Topaz is the one model on Clipwave with no prompt box, and that is the honest starting point for this guide. It upscales what you feed it, so using it well comes down to two decisions: which input you give it — the sharpest, least-compressed version you have — and which settings you pick: image or video mode, 2x or 4x scale. It sharpens and enlarges detail that exists; it does not invent detail the source never had. Every example below is an input-plus-settings scenario, not prompt text.
Output quality is decided before you click anything: feed the original export, not a screenshot of it, not a re-compressed download. Every generation of compression you remove from the input shows up in the output.
Know where the asset ends up — web hero, 4K timeline, print — and its required pixel dimensions. Upscaling without a target is just making files bigger.
Output = input × scale, nothing more mystical. 1920×1080 at 2x is exactly 4K (3840×2160); 2048px at 4x is 8192px. Choose the factor that hits the target, not the biggest one.
Two separate tools on Clipwave. Use the video tool for clips — it processes the footage as footage — and the image tool for stills, including frames you export as images.
Upscaling is a finishing step. Generate, edit and pick winners at native resolution; upscale once, at the end, on the final asset.
After upscaling, inspect at 100% zoom and at final display size. Sharpened compression artifacts and plastic-smooth textures are the two failure modes to check for.
Template
[sharpest source you have] + [mode: image or video] + [scale: 2x to hit the delivery resolution, 4x only when the arithmetic demands it] -> run as the last step -> inspect at 100%Input: a clean 1344x768 generated image, the selected final take. Settings: image mode, scale 2x. Output: 2688x1536.
2x covers retina density at typical hero widths. Run it on the one image you actually ship, after selection — not on every candidate in the grid.
Input: a 2048x2048 final image. Settings: image mode, scale 4x. Output: 8192x8192 — roughly 27 inches square at 300 DPI.
Print is the legitimate 4x case: paper demands pixel densities screens never do. Do the DPI math first (pixels ÷ 300 = inches) so you know 4x actually reaches your print size.
Input: the finished 1920x1080 export of your edit. Settings: video mode, scale 2x. Output: 3840x2160 — exact 4K.
Upscale the finished master, not the raw clips: one pass over one file, and every cut, caption and grade inherits the resolution together.
Input: a sharp 1080x1920 master for Reels/TikTok. Settings: video mode, scale 2x. Output: 2160x3840, uploaded as-is.
Platforms re-encode everything you upload. Handing their encoder more resolution than it needs typically leaves the clip crisper after compression than uploading at the minimum.
Input: a soft 854x480 draft generation you like. Settings: none — regenerate at 1080p instead, then decide if you still need Topaz.
Included deliberately: 4x on soft 480p gives you large, sharp-edged blur, not 4K detail. When the generating model offers a higher native resolution, rerunning the prompt there beats upscaling the draft every time.
Input: a sharp 4K product photo. Crop to the detail region you want (say, a quarter of the frame), then run the crop at 4x.
Crop first, upscale second — scaling the full frame and then cropping wastes most of the pixels you paid for in processing. This is how you fake a macro shot from a standard photo.
Input: a 1280x720 archive clip that must sit in a 1080p timeline next to sharp footage. Settings: video mode, scale 2x to 2560x1440, then let the editor downscale it to 1080p.
Upscale-then-downscale oversamples: the clip lands at 1080p noticeably crisper than dropping the raw 720p into the timeline and letting the editor stretch it.
Input: a UI screenshot or screen recording headed for a demo video. Settings: 2x — and only if the text is already legible.
Text is the stress test for any upscaler: it sharpens existing letterforms but cannot re-draw ones that compression already broke. If the text is mushy at 100%, re-capture at higher resolution instead of upscaling.
Input: a frame exported as an image from your best clip. Settings: image mode, 2x. Output: a thumbnail that holds up at full-player size.
Frame grabs are softer than photographs — motion and video compression both cost detail. A 2x image pass is usually the difference between a mushy thumbnail and a usable one; go 4x only if the source frame was small.
Input: a session of 20 candidate generations. Settings: none during iteration — pick the winner at native resolution, then run a single upscale pass on it.
Upscaling every iteration slows the loop and multiplies processing for assets you will throw away. Native resolution is enough to judge composition and quality; the heavy pass is for the keeper.
2x or 4x. Choose by arithmetic, not ambition: output = input × scale. 2x covers almost every web, social and 4K-from-1080p case; 4x is for print, heavy crops and small sources that are genuinely sharp.
Image and video upscaling are separate tools on Clipwave. Run clips through the video tool rather than exporting frames through the image tool — it processes the clip as a clip.
There are deliberately no other knobs — no prompt, no creativity slider. If the output looks wrong, change the input, not the settings: feed a cleaner, less-compressed, higher-resolution source.
Every upscaling decision reduces to one line of arithmetic: output = input × scale. Work it backwards from the destination.
When the source is an AI generation, you often have a choice Topaz users of real footage never get: rerun the prompt at a higher native resolution. Native wins on detail — the model synthesizes new information at the larger size, while an upscaler can only work with what the smaller render contains. Topaz wins when the exact take is locked: the one generation where the pose, the face and the light all landed, a real photograph, a finished edit. The rule of thumb — regenerate to explore, upscale to preserve.
The order of operations decides how much of the upscale survives. Edit at native resolution, export one finished master, upscale that master once, then upload — the platform's encoder gets more resolution than it needs and throws away compression noise instead of picture. Two refinements: burn subtitles and text overlays after upscaling, not before, so the type stays vector-crisp instead of being processed as picture; and never upscale an already-uploaded, re-compressed download when you still have the original export — the original is always the better input.
Single images and exported frames — stills headed for web, thumbnails or print. 2x or 4x.
Finished clips and masters — processes the footage as footage. 2x or 4x.
No. On Clipwave it exposes exactly one setting — the scale factor, 2x or 4x — on two tools, image and video. The input file is the real "prompt": the quality of what you feed it decides the quality of what you get.
Do the arithmetic: output = input × scale, so pick the factor that reaches your delivery size. 2x covers most real cases — 1080p to 4K is exactly 2x. Reserve 4x for print, heavy crops, and genuinely sharp small sources.
It makes footage larger and sharper; it cannot reconstruct detail the source never recorded. Soft input becomes bigger, cleaner-edged soft output. If the source is an AI generation, regenerating at higher resolution beats upscaling a soft draft.
Last. Generate, select and edit at native resolution, then run one upscale pass on the final asset before delivery. Upscaling intermediates wastes processing and stacks artifacts if anything is re-compressed in between.
Because the source was soft — upscaling enlarges and sharpens what exists, and blur is part of what exists. Go back one step: find a sharper, less-compressed version of the input, or regenerate the asset at a higher native resolution, then upscale that.