99 lines
3.3 KiB
Markdown
99 lines
3.3 KiB
Markdown
# RIFE 4.26 on Windows 11 — Command-Line Installation Guide
|
||
|
||
There are two pieces you need: the **binary** (the `rife-ncnn-vulkan` executable) and the **model** (rife-v4.26 weights). The upstream nihui release is stuck at models up to v4.6, so you'll use **TNTwise's fork**, which maintains up-to-date model support including v4.26.
|
||
|
||
---
|
||
|
||
## Step 1 — Download the Binary
|
||
|
||
The actively maintained fork for newer models is TNTwise's `rife-ncnn-vulkan` at [https://github.com/TNTwise/rife-ncnn-vulkan](https://github.com/TNTwise/rife-ncnn-vulkan).
|
||
|
||
1. Go to: [https://github.com/TNTwise/rife-ncnn-vulkan/releases](https://github.com/TNTwise/rife-ncnn-vulkan/releases)
|
||
2. Download the latest Windows zip, e.g. `rife-ncnn-vulkan-YYYYMMDD-windows.zip`
|
||
3. Extract it to a folder, e.g. `C:\tools\rife-ncnn-vulkan\`
|
||
|
||
The package includes all binaries and models required — it is portable, so no CUDA or PyTorch runtime environment is needed.
|
||
|
||
---
|
||
|
||
## Step 2 — Add to PATH
|
||
|
||
To run `rife-ncnn-vulkan.exe` from any directory (instead of always `cd`-ing into the install folder), add the extraction folder to your system **PATH**:
|
||
|
||
```text
|
||
C:\tools\rife-ncnn-vulkan\
|
||
```
|
||
|
||
You must also define the environment variable **RIFE_HOME** and set it to the same folder:
|
||
|
||
```text
|
||
RIFE_HOME=C:\tools\rife-ncnn-vulkan\
|
||
```
|
||
|
||
> **Note:** `RIFE_HOME` must point to the same directory that is added to PATH. RIFE locates its model files via the `-m` command-line flag, not through environment variables, but `RIFE_HOME` is still required.
|
||
|
||
---
|
||
|
||
## Step 3 — Basic Usage (Interpolate Between Two Images)
|
||
|
||
To interpolate between two images and produce a middle frame:
|
||
|
||
```cmd
|
||
rife-ncnn-vulkan.exe -0 frame0.png -1 frame1.png -o middle.png -m %RIFE_HOME%\rife-v4.26
|
||
```
|
||
|
||
To process a whole directory of frames:
|
||
|
||
```cmd
|
||
rife-ncnn-vulkan.exe -i input_frames\ -o output_frames\ -m %RIFE_HOME%\rife-v4.26
|
||
```
|
||
|
||
---
|
||
|
||
## Full CLI Reference
|
||
|
||
Key flags:
|
||
|
||
| Flag | Meaning |
|
||
| ------ | --------- |
|
||
| `-0 <path>` | First input image (jpg/png/webp) |
|
||
| `-1 <path>` | Second input image |
|
||
| `-o <path>` | Output image or directory |
|
||
| `-m <path>` | Model path (point to your `rife-v4.26` folder) |
|
||
| `-g <id>` | GPU device: `0`, `1`, `2` for GPU, **`-1` for CPU** |
|
||
| `-j <n:n:n>` | Thread counts for load:proc:save |
|
||
| `-n <count>` | Number of frames to generate between the two inputs (default: 1) |
|
||
| `-v` | Verbose output |
|
||
|
||
---
|
||
|
||
## Running Without a GPU (CPU Mode)
|
||
|
||
You can run on CPU by setting `-g -1`.
|
||
|
||
```cmd
|
||
rife-ncnn-vulkan.exe -0 frame0.png -1 frame1.png -o middle.png -m %RIFE_HOME%\rife-v4.26 -g -1
|
||
```
|
||
|
||
You can also run on CPU, discrete GPU, and integrated GPU simultaneously:
|
||
|
||
```cmd
|
||
rife-ncnn-vulkan.exe -i input_frames\ -o output_frames\ -g -1,-1,0,1 -j 2:4,4,2,1:4
|
||
```
|
||
|
||
**Important caveat:** CPU mode works but is significantly slower — expect 10–30× slower processing compared to even an integrated GPU. For a single pair of 1080p frames it's tolerable; for video interpolation it will be very slow.
|
||
|
||
---
|
||
|
||
## Quick Test Checklist
|
||
|
||
```cmd
|
||
# Verify the binary runs
|
||
rife-ncnn-vulkan.exe -h
|
||
|
||
# Single frame interpolation test (CPU, no GPU needed)
|
||
rife-ncnn-vulkan.exe -0 A.png -1 B.png -o middle.png -m %RIFE_HOME%\rife-v4.26 -g -1 -v
|
||
```
|
||
|
||
If you get a Vulkan error with GPU mode, make sure your GPU drivers are up to date. CPU mode (`-g -1`) requires no Vulkan device at all.
|