issues with GPU list
- Minor UI improvements (uniform/updated PayPal, Discord icons)
- Minor UX improvements (clarify that image sequences require manual FPS input)


Flowframes 1.32.0 Changelog:
- Added slow-motion option to easily create slow-mo interpolations
- Auto-Encode now works with Image Sequence exports
- Option to create both limited and unlimited FPS outputs now works with image sequences
- Added automatic ping-based server picker to download models from US or EU server
- Added config option to enable custom interpolation factor (if compatible)
- ProRes now supports Alpha output when used with profile 4444 or 4444xq
- Fixed ProRes being locked to YUV420 colors on all presets
- Fixed WEBP image sequence export not working correctly


Flowframes 1.31.1 Changelog:
- Fixed problems that broke XVFI when using embedded python runtime
- Fixed model downloader not offering the option to download XVFI models


Flowframes 1.31.0 Changelog:
- Added XVFI interpolation AI, allows all interpolation factors from 2-10
- Unlocked 16x interpolation for RIFE-CUDA (quality will be worse than lower factors!)
- Unlocked 3x, 5x, 6x, 7x factors for DAIN-NCNN
- GPUs are now listed in the application's window title
- Batch: Notifications now only show on queue completion, not after each interpolation
- If interpolation fails, the last 10 lines of the corresponding log are shown
- Minor UI improvements
- Fixed custom output directory not being applied when using the batch queue
- Some internal refactoring and code cleanup


Flowframes 1.30.0 Changelog:
- Added the option to Close/Sleep/Hibernate/Shutdown after interpolation
- Added notifications that will show if an interpolation finished in the background
- Added option to change default output directory (same as input or custom)
- Added basic command line arguments to automate Flowframes interpolation
- Added move up/down buttons to batch processing queue
- Fixed a bug that would set incorrect default values for certain settings


Flowframes 1.29.0 Changelog:
- Added option to change the color space for most codecs (e.g. for 10-bit encoding)
- Added a button in the settings to reset the config to default
- Added a shortcut button to the Encoding Settings on the main tab
- Added option to disable Preview tab
- Fixed problem where videos with BT470M color format would fail to encode
- Removed compatibility warning when Flowframes is run on Windows 11


Flowframes 1.28.1 Changelog:
- Added option to auto-pause interpolation if auto-encode can't keep up
- Auto-Encode: Old frames now get deleted instead of being replaced by dummy files
- Fixed speed regression on RIFE 3.x (up to ~1.5x as fast as in 1.28.0)


Flowframes 1.28.0 Changelog:
- Updated RIFE CUDA to 3.8 (Skipped previous models)
- Added native transparency support with RIFE CUDA 3.8
- Added model downloader to pre-download all models at once
- Added support for loading custom AI models
- Improved model loading (faster file validation and lots of internal changes)
- Step-by-step mode: Previous steps now run automatically if needed


Flowframes 1.27.2 Changelog:
- Added Backup option - Will save an incomplete video after every chunk encode
- Custom FFmpeg args now also get applied for GIF encoding, not just videos
- Fixed low disk space warning popping up even if interpolation is already paused
- Fixed certain temp files not being deleted correctly


Flowframes 1.27.1 Changelog:
- Image sequence inputs can now be used without any importing/copying (if compatible)
- Added the FPS limit ("Maximum Output Framerate") to Quick Settings tab
- Batch Processing: AI Model name is now shown in queue
- Batch Processing: Fixed problems where GIFs/videos would play too slow
- Fixed problem where GIF/APNG interpolations would have unnecessary dupe frames
- Fixed problems with alpha interpolation when JPEG frames are enabled


Flowframes 1.27.0 Changelog:
- Updated RIFE-CUDA to v3 - Better quality and up to 25% faster!
- Updated RIFE-NCNN to v3
- Added new RIFE models 3.0 and 3.1 (both CUDA and NCNN)
- AV1 encoding is now up to 35% faster (updated to SVT-AV1 0.8.7)
- Renaming existing files to ".old" now also works for looped outputs and folders
- Added image sequence copy support for BMP and WEBP (no PNG conversion on import)
- Invalid (non-image) files now get ignored when loading an image sequence folder
- New model files hoster, should make model downloading more reliable
- Fixed the encoding log sometimes stopping, even if ffmpeg is still running
- Fixed PNG image sequence outputs getting re-encoded unnecessarily
- Fixed some problems regarding the FPS limit feature
- Fixed problem where toggling JPEG frames wouldn't get applied
- Internal data handling improvements (JSON for config files)


Flowframes 1.26.1 Changelog:
- Fixed image sequence copying not working correctly in some cases
- Fixed loop not working correctly when using JPEG frames
- Fixed batch queue not working correctly
- Fixed encoding failing for videos with bt470bg color space
- Fixed encoding failing when using lossless h265


Flowframes 1.26.0 Changelog:
- Image Sequence will now be copied instead of re-encoded if they are compatible
- Color space (and transfer function) will now be retained from the input video
- Aspect ratio of input video will now be preserved (for non-square pixel videos)
- Added "Debug Tools" window with a log viewer and a config editor tab
- Interpolation will now pause, not cancel, when disk space is low
- Existing output files now get renamed instead of overwritten if they already exist
- Several processing functions now run in the background, so the UI no longer freezes
- Fixed h265 CRF 0 not being lossless
- Fixed critical bug where Auto-Encode ran too early, causing FFmpeg errors
- Fixed bug where deduplicated videos would play too fast if JPEG frames are used
- Fixed some cases where the program wouldn't launch because of undhandled errors


Flowframes 1.25.1 Changelog:
- Added a button to pause/resume the interpolation
- Added AV1 codec support for MP4/MKV output
- Added support for JPG or WEBP image sequence output instead of only PNG
- NVENC encoding is now slightly more efficient (P7 Preset)
- Fixed accurate deduplication not working with JPEG frames


Flowframes 1.25.0 Changelog:
- Added support for FPS limiting (resampling) on Image Sequence outputs
- Added MagicYUV codec for AVI exports (super fast, better compression than huffyuv)
- Added option to import frames as JPEG for YUV420 content (almost lossless)
- Added [RES] (resolution) and [H] (height) variables for export filename pattern
- Added support for fractional input for "Maximum Output Frame Rate" option
- Encoding Speed option in Developer Options now also affects VP9
- Updated icon to a new colorful variant (same as FF Installer)
- Updated FFmpeg build (Release 4.4)
- Changed GIF encoding flags to fix potential transparency problems
- Better error messages for overclocking bugs and model load errors
- Fixed changing speed by adjusting input FPS not working correctly (allowCustomInputRate)
- Fixed encoding failing on non-NTFS drives


Flowframes 1.24.6 Changelog:
- GIF encoding framerate will now automatically be downsampled to 50 if it's higher
- Fixed CUDA AIs not actually starting on undetected GPUs or CPU
- Fixed muxing not working if metadata transfer was enabled but input has no metadata (MKV)
- Fixed audio/subtitle transfer sometimes breaking missing when using trim feature
- Fixed transparency interpolation not working if input resolution is not divisible by 2
- Fixed cursor skipping when entering fractions in the FPS input box


Flowframes 1.24.5 Changelog:
- Added option to choose dithering type for GIF encoding (Bayer/Floyd-Steinberg/None)
- Added option to preserve video metadata (for embedded fonts in MKV, etc)
- NVENC Improvements: CRF is more accurate, CRF 0 is now lossless
- Manual FPS input now accepts both floats ("23.97") and fractions ("24000/1001")
- Fixed manual FPS input (for image sequences etc) not working
- Several under-the-hood improvements and better error handling
- Fixed manual deduplication being ignored (Hotfix 1)


Flowframes 1.24.4 Changelog:
- Drastically improved ffmpeg encoding speed (+350% NVENC, +55% VP9/h264, +35% ProRes)
- Added compatibility checks showing warnings if not running Windows 10 x64
- Paths containing unicode chars now work with Pytorch implementations, removed warning
- Fixed scene detection somehow being broken once again


Flowframes 1.24.3 Changelog:
- Running CUDA AIs without a (detected) Nvidia GPU is now possible (for CPU or undetected cards)
- Fixed scene detection using wrong frames
- Fixed blended scene transitions not working with deduplication or VFR inputs
- Fixed manual deduplication and scene transition choices being ignored (step-by-step mode)
- FFprobe frame counting should now be multithreaded (-threads 0)


Flowframes 1.24.2 Changelog:
- Fixed problems with image sequence inputs (broken scene detection, etc)
- Added option to repeat the last frame, so the duration matches the input
- Framerates are now handled as fractions (24000/1001 instead of 23.97...) to avoid timing problems
- Fixed Variable Framerate (VFR) videos not being processed correctly
- Fixed DPI scaling breaking when opening a file browser dialog
- Other minor fixes and improvements

Known Issue: Blended scene transitions might not work correctly with deduplication or VFR inputs


Flowframes 1.24.1 (Hotfix) Changelog:
- Fixed a critical bug that would disable subtitle transfer, even when enabled and compatible


Flowframes 1.24.0 (Pre-Releases 1/2/3 + Final) Changelog:
- [PR1] Added experimental FLAVR integration
- [PR1] Fixed missing subtitle compatibility check when muxing audio/subs directly from input
- [PR1] Fixed bug that could break Image Sequence exports
- [PR1] Fixed flaw where certain online checks ran before the UI opens up instead of afterwards
- [PR2] Added optional blending for scene transitions (instead of duplicating previous frame)
- [PR2] Fixed a bug where the next-to-last frame can't get detected as a scene change
- [PR3] Added config option to allow custom input FPS (might break trimming and progress bars!)
- [PR3] Fixed scene change blending using the wrong frames
- [PR3] Fixed scene change blending not working in step-by-step mode
- [PR3] Fixed a problem in step-by-step mode that made manual deduplication impossible
- [PR3] Fixed a rare bug where the video resolution gets reported incorrectly
- [FNL] Fixed audio fallback not working for incompatible audio codecs
- [FNL] Added MP3 and WMA (v1/v2) lossless transfer (MKV output)


Flowframes 1.23.4 Changelog:
- Scene detection now also works for Image Sequence inputs
- Added a disk space check that will cancel the interpolation if disk space is too low
- Model name is included in output name pattern by default
- Fixed bug where framerate in output filename is incorrect when using FPS limiter
- Fixed FFmpeg MKV muxing bug (caused audio stuttering or complete lack of audio)
- Fixed models.txt files getting deleted if update fails
- Fixed log box flickering in certain situation


Flowframes 1.23.3 Changelog:
- Added option to customize the output filename pattern using placeholders/variables
- Audio/subtitle streams are now pulled from the input file, no longer requires extraction
- Trimming is now a lot faster for longer videos (like movies)
- Fixed UHD mode not working with RIFE-CUDA
- Fixed rare bug where videos with specific frame counts got cut off at the end
- Fixed scene detection not working for the last frame of a video
- Fixed problem that sometimes made the output video shorter, especially when trimming


Flowframes 1.23.2 Changelog:
- Added RIFE 2.4 model (CUDA & NCNN)
- Model list now gets loaded from GitHub, model updates won't require a Flowframes update
- Fixed manual FPS input for image sequence or videos where no framerate was detected
- Fixed a bug where encoding would sometimes fail
- Fixed a bug with trimming where the end of video would get cut off when it shouldn't


Flowframes 1.23.1 Changelog:
- Added RIFE 2.3 model (CUDA)
- Added AC3, DTS and PCM/Raw lossless audio transfer support for MOV output
- Fixed problem where muxing would fail if there's an incompatible subtitle stream
- Fixed "generating frame order information" optimizations being disabled
- Fixed scene detection running even if it's disabled (in Step-By-Step mode)
- Minor UX improvements to trimming tool
- Removed ability to change input FPS as it caused problems and was confusing to some


Flowframes 1.23.0 Changelog:
- Added "Fast Mode" for RTX cards, uses less VRAM and is at least 20% faster (RIFE-CUDA)
- Added support for keeping multiple audio tracks
- Added trimming feature, allow you to selected start and end time for your interpolation
- Added "Quick Settings" tab that allows you to change common settings faster
- Several UX improvements to Step-By-Step mode
- Peak output FPS is shown after interpolation (for benchmarking)
- Increased max color amount for GIF output to 256 (GIF's maximum) instead of 128
- Minor RIFE-CUDA optimization that should increase performance by an additional 3-5%
- Optimized "generating frame order information" process, making it ~60x faster
- Loading huge (>5gb) videos is now much faster and stable
- If NVENC is enabled, Flowframes will run an NVENC compatibility check before starting
- Flowframes now shows an error if someone tries to run it without unzipping (sigh...)
- Fixed bug where the output video got deleted if one stream (e.g. audio) failed to encode
- Fixed minor problems with transparent GIF encoding


Flowframes 1.22.3 Changelog:
- Added new RIFE-CUDA models: 2.1 and 2.2
- Added new RIFE-NCNN model: 2.0
- Improved interpolation progress bar and preview (updates faster, lower disk usage)
- Fixed looped exports having a double frame
- Fixed GIF encoding being locked even if framerate gets resampled to <=50 FPS
- Fixed extreme RAM usage (or even disk caching) for accurate deduplication
- Fixed bugs regarding alpha extraction


Flowframes 1.22.2 Changelog:
- Updated RIFE to 2.0 - Faster and better than ever!
- Added progress bar for loading metadata of very long videos
- Fixed freezing when loading video metadata
- Fixed scene detection running when extracting frames in Step-By-Step mode
- More error handling for Step-By-Step mode
- Fixed rare bug where Auto-Encoding would get stuck after cancelling an interpolation


Flowframes 1.22.1 Changelog:
- Added progress bars for ffmpeg actions (extraction/encoding)
- Added/fixed subtitle support for foreign languages/encodings
- Added experimental APNG alpha support
- Added video duration to "Current Input Info" label
- Added backup model download mirror (Los Angeles)
- Fixed batch processing not working at all
- Fixed multi-gigabyte input videos freezing the program or causing errors
- Fixed alpha not working correctly for image sequence inputs
- Fixed the AI selection saving the last selected AI not working reliably
- Fixed progress bar being stuck at 100% and not resetting in some cases
- Raised the hard framerate limit from 500 to 1000 FPS (out)
- Lots of under-the-hood work on resuming (not usable in 1.22.1 yet)


Flowframes 1.22.0 Changelog:
- Added experimental alpha (transparency) support for GIF and PNG inputs/outputs
- RIFE-NCNN now has all RIFE models from 1.5 to 1.8, same as RIFE CUDA
- The progress bar is now also being shown on the task bar icon
- The "Minimum Video Length" option is now only active if loop is enabled
- Fixed NVENC using default settings instead of the set CRF value
- Fixed frame extraction in step-by-step mode not clearing old frames
- In step-by-step mode, image sequences are now copied instead of moved on export
- Inputs resolutions not divisible by 2 now get padded instead of cropped


Flowframes 1.21.1 Changelog:
- Improved RIFE-CUDA image writing, making it run up to 20% faster
- Re-integrated DAIN-NCNN and reduced its VRAM usage
- The resolution, frame rate and frame count of your input is now shown in the GUI
- Fixed deduplicated loops not having the duplicate frames at the end removed
- Fixed scene detection using the wrong frame to replace the interpolation frames
- Fixed Auto-Encode making the GUI unresponsive when interpolating >10k frames
- Fixed last frame missing from exported video


Flowframes 1.21.0 Changelog:
- Added support for optionally using MKV instead of MP4 as h264/h265 container
- Added support for keeping subtitles (for MKV/MP4/WEBM/MOV outputs, MKV recommended)
- Added support for NVENC encoding on Nvidia GPUs
- Added AI model selection (RIFE 1.5 through 1.8)
- AI models are now dynamically loaded from the web, no longer included in the installation
- Downscaling ("Max Video Size") now also works for folder inputs
- Settings GUI Cleanup, removed some obsolete settings
- Fixed a bug where the progress bar for RIFE-NCNN wouldn't show
- Fixed scene detection not working if first few frames are dupes followed by a scene cut
- Fixed a problem where videos would get cut off if they end with a static image (dupes)
- Removed DAIN-NCNN as it was deemed obsolete, and the NCNN version had many limitations
- Other minor fixes and improvements

Known Issue:
Auto-Encode can get laggy if you're interpolating >5000 frames. Fixed in the next version.


Flowframes 1.20.4 Changelog:
- Fixed problem with Auto-Encode getting stuck at the end
- Fixed performance problems when using Auto Encode on long (>10k frames) videos
- Fixed freezes when trying to load certain video files (e.g. cut MTS files)
- Added warning that GIF is not suited for high-FPS videos


Flowframes 1.20.3 Changelog:
- Added option to clear the log box when selecting a new input file (on by default)
- Deduplication stats are now also shown when deduplicating during extraction
- Improved deduplication logging messages
- The encoding log is now being shown while waiting for Auto-Encode to fini
- Fixed outputs sometimes being too slow (especially with GIFs as input)
- Fixed scene detection not working
- Fixed minor problem where the AVI color space would be overwritten


Flowframes 1.20.2 Changelog:
- Added AVI encoding options: Codec (rawvideo/ffv1/huffyuv) and Color Space
- Added FPS limiting support to GIF exports
- Fixed FPS limiting not working correctly
- FPS limiting no longer re-encodes the video, which improves the quality
- Fixed video duration/timing when using scene detection with deduplication
- Fixed progress bar in step-by-step mode not showing up
- Fixed image sequence input not working at all
- AutoEncode "Delete Frames Once Encoded" option works again


Flowframes 1.20.1 Changelog:
- Updated RIFE-NCNN model to 1.7 (2D animation support)
- Updated RIFE-NCNN with UHD support
- Fixed GIF export not working correctly
- Fixed bug where Auto-Encode gets stuck and never finishes
- Fixed bug where old deduplication data was used even with deduplication disabled

KNOWN ISSUES:
- Scene detection will shorten the video - avoid it for now
- Exporting FPS-limited videos might not work correctly with formats other than MP4
- The AutoEncode mode "Delete Frames Once Encoded" is currently disabled as it's unstable


Flowframes 1.20.0 Changelog:
- Added export support for WEBM (VP9), MOV (ProRes) and uncompressed AVI
- Updated RIFE to 1.7 which works better for 2D animation
- Switched to a more stable approach for video timing, fixing desync issues
- Fixed looped videos having a duplicate frame between loops
- Embedded Python now supports RTX 3000 cards (Download "FF-1.20.0-RTX3000.7z")
- The python runtime will now automatically be compressed to save 1-2 GB of disk space

KNOWN ISSUES:
- Using "Remove during extraction" deduplication might slow down the output video a lot
- Exporting FPS-limited videos might not work correctly with formats other than MP4
- The AutoEncode mode "Delete Frames Once Encoded" is currently disabled as it's unstable
- RIFE-NCNN does not have the 1.7 model yet (Coming in 1.20.1)