Upscale
Upscalers grow an image to delivery resolution — the finishing step after you've generated and curated. They are paid nodes: the source image goes to the upscaling provider with your own key, and cost scales with output pixels, not input.
Upscale Image
Wire one image in (from Image In, any generator, a Shot Board tile…), pick a model, run. The upscaled image flows out the bottom port like any other image — into a Preview, a Conform, or straight to disk via the ledger.
- Batch: wire several images (an Image In folder, a curated Shot Board pool) and each one upscales as its own job — two in parallel, with a per-image cache, so a re-run only pays for what changed and a partial failure retries just the failures. The estimate line sums the whole set, and each output's 🔍 compares against its input.
- Caching works as everywhere: unchanged input + unchanged dials = free replay; the 🕘 history lets you re-pin any past result.
- The node shows a live estimate — input size → output size, megapixels, and the approximate cost — before you spend anything.
Models (Topaz Labs)
| Model | What it does |
|---|---|
| Gigapixel Standard 2 / High Fidelity 2 / Low Resolution 2 | Faithful precision upscaling — no invention, tuned respectively for everyday use, maximum preservation, and tiny sources. |
| Gigapixel Art & CGI | Precision upscaling for renders, illustrations and graphics — keeps clean edges and flat color, avoids photographic texture artifacts. The archviz workhorse. |
| Redefine / Standard MAX / Recover 3 | Generative enhancement (photoreal, true-to-input, or restoration) with an optional guidance prompt and a 1–9 creativity dial. |
Topaz bills credits (see the provider page); the node shows your live balance and a per-run credit estimate, and the ledger records the credits each job actually billed.
Models (Magnific)
| Model | What it does |
|---|---|
| Upscaler Creative | The famous "reimagining" upscaler: 2x/4x/8x/16x with a guidance prompt and the creativity / HDR / resemblance / fractality dials, plus an engine choice (Illusio = smooth, Sharpy = crisp, Sparkle = detailed). It invents plausible detail — great for stylized finishes, wrong for strict fidelity. Input capped at 25.3M pixels. |
| Upscaler Precision V2 | Faithful, detail-preserving upscale from 2x to 16x with sharpen / smart grain / ultra detail knobs and a flavor per image type: photo, sublime (illustration), photo_denoiser (grainy or low-light sources). The archviz default. |
Compare — before / after
The 🔍 on the node's result opens straight into a full-screen before/after comparison of the input against the upscaled output: drag the divider to sweep between them (the input is shown at the same size, so you're comparing detail, not scale). The mouse wheel zooms — the fast display copy is shown until you zoom, then the original files load so you're judging real pixels — drag pans when zoomed in, double-click resets, Esc or a click outside closes. The same wheel-zoom works in every magnified preview across Brut.
Presets
Your favorite dial combinations can be saved as named upscaling presets — the 💾 button next to the preset select stores the current model + dials in Brut's global library (all projects, both dev and installed app). Picking a preset applies it; editing any dial detaches it (your params are always the truth); ✕ deletes it from the library.
Upscale Video
The clip finisher: wire one clip in (a Clip Board tile, an Image → Video output, a whole Sequence), pick a provider and model, choose a scale — plain 2x / 4x or a standard target: → HD, → 4K, → 8K (the target sizes the clip's short side, so a portrait 9:16 reel "at 4K" comes out 2160 wide) — and run. The upscaled clip flows out — audio is kept.
Two providers, same models:
Topaz Labs (cloud) — bills credits per job, works anywhere.
Topaz Video AI (local) — free: if you own the Topaz Video desktop app, Brut runs the upscale on your GPU. No key, no credits, no confirm dialog, and the clip never leaves your machine. See the provider page.
Nothing bills without your OK — every cloud run opens an itemized confirmation with the credit estimate first (cached and pinned replays never ask; free local runs don't either). Topaz's own analysis is the billing truth: the lower bound of its quote is billed and lands in the ledger.
Cloud jobs can take a while on real clips; they run engine-side, survive restarts, and reattach. Local jobs simply re-run if interrupted — nothing was billed.
Output caps at 8K per side; cloud uploads at 500MB.
| Model | When |
|---|---|
| Proteus 4 | the general-purpose default — upscale, denoise, sharpen in one pass |
| Rhea | natural-texture enhancement, higher scales |
| Gaia CG | rendered/CG clips — archviz animations, clean edges |
Cost
Upscales are priced by output image area (and for video, × duration). Magnific's doc rates work out to roughly €0.025 per output megapixel with a €0.10 minimum — Brut shows this as an approximate $ figure before the run, and the ledger records the estimate from the actual output pixels after it. Rough feel: a 1080p still → 4K is ~€0.20; pushing a 640×480 crop to 16x lands near €2.
As everywhere in Brut: your key, your bill, zero markup.