aa71eb96d7
Plugins: - misc_fixes v2.8: Pre-processing, container remux, stream conforming - stream_organizer v4.8: English priority, subtitle extraction, SRT conversion - combined_audio_standardizer v1.13: AAC/Opus encoding, downmix creation - av1_svt_converter v2.22: AV1 video encoding via SVT-AV1 Structure: - Local/ - Plugin .js files (mount in Tdarr) - agent_notes/ - Development documentation - Latest-Reports/ - Error logs for analysis
31 lines
2.5 KiB
Markdown
31 lines
2.5 KiB
Markdown
# SVT-AV1 Optimization Guide
|
|
|
|
**Target Goal**: "Relatively fast encoding, high visual quality, ignore file size."
|
|
|
|
Based on analysis of the plugin constraints and current SVT-AV1 (2024/v2.x+) best practices, here is the recommended configuration.
|
|
|
|
## Recommended Settings
|
|
|
|
| Setting | Recommended Value | Default | Reasoning |
|
|
| :--- | :--- | :--- | :--- |
|
|
| **Preset** | **6** | `10` | Preset 6 is the modern "sweet spot" for efficiency/speed. It is significantly better quality than 10, and much faster than the slow presets (0-4). If 6 is too slow for your hardware, use **8**. |
|
|
| **CRF** | **24** | `29` | Lower CRF = Higher Quality. CRF 24 is excellent for visual fidelity (often VMAF > 95). Since file size is not a concern, we can drop from the default 29 to ensure no artifacts. |
|
|
| **Input Depth** | **10** | `8` | **Crucial Change**. 10-bit encoding prevents color banding and improves visual fidelity with **negligible speed penalty** on SVT-AV1. Always use this for quality. |
|
|
| **Tune** | **0** | `0` | Keep at 0 (Visual Quality) to prioritize human perception over metrics. |
|
|
| **Film Grain** | **0** (or 5-10) | `0` | Keep at 0 for clean sources. If your source is older/grainy, set to **5-10** to synthesize grain instead of struggling to compress it (which looks blocky). |
|
|
| **AQ Mode** | **2** | `2` | DeltaQ mode is best for perceptual quality. |
|
|
| **Max Resolution**| **As Source** | `none` | Don't downscale unless necessary for compatibility. |
|
|
|
|
## Why these settings?
|
|
|
|
1. **Preset 6 vs 10**: The default Preset 10 is designed for *real-time* applications. It sacrifices a lot of efficiency. Preset 6 optimizes block partitioning much better, resulting in a cleaner image at the same bitrate, or simply a better looking output.
|
|
2. **CRF 24 vs 29**: Default 29 is a "safe" bet for small files. 24 moves firmly into "High Quality/Archival" territory without being wasteful (like CRF 18 might be).
|
|
3. **10-bit**: Modern encoders handle 10-bit very efficiently. Even if your source is 8-bit, encoding in 10-bit avoids internal rounding errors that cause banding in gradients (sky, dark walls).
|
|
|
|
## Summary for Tdarr Plugin Inputs
|
|
* `crf`: **26** (or **24** via manual input if possible, otherwise pick closest lower option like 26)
|
|
* *Note: Plugin dropdown options are: 20, 26, 28...* -> **Select 26**.
|
|
* `preset`: **6**
|
|
* `input_depth`: **10**
|
|
* `fast_decode`: **0** (Optional: Turn off fast decode optimization for slightly better compression efficiency, though enable is fine if playback on low-end devices is a concern).
|