SVT-AV1 Parameters for Sharper Video Output

This article outlines the specific internal parameters within the SVT-AV1 encoder (libsvtav1) that directly influence and control the sharpness of the final video output. We will examine the core settings—including the loop filter sharpness, quantization matrices, and film grain synthesis—explaining how to configure them to prevent blurriness and preserve high-frequency details in your encodes.

The Sharpness Parameter (sharpness)

The most direct way to control image sharpness in SVT-AV1 is through the internal sharpness parameter (configured via -svtav1-params sharpness=X in FFmpeg). * Range: -2 to 7 (with 0 as the default). * Function: This parameter directly adjusts the loop filter’s sharpness metric. Higher positive values restrict the deblocking filter from smoothing out edge transitions, resulting in a noticeably sharper image. Negative values allow the loop filter to smooth out more edges, reducing blockiness at the expense of sharpness.

Deblocking Loop Filter (enable-dlf)

AV1 uses a Deblocking Loop Filter (DLF) to smooth out block artifacts caused by compression. * Function: While the DLF improves objective metrics, it can introduce unwanted blur. If preserving maximum sharpness is your primary goal, you can disable this filter entirely using enable-dlf=0 (or disable-dlf=1 depending on your SVT-AV1 version). * Caveat: Disabling the loop filter altogether may introduce visible block artifacts, especially at lower bitrates.

Quantization Matrices (enable-qm)

Quantization matrices control how distortion is distributed across different frequency bands. * Function: By enabling quantization matrices using enable-qm=1, the encoder allocates bits more aggressively to high-frequency details (which represent sharp edges and fine textures) rather than flat surfaces. * Adjustment: You can fine-tune this with qm-min and qm-max (ranging from 0 to 15) to set the minimum and maximum flatness allowed for the quantization matrices, preventing the encoder from over-smoothing complex scenes.

Film Grain Synthesis (film-grain)

At lower to medium bitrates, AV1 encoders naturally discard high-frequency noise and fine textures to save space, resulting in a “washed out” or overly smooth look. * Function: Enabling film grain synthesis (e.g., film-grain=8 or higher) instructs the encoder to analyze the original video’s grain, discard it during compression, and then synthetically regenerate it during playback. * Result: While this does not increase actual resolution sharpness, the simulated grain restores perceived texture and sharpness, preventing the video from looking artificially soft.

Visual Quality Tuning (tune)

The mathematical algorithm the encoder uses to evaluate video quality heavily impacts sharpness. * Function: Setting tune=0 (Visual Quality / VQ) optimizes the encoder for subjective human vision. This tuning path prioritizes the preservation of sharp edges and fine details. * Alternative: Avoid tune=1 (PSNR) or tune=2 (VMAF) if sharpness is your priority, as these modes often smooth out textures to achieve better mathematical compression scores.