Popular Models
Check out our top-performing models
The fastest way to great results. These are the models developers reach for first. Each one handles a wide range of sources on its own. Pick the one that fits your goal and you're one request away from production-quality output.
Video Models
Image Models
Video Models
Starlight Precise 2.6
Our highest-quality video model. Starlight Precise 2.6 uses diffusion to rebuild footage frame by frame, restoring realistic faces, skin, texture, and text that other models can't recover. It's the model to reach for when quality matters more than anything else.
Best-in-class restoration for GenAI, archival, and heavily degraded video
Natural faces and skin tones with stable, consistent motion
Clearer text and finer texture detail
Start here if you need the best possible result.
Starlight Fast 3
Generative quality at production speed. Starlight Fast 3 brings diffusion upscaling to high-volume pipelines. It's our fastest generative video model, up to about 4× faster than Starlight Precise 2.6, so you can deliver Starlight-class results on tight timelines and large libraries.
Our fastest diffusion upscaler
Built for batch processing and quick turnaround
A big step up in detail from traditional upscaling
Start here if you want generative results at scale.
Hyperion 2.5
Turn SDR into stunning HDR. Hyperion 2.5 converts standard dynamic range footage into true HDR, expanding luminance range, bit depth, and contrast so your content shines on modern HDR displays and gives colorists more room to work.
Real HDR from existing SDR libraries
Professional output formats for grading and VFX workflows
Breathes new life into older content for HDR platforms
Start here if you're delivering HDR or need more latitude in the grade.
Proteus
The trusted workhorse. Proteus upscales, denoises, and sharpens video in a single pass, and it's the go-to starting point for video on the API. It's fast, predictable, and fully controllable: let it tune itself automatically, or dial in every setting for your pipeline.
Reliable results on almost any camera footage
Automatic parameter estimation, with manual control when you want it
Combines upscaling, noise reduction, and sharpening in one filter
Start here if you're not sure which video model to use.
Image Models
Wonder 4
One pass. Every fix. Wonder 4 upscales your image while it removes noise, recovers blur, and repairs compression, deciding on its own what each part of the image needs. Instead of chaining three or four models, you send one request and get back a sharper, cleaner, larger image with natural, believable detail.
Finer detail and more realism than any Wonder model before it
Biggest gains on low-resolution, compressed, and heavily degraded photos
Crisper text, charts, and graphics
Strong across portraits, landscapes, wildlife, and complex textures
Start here if you're not sure which image model to use.
Bloom 2
Creative detail for GenAI images. Bloom 2 is our creative upscaler for native GenAI content: AI artwork, illustrations, and CG renders. It enhances quality while introducing new detail, texture, and visual elements, with the range to subtly refine the source or substantially reimagine it.
Adds rich new detail and texture to AI artwork and illustrations
Improves skin and facial detail in GenAI portraits
Upscales CG renders with added realism
Preserves source colors while adding detail
Start here if you want creative upscaling for GenAI or stylized images.
Denoise Max
Clean images from impossible light. Denoise Max is our most advanced denoising model, built on the same generative foundation as Wonder. It strips out heavy noise and grain, sharpens the areas that went soft, and leaves already-sharp detail alone, with far fewer artifacts than traditional noise reduction.
Rescues high-ISO, low-light, and underexposed shots
Sharpens soft regions without over-processing sharp ones
Natural texture instead of the smeared, plastic look of older denoisers
Start here if noise is the main problem with your images.
Standard 2
Bigger, and exactly the same. Standard 2 is our precision upscaler: it increases resolution and cleans up the image without inventing a single detail that wasn't in the source. What you send is what you get back, just larger and clearer.
Faithful to the source, pixel for pixel
Ideal for product photography, print reproduction, and archival masters
Clean results on graphics and imagery that will be compared side by side with the original
Start here if the output has to stay true to the input.
Keep Learning
Check out some of our other helpful resources.
Last updated
Was this helpful?