Model Selection

We recommend starting with what you are trying to do, then selecting the best model from there.

Image

chevron-rightUpscale an Imagehashtag

Three tiers, each with a different tradeoff:

Gigapixel
Wonder
Bloom

Tier

Precision

Generative

Creative

Description

Precision Image Upscale & Enhancement

Generative Image Upscale & Enhancement

Creative Image Upscale & Enhancement

Intent

Improve resolution, preserve source characteristics.

Improve and restore by adding detail and texture, while preserving intent.

Transform with new, creative detail or stylization.

Use Case

General purpose upscaling & enhancement.

Upcaling, restoring and improving low resolution or compressed sources.

Upscaling, improving, and transforming GenAI sources.

Gigapixel

Model
Use Case

Default — works for most images

Maximum source preservation

Very small inputs (thumbnails, web scrapes)

Illustrations, renders, game assets, digital art

Documents, screenshots, text-heavy images

Portraits where facial detail is the priority

PNGs with alpha channels

Wonder

Model
Use Case

Default generative upscaling

Stronger generation for more detail

Generative enhancement grounded in photorealism

Generative enhancement with more creative freedom

Heavily degraded or damaged sources

Bloom

Model
Use case

Maximum stylization

Creative but more grounded

chevron-rightSharpen or Deblurhashtag

Standard — use when you know the blur type:

Model
Use case

General-purpose — good starting point

Noticeably soft images

Subtle sharpening that avoids over-processing

Optical blur from lens softness or missed focus

Blur from camera or subject movement

Slightly out-of-focus images

Optimized for faces and skin texture

Optimized for fur, feathers, scales

Generative — use when the source is heavily degraded or you're unsure of the blur type:

Model
Use case

Generative sharpening that recovers detail beyond the original

chevron-rightReduce Noise or Grainhashtag

Pick based on noise severity:

Model
Use case

Light noise — well-lit conditions

Moderate noise — high ISO, low light

Heavy noise — very high ISO, underexposed

Film scans or archival images with physical artifacts

chevron-rightRemove Objects or Backgroundshashtag
Model
Use case

Remove the background entirely

Extract subject with transparency for compositing

chevron-rightCorrect Color or Lightinghashtag
Model
Use case

Auto white balance and color correction

Exposure and contrast adjustment

Convert black-and-white to color

Video

chevron-rightUpscale or Enhance a Videohashtag

Three families, each with a different tradeoff:

Proteus
Starlight
Astra

Tier

Precision

Generative

Creative

Description

Precision Video Upscale & Enhancement

Generative Video Upscale & Enhancement

Creative Video Upscale & Enhancement

Intent

Improve resolution, preserve source characteristics.

Improve and restore by adding detail and texture, while preserving intent.

Transform with new, creative detail or stylization.

Use Case

General purpose upscaling & enhancement.

Upcaling, restoring and improving archival and GenAI

Upscaling, improving, and transforming GenAI sources.

Proteus

Don't know where to start? Use Proteus. Then narrow down if your source has specific characteristics:

Model
Use case

Default — handles most content well

Lighter processing, stays closest to the original

Match to your source quality — HQ for clean footage, LQ for degraded

Visible aliasing or moire patterns

Halo artifacts from older upscaling or edge enhancement

Interlaced broadcast (TV) or camcorder (DV) footage

Interlaced content with heavy degradation

Interlaced content with halo artifacts

Animation and single-frame animation

High-quality or general CG and rendered content

Legacy content at medium or low quality

General enhancement with natural texture

Push detail enhancement — more aggressive

Preserve original look — lighter touch

Starlight

Quality and Fast subtiers. Use Quality when output matters most, Fast when speed or cost is the priority.

Model
Subtier
Use case

Quality

Maximum quality generative output

Quality

Default for most generative work

Quality

Highest fidelity to source

Quality

Emphasis on sharpness and edge detail

Fast

Best fast model — start here for speed

Fast

Speed-optimized generative enhancement

Fast

Lightest — previews or cost-sensitive jobs

Astra

Model
Use case

Creative video transformation

chevron-rightReduce Noise or Grainhashtag

Denoise

Quality and Fast subtiers via the Nyx family:

Model
Subtier
Use case

Quality

Default — balanced noise reduction and detail

Quality

Stronger reduction for heavy noise

Quality

Maximum detail preservation, lighter reduction

Fast

Speed-optimized denoising

chevron-rightChange Frame Rate or Create Slow-Mohashtag

Frame Interpolation

Quality and Fast subtiers. Generates new frames to increase frame rate or create slow motion.

Model
Subtier
Use case

Quality

Default — best all-around quality

Quality

Alternative algorithm for fast action or complex scenes

Quality

Additional option for specific content types

Apollo Fast

Fast

Speed-optimized

Chronos Fast

Fast

Speed-optimized alternative algorithm

chevron-rightDeblur or Colorizehashtag

Video Utilities

Model
Use case

Remove motion blur from camera shake or fast movement

Convert B&W or desaturated footage to color

Last updated