Lots of new parameters and restructuring

av1_presets.yml

@@ -1,24 +1,40 @@
 "7.1":
   alayout: "7.1"
   ba: "450k"
-  cpu: "2"
+  cpu: "3"
   crf: "24"
   mf: "1"
+  dnl: "5"
+  arnr_strength: "1"
 "5.1":
   alayout: "5.1"
   ba: "256k"
-  cpu: "2"
+  cpu: "3"
   crf: "24"
   mf: "1"
+  dnl: "5"
+  arnr_strength: "1"
 stereo:
   alayout: "stereo"
   ba: "128k"
-  cpu: "2"
+  cpu: "3"
+  crf: "24"
+  mf: "1"
+  dnl: "5"
+  arnr_strength: "1"
+anime:
+  alayout: "stereo"
+  ba: "128k"
+  cpu: "3"
   crf: "25"
   mf: "1"
+  dnl: "-1"
+  arnr_strength: "0"
 lq_mono:
   alayout: "mono"
   ba: "64k"
   cpu: "2"
   crf: "42"
   mf: "1"
+  dnl: "10"
+  arnr_strength: "1"

hav1w.py

@@ -1,4 +1,6 @@
 import argparse
+import os
+import pathlib
 import subprocess
 
 import yaml
@@ -6,6 +8,9 @@ import yaml
 # Hannes' AV1 wrapper script
 # Version 2.0.0
 
+# This is a good guide for optimizing AV1 parameters with the AOM reference encoder:
+# https://www.reddit.com/r/AV1/comments/t59j32/encoder_tuning_part_4_a_2nd_generation_guide_to/
+
 parser = argparse.ArgumentParser(prog="hav1w")
 parser.add_argument("input")
 parser.add_argument("output")
@@ -15,12 +20,15 @@ parser.add_argument("--ba")
 parser.add_argument("--alayout")
 parser.add_argument("--cpu")
 parser.add_argument("--mf")
+parser.add_argument("--dnl")
+parser.add_argument("--arnr-strength")
 parser.add_argument("--crop")
-parser.add_argument("--no10bit")
+parser.add_argument("--no10bit", action="store_true")
 parser.add_argument("--dry", action="store_true")
 args = parser.parse_args()
 
-with open("av1_presets.yml", "r") as stream:
+preset_path = os.path.join(pathlib.Path(__file__).parent.resolve(), "av1_presets.yml")
+with open(preset_path, "r") as stream:
     presets = yaml.safe_load(stream)
 preset = presets[args.preset]
 
@@ -34,15 +42,61 @@ if args.cpu is not None:
     preset["cpu"] = args.cpu
 if args.mf is not None:
     preset["mf"] = args.mf
+if args.dnl is not None:
+    preset["dnl"] = args.dnl
+if args.arnr_strength is not None:
+    preset["arnr_strength"] = args.arnr_strength
+
+cmd = ["ffmpeg", "-i", args.input, "-c:v", "libaom-av1", "-c:a", "libopus", "-c:s", "copy", "-map", "0"]
+
+# AV1 Settings
+
+# Constant rate factor, determines target quality
+cmd += ["-crf", preset["crf"], "-b:v", "0"]
+
+# Set minimum keyframe interval to 12 to prevent placing too many keyframes... just in case
+cmd += ["-keyint_min", "12"]
+
+# Set lookahead to 48 frames
+# Reddit post says it's worth it...
+cmd += ["-lag-in-frames", "48"]
+
+# Temporal filtering parameters
+# Set arnr_strength to 0 for animation and low variance stuff, 1 for content with more variance and fast motion scenes
+cmd += ["-arnr-strength", preset["arnr_strength"], "-arnr-max-frames", "3"]
+
+# Disable deltaq-mode for higher fidelity
+aom_params = ["deltaq-mode=0"]
+
+# Enable quantization matrices
+# The Reddit post told me to do so, no matter what :P
+aom_params += ["enable-qm=1"]
+
+# Grain synthesis
+# Set to -1 for disabling grain synthesis: makes sense for animation content (animes, ...) as
+# a) this is not a strength of the aom encoder
+# b) animation content usually does not have a lot of noise
+#
+# If you need to do grain synthesis on animation content, maybe SVT-AV1 is a better choice?
+# And read this: https://www.reddit.com/r/AV1/comments/n4si96/encoder_tuning_part_3_av1_grain_synthesis_how_it/
+if preset["dnl"] != "-1":
+    aom_params += ["enable-dnl-denoising=0", "denoise-noise-level=" + preset["dnl"]]
+
+cmd += ["-aom-params", ":".join(aom_params)]
+
+if not args.no10bit:  # Encode to 10 bit per default
+    cmd += ["-pix_fmt", "yuv420p10le"]
 
-cmd = ["ffmpeg", "-i", args.input, "-c:v", "libaom-av1", "-c:a", "libopus", "-c:s", "copy", "-map", "0",
-       "-crf", preset["crf"], "-b:v", "0", "-keyint_min", "12"]
 if args.crop is not None:
     cmd += ["-vf", "crop=" + args.crop]
-if args.no10bit is None:  # Encode to 10 bit per default
-    cmd += ["-pix_fmt", "yuv420p10le"]
-cmd += ["-b:a", preset["ba"], "-af", f"aformat=channel_layouts={preset['alayout']}", "-mapping_family", preset["mf"],
-        "-cpu-used", preset["cpu"], "-row-mt", "1", "-tiles", "2x2", args.output]
+
+# Encoding efficiency settings (tiles also affect decoding performance)
+cmd += ["-cpu-used", preset["cpu"], "-row-mt", "1", "-tiles", "2x2"]
+
+# Audio/Opus settings
+cmd += ["-b:a", preset["ba"], "-af", f"aformat=channel_layouts={preset['alayout']}", "-mapping_family", preset["mf"]]
+
+cmd += [args.output]
 
 if args.dry:
     print(" ".join(cmd))