Correction, 2026-10-08: this page stated as fact that AI upscalers keep edges sharp and faces from going waxy, and that a Topaz upscale "goes up cleanly"; we have not measured that, and an upscaler that re-draws detail can change faces, so those lines are now things to check. It grouped Magnific with the diffusion upscalers that re-draw detail, but Magnific also has a Precision mode built to stay faithful; the page now names its Creative mode. It did not say that Topaz on 8frame always enlarges 4x and bills per started 24-megapixel block of the output, up to 281 credits, or that no video model on 8frame outputs 4K, and its closing step said to connect an upscale node to any generation output, but the upscalers take still images only. FAQ and sources were added.
Correction, 2026-10-01: an earlier version of this page said Clarity turns a 1024-pixel image into a 4096-pixel one (8frame runs Clarity at 2x) and that upscalers do not invent content, which is not true of diffusion-based upscalers such as Clarity. BRIA Increase Resolution, added to 8frame in September, is now listed.
AI upscaling is the use of machine learning to increase the resolution of an image or video while restoring fine detail that a simple pixel-stretch would smear or make blocky.
Traditional upscaling algorithms, like bicubic interpolation, work by averaging existing pixels to fill in new ones. The result is softer and blurrier at scale. AI upscalers work differently: they are trained on pairs of low-resolution and high-resolution images, so they learn what detail is likely to be present at higher resolution, and they reconstruct it. Compare the two at full size: an AI upscale is meant to keep edges and textures crisp where a bicubic enlargement goes soft. Check faces, lettering and logos as well, because a model that adds detail can also change it.
For video, the same idea runs on every frame, and a video upscaler also has to keep the added detail consistent from one frame to the next, or it flickers.
How AI upscaling works
Early AI upscalers were convolutional neural networks: SRCNN (Dong et al., 2014) "directly learns an end-to-end mapping between the low/high-resolution images". Generative adversarial networks (GANs) followed; the ESRGAN paper describes SRGAN as "capable of generating realistic textures". Newer upscalers use diffusion models: Clarity, for example, is built on Stable Diffusion with ControlNet, so it re-draws detail rather than only sharpening what is there.
The core inference loop:
- The input image or video frame is fed into the model at its native resolution.
- The model predicts what additional high-frequency detail should exist at the target resolution, based on patterns from training data.
- That predicted detail is added to a high-resolution canvas alongside the upsampled base image.
- For video, a temporal step compares adjacent frames so the added detail does not flicker.
The output resolution is typically 2x or 4x the input. A 1080p frame becomes 2160p (4K). A 540p clip becomes 1080p. Upscalers differ in how much they invent: some are built to preserve the original content, while generative ones re-draw detail. Clarity is diffusion-based, and Magnific's Creative mode "adds or infers new detail"; that looks sharper but can change lettering, logos or faces. Magnific's Precision mode, like BRIA on 8frame, is built to stay faithful to the source.
When you use AI upscaling
The clearest use case is delivery gap coverage. Many AI video models output 720p or 1080p, and on 8frame none outputs 4K. If your deliverable is 4K (broadcast, large-format display, premium social), generate at the highest resolution the model offers, then bridge the rest with a video upscaler. On 8frame that step happens outside the canvas, after export: there is no video upscaler here.
Other common situations:
Archival footage. Older brand assets or footage captured at lower resolutions gets a resolution lift before it goes back into a campaign mix.
Social export variants. A single 1080p master gets upscaled to 4K for YouTube and downscaled to 720p for Story formats, from one source.
Post-generation polish. You generated a strong image, or pulled a still from a clip, but its native resolution isn't high enough for print, OOH, or broadcast specs.
Client deliverable compliance. Many agencies and in-house brand teams have minimum resolution requirements. Upscaling lets you meet them from AI-generated source material.
Examples in 8frame workflows
Topaz Image Upscale handles stills, and the upscale tool opens on it. On 8frame it always enlarges 4x and bills 11 credits per started 24-megapixel block of the output, up to 281. A 1024 x 1024 image becomes 4096 x 4096 (about 16.8 megapixels, one block) for 11 credits; a 2048 x 2048 image becomes 8192 x 8192 (about 67 megapixels, three blocks) for 33. Use it when the file goes to print or a large format. To be clear about what is not here: 8frame does not upscale video. Topaz Labs sells its video upscaler, Topaz Video, separately (a Mac and Windows app, with optional cloud rendering), and it does not run on this canvas. If a clip needs a video upscale, that happens in your own finishing tool after export.
Clarity Upscaler, at 13 credits, is diffusion-based: it re-draws detail as it enlarges, and 8frame runs it at a fixed 2x, so a 1024 x 1024 image becomes 2048 x 2048. BRIA Increase Resolution, at 6 credits, does 2x or 4x up to 8192 x 8192, and its model page says it preserves the original content with no regeneration, so try it first for labels, logos and faces.
All three run inside 8frame's canvas alongside the generation models, so you don't export and re-import between apps. More on Clarity in what is Clarity Upscaler.
FAQ
What is AI upscaling? AI upscaling uses a trained model to enlarge an image or video frame, typically 2x or 4x, and to fill in the detail a plain enlargement would blur. On 8frame it works on still images only, with Topaz, Clarity or BRIA Increase Resolution.
Does AI upscaling add detail that was not in the original? It can. Generative upscalers such as Clarity (diffusion-based) and Magnific's Creative mode re-draw detail, so check lettering, logos and faces at full size. Content-preserving upscalers such as BRIA Increase Resolution and Magnific's Precision mode are built to stay faithful to the source.
Can 8frame upscale video? No. Topaz Image Upscale, Clarity and BRIA Increase Resolution work on still images only. Generate the clip at the highest resolution the model offers, and upscale it in a desktop tool after export if you need more.
How much does AI upscaling cost on 8frame? Topaz Image Upscale is 11 credits per started 24-megapixel block of output (always 4x, up to 281), Clarity is a flat 13 (always 2x), and BRIA Increase Resolution a flat 6 (2x or 4x, up to 8192 x 8192). A credit is about $0.01 at pack rate.
Related concepts
- Topaz vs Clarity Upscaling compares the still-image upscalers on the canvas, what each costs and why none covers video.
- Best Free AI Image Upscalers in 2026 covers Upscayl, waifu2x, Bigjpg and imgmend with the limits each one states, and when paid upscaling earns its cost.
Sources
- How early AI upscalers learn ("directly learns an end-to-end mapping between the low/high-resolution images"): Dong et al., Image Super-Resolution Using Deep Convolutional Networks, arxiv.org/abs/1501.00092, checked 2026-10-08
- GAN-based super-resolution (SRGAN "is capable of generating realistic textures"): Wang et al., ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks, arxiv.org/abs/1809.00219, checked 2026-10-08
- What Clarity is built on (Stable Diffusion, ControlNet, tile processing, MultiDiffusion; "free and open-source Magnific Alternative"): github.com/philz1337x/clarity-upscaler, checked 2026-10-08
- Magnific Creative mode ("adds or infers new detail") and Precision mode ("faithful super-resolution without hallucinations"): docs.magnific.com/api-reference/image-upscaler-creative/image-upscaler, checked 2026-10-08
- BRIA Increase Resolution (preserves the original content with no regeneration; 2x or 4x; up to 8192 x 8192): fal.ai/models/bria/increase-resolution, checked 2026-10-08
- Topaz Video (video upscaling; Mac and Windows; local or cloud rendering): topazlabs.com/topaz-video, checked 2026-10-08
- 8frame upscalers (the upscale tool opens on Topaz; Topaz fixed 4x, 11 credits per started 24 MP block of output, capped at 281, calculateTopazCost; Clarity 13, always 2x; BRIA 6 for 2x and 4x; no video upscaler) and video output resolutions (no 8frame video model outputs 4K): 8frame tool registry, upscale tool code and content fact sheet, read 2026-10-08
Ready to add it to your chain? Open the 8frame canvas and connect an upscale node to any image on the board.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.