github
/
flowframes
mirror of https://github.com/n00mkrad/flowframes
1
Fork 0
Code Releases Activity

Update README.md

This commit is contained in:
N00MKRAD 2021-05-03 15:13:54 +02:00
parent 9dc732dcf6
commit 5940011e3a
1 changed files with 19 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

43
README.md
View File

@ -1,6 +1,6 @@
# Flowframes - Windows GUI for Video Interpolation
## [itch.io (Free Builds)](https://nmkd.itch.io/flowframes) | [Patreon (Latest Builds)](https://www.patreon.com/n00mkrad) | [Discord](https://discord.com/invite/eJHD2NSJRe)
## [itch.io (Free Old Builds)](https://nmkd.itch.io/flowframes) | [Patreon (Latest Builds)](https://www.patreon.com/n00mkrad) | [Discord](https://discord.com/invite/eJHD2NSJRe)
Flowframes Windows GUI for video interpolation - Supports RIFE (Pytorch & NCNN), DAIN (NCNN), and FLAVR (Pytorch) implementations.
@ -13,8 +13,8 @@ However, **I do not provide support for self-built versions** as I can't guarant
## Quick Installation
* Download on [itch](https://nmkd.itch.io/flowframes) or, for the most recent beta versions, on [Patreon](https://www.patreon.com/n00mkrad). This repo does not provide builds.
* Extract the 7z archive (7-zip or similar software required!) to any folder
* Run Flowframes.exe
* Follow the instructions in the installer and wait for it to complete
* Run Flowframes
@ -27,47 +27,43 @@ However, the official RIFE implementation run best via its original Pytorch impl
The requirements to run these are the following:
* A **modern Nvidia GPU** (750 Ti, 900/1000/1600/2000/3000 Series).
* A **Python** installation including Pytorch (1.5 or later) as well as the packages `opencv-python` and `imageio`.
* The python runtime with pytorch is included in the Flowframes "Full" and "Full-RTX3000" downloads.
* A **Python** installation including Pytorch (1.5 or later) as well as the packages `opencv-python`, `sk-video`, `imageio`.
* The Flowframes Installer will automatically download all dependencies by default if these requirements are not fullfilled.
[More Details On Python Dependencies](PythonDependencies.md)
## Running A Pytorch Implementation on Nvidia Ampere GPUs
The python runtime included in the "Full-RTX3000" package includes support for Ampere GPUs. If you want to use your own installation, [use Pytorch 1.7.1](https://pytorch.org/get-started/locally/) or newer on Python 3.8.x.
Important: Ampere GPUs perform worse than they should on cuDNN 8.04 and older. If your cuDNN version is not >=8.05, you can manually update it by downloading it from Nvidia and replacing the DLLs in the torch folder. The embedded runtime already includes those files.
## Configuration
All Settings have reasonable defaults, so users do not need to do any configuration before using the program.
Here is an explanation of some of the more important settings.
### General
### Application
* Processing Style: Either run all steps at once, or each step manually, in case you want to edit frames, or deduplicate manually.
* Maximum Video Size: Frames are exported at this resolution if the video is larger. Lower resolutions speed up interpolation a lot.
* Export Name Pattern: Customize the pattern of the filenames of outputs using variables.
### Interpolation
* Copy Audio: Audio will be saved to a separate file when extracting the frames and will afterwards be merged into the output.
* Not guaranteed to work with all audio codecs. Supported are: M4A/AAC, Vorbis, Opus, MP2, PCM/Raw.
* Remove Duplicate Frames: This is meant for 2D animation. Removing duplicates makes a smooth interpolation possible.
* You can disable this completely if you only use content without duplicates (e.g. camera footage, CG renders).
* Animation Loop: This will make looped animations interpolate to a perfect loop by copying the first frame to the end of the frames.
* Don't Interpolate Scene Changes: This avoids interpolating scene changes (cuts) as this would produce weird a morphing effect.
* Input Media To Preserve: Toggle transfer of Audio, Subtitles and MKV Metadata.
* Enable Transparency: Interpolate transparency. Only active if the input **and** output support transparency (PNG/GIF).
* Import HQ JPEGs: Will extract JPEG instead of PNG frames from videos. Fast and lightweight, but with a tiny (invisible) quality loss.
* Frame De-Duplication: This is meant for 2D animation. Removing duplicates makes a smooth interpolation possible.
* You should disable this completely if you only use content without duplicates (e.g. camera footage, CG renders).
* "During Extraction" works for most content. Use "Accurate (After Extraction)" for fine-tuning the sensitivity.
* Loop Interpolation: This will make looped animations interpolate to a perfect loop by interpolating back to the first frame at the end.
* Fix Scene Changes: This avoids interpolating scene changes (cuts) as this would produce weird a morphing effect.
* Auto-Encode: Encode video while interpolating. Optionally delete the already encoded frames to minimize disk space usage.
* Save Output Frames As JPEG: Save interpolated frames as JPEG before encoding. Not recommended unless you have little disk space.
### AI Specific Settings
* RIFE - UHD Mode - This mode changes some scaling parameters and should improve results on high-resolution video.
* GPU IDs: `0` is the default for setups with one dedicated GPU. Four dedicated GPUs would mean `0,1,2,3` for example.
* NCNN Processing Threads: Increasing this number to 2, 3 or 4 can improve GPU utilization, but also slow things down.
* RIFE CUDA Fast Mode: Utilizes Half-Precision (fp16) to speed things up and reduce VRAM usage, but can be unstable.
### Video Export
@ -78,7 +74,6 @@ Here is an explanation of some of the more important settings.
### Debugging / Experimental
* Show Hidden CMD Windows: This will show the windows for AI processes. Can be useful for debugging.
* FFprobe: Count Frames Manually: This uses a slower way of getting the input video's total frame count, but works more reliably.
@ -92,7 +87,7 @@ Here is an explanation of some of the more important settings.
* Modern CUDA-capable GPU (Nvidia Maxwell or newer) with 6 GB VRAM or more
* 16 GB RAM
* Modern CPU (Intel Core 7xxx Series or newer, AMD Ryzen Series)
* Modern CPU (Intel Core 7000 Series or newer, AMD Ryzen 1000 Series or newer)
@ -114,4 +109,4 @@ Here is an explanation of some of the more important settings.
**A:** It encodes your output video during interpolation, instead of afterwards. Enable it unless you have a very weak CPU.
**Q:** I downloaded a "Full" package but now want to switch to my own system Python installation. How do I do that?
**A:** Go to `FlowframesData/pkgs/` and delete the folders `py-tu` or `py-amp`, whichever you have. Flowframes will now try to use system python.
**A:** Go to `FlowframesData/pkgs/` and delete the folders `py-tu` or `py-amp`, whichever you have. Flowframes will now try to use system python.