forked from n00mkrad/flowframes
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathchangelog.txt
203 lines (165 loc) · 10.1 KB
/
changelog.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
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)
Flowframes 1.19.1 Changelog:
- Fixed the UHD threshold option using the original (instead of scaled) resolution
- Fixed Accurate De-Duplicaiton not working properly with Scene Detection
- Improved Scene Detection to avoid issues with a very low threshold
- Size of the image buffer for de-duplication is now dynamic based on the video resolution
- The audio stream no longer gets re-encoded when converting the frame rate of a video
Flowframes 1.19.0 Changelog:
- Added RIFE's UHD Mode which works better on high-resolution content (CUDA Only)
- Updated RIFE model to 1.6
- Removed the video format whitelist, enabling support for Apple ProRes, 3GP, APNG, and more
- Scene Detection now works for "Image Sequence" output mode
- Image Sequence output folders now have a better name with a suffix, like the output videos
- Fixed incorrect progress/FPS when running a chained interpolation (like 4x/8x RIFE-NCNN)
- Fixed an encoding problem where some videos were too short or had duplicate frames
- Fixed Auto-Encode mode with NCNN producing short or broken videos
- Improved updater to show latest versions for both Patreon and Itch.io
Flowframes 1.18.2 Changelog:
- Fixed exporting with both loop and FPS limit not working correctly
- Fixed loop mode not working with JPEG mode
Flowframes 1.18.1 Changelog:
- Fixed last frame not being included in output video (Thanks @ChristianUndCo)
- Reduced RIFE's buffer size which should decrease RAM usage a bit
- Limited deduplication buffer to 750 frames to avoid freezes and extreme RAM usage
- The highest preset for video size is now 4320p (8K), but you can still enter custom values
- Fixed JPEG interpolation output mode
- Fixed bug where files with uppercase extensions (like "video.MP4") were rejected
- Fixed multi-GPU not working with NCNN AIs
- Better error handling for Vulkan and Python errors
KNOWN ISSUE: Scene detection might not work with de-duplication, a fix is in the works
Flowframes 1.18.0 Changelog:
- Added "Auto-Encoding" which will encode your video while interpolating (instead of afterwards)
- Updated RIFE code which greatly reduces VRAM usage - Makes 4K on 8 GB VRAM possible
- Updated the RIFE model to 1.5, resulting in a great quality boost
- RIFE-NCNN has been updated and should now be about 5x as fast as before
- GIF output now has correct timings again (handled by ffmpeg)
- Dropping multiple files into the main view directly loads them into Batch Processing
- When using Step-By-Step mode, you can now re-use a video's existing temp folder
- Step-By-Step no longer generates 2x/4x/8x timecodes, now only for the selected factor
- Several fixes regarding Step-By-Step mode and Batch Processing
- Fixed the Video Conversion util's audio bitrate being 1000x too low
- CAIN-NCNN has been removed as it is too buggy, please use RIFE-NCNN instead
Flowframes v17 [RC1] Changelog:
- Added Scene Detection - Avoids dirty interpolations of cuts
- Added "Step-By-Step Mode", where you can run each processing step separately
- Added 4x/8x support for RIFE-NCNN
- Added support for Opus audio and WMV video
- Fixed support for using a frame folder as input
- Folder input now accepts any image format that's compatible with FFmpeg (JPG, etc)
- Folder input now also works for non-consecutive filenames instead of a fixed format
- Fixed image sequence output mode
- The accurate de-duplication method is now up to 2x as fast
- Looping the output video is now massively faster
Flowframes v16 Changelog:
- Added new timecode-based deduplication and encoding (can be configured in the Settings)
- Added RIFE-NCNN, slower than the CUDA version but runs on AMD GPUs
- Added support for Raw/PCM audio as well as MP2
- Added new "landing page" with changelogs and Patreon supporters
- Added new experimental options (NCNN threads, ffmpeg threads, ffmpeg preset)
- Improved RIFE-CUDA error handling
- Fixed freezes when loading a video that has a lot of metadata
- Fixed JPEG frame mode not working