For the complete documentation index, see llms.txt. This page is also available as Markdown.

Starlight Precise 2.6

High-quality upscaling and enhancement for GenAI and archival video with realistic faces, textures, text clarity, and temporal consistency

Model Overview

Starlight Precise 2.6 is a diffusion video enhancement model designed to upscale footage up to 4K while improving realism and detail across both AI-generated and archival video. Building on Precise 2.5, it enhances faces, fabrics, materials, and textures while improving the clarity of text, labels, logos, and other fine details.

Starlight Precise 2.6 delivers up to 15% faster rendering across both cloud and local processing, with improved temporal consistency that reduces periodic flickering and produces more stable results across longer sequences and multiple clips. A new Sharpen parameter gives the ability to dial back over-sharp details for a smoother, softer look.

Key Capabilities

  • Upscales and enhances GenAI and archival video up to 4K.

  • Improved faces, materials, textures, text, labels, and fine detail.

  • Up to 15% faster rendering across both cloud and local processing.

  • Improved temporal consistency reduces flickering and provides greater stability across longer sequences and multiple clips.

  • New Sharpen parameter controls the balance between softer, natural look and sharper generative enhancement.

Pricing

Frames Per Credit — 1080p
Frames Per Credit — 4k

26.0417

11.9179

For more accurate estimates we recommend checking out our Credit Calculator.

Endpoint

Parameters

model

Model name:


inputFrameCount

Input video number of frames.


inputFrameRate

Input video frame rate in frames per second.


inputHeight

Input video height in pixels.


inputWidth

Input vidoe width in pixels.


outputHeight

Output video height in pixels.


outputWidth

Output video width in pixels.


videoBitDepth

Output video bit depth.


videoCodec

Output video codec.

Possible values: , ,


videoProfile

Output video profile.

Possible values: , ,


watermark

Whether to watermark the output.


sharpness

Controls how much sharpening the model applies to the output.

Range: to

Examples

Last updated