MagnificUpscaling & exportParameters & settings
Magnific Precision vs Creative Upscaler: Which to Use
Use Magnific's Precision upscaler when the image must stay exactly as it is: photos, products, logos, UI and anything a client has approved. Use Creative when you want the upscaler to invent texture and detail, guided by a prompt and sliders such as creativity, HDR, resemblance and fractality: AI images, renders, concept art and illustration. Precision sharpens what exists; Creative adds what was not there.
Magnific made its name with an upscaler that does more than enlarge: it reimagines detail, so a soft 1,024-pixel AI image can come back with pores, fabric weave and stone texture. That is exactly what you want for a concept render and exactly what you do not want for a product photo with a legal label. The answer is two modes. This guide explains what each does according to Magnific's own documentation, the controls each exposes, and a decision table for common jobs. Interface labels move from time to time; the behaviour below is what matters.
The two modes in one sentence each
- Creative "adds or infers new detail guided by your prompt", in Magnific's words, for stylised, higher-impact upscales.
- Precision is faithful super-resolution that preserves the original content, texture and structure "without hallucinations", and that does not take a prompt.
Both increase resolution and improve quality; only Creative changes what the image depicts at the detail level. Magnific's documentation lists Creative's scale factors as 2x, 4x, 8x and 16x in the API. Its help pages give different maximums for the web app in different places, so check the scale selector in your interface.
Creative mode controls
Creative exposes a prompt and four sliders. In the API each slider runs from −10 to 10 with a default of 0; the web interface may show them on a different scale.
| Setting | What it does | Start with |
|---|---|---|
| Prompt | Tells the upscaler what the details should be | Reuse the image's original prompt, or a short literal description |
| Creativity | How much new detail is invented | Default; lower for faces and products |
| HDR | Level of definition and detail | Default or slightly lower to avoid crunchy edges |
| Resemblance | How close the result stays to the source | Raise for approved compositions |
| Fractality | Strength of the prompt and intricacy per square pixel | Default; lower if small repeated patterns appear |
| Engine | Automatic, Illusio, Sharpy or Sparkle | Automatic, then compare |
| Optimised for | A style preset such as portraits, illustration, 3D renders or landscapes | The preset that matches the image |
Two tips from the documentation are worth following. First, reusing the same prompt that generated an AI image improves the upscale. Second, send the original file: Magnific notes that re-encoding a PNG as JPEG or resizing the image before upscaling reduces quality, and that browser canvas exports can lose around 8 to 20 per cent.
The style presets in the API are standard, soft portraits, hard portraits, art and illustration, video game assets, nature and landscapes, films and photography, 3D renders, and science fiction and horror. Our guide to upscaler settings explains each slider with examples.
Precision mode controls
Precision has no prompt and fewer, gentler controls. In the API they are sharpen (default 50), smart grain (default 7) and ultra detail (default 30), each from 0 to 100. Magnific lists what Precision is for: preserving edges with minimal halos, reducing noise and compression artefacts, keeping colour consistent and improving the readability of small text and UI. Its suggested uses are e-commerce product photos, logos and icons, screenshots, restored scans, technical drawings and images for high-DPI screens.
- Lower sharpen if edges start to glow.
- Add a little smart grain to stop skin and gradients looking plastic.
- Raise ultra detail carefully; it brings back fine texture but can exaggerate noise in a poor source.
Which mode for which job
| Setting | What it does | Start with |
|---|---|---|
| Client-approved photo or product shot | Must not change | Precision |
| Logo, icon, UI screenshot, diagram | Lines and text must stay exact | Precision |
| Old scan or compressed JPEG | Clean up artefacts and noise | Precision, low sharpen |
| Soft AI image for a poster | Needs real texture at size | Creative, the image's own prompt |
| Architectural or 3D render | Add material detail | Creative, 3D renders preset |
| Portrait from an AI model | Skin detail without a new face | Creative, portrait preset, high resemblance |
| Concept art or game asset | More intricacy welcome | Creative, higher creativity and fractality |
A useful combination: run Creative at 2x to add texture to an AI image, choose the best result, then use Precision for any further enlargement so the detail you approved is not reinvented a second time.
How to compare the two fairly
Judge upscales the way the image will be used. Open both results at 100% side by side and look at three places: an edge with high contrast, an area of fine texture such as hair or fabric, and anything that must stay exact, such as a label or a face. Then view both at the final display or print size, because detail that looks invented at 100% can read as natural texture on paper. Pick per image, not per project: a campaign can mix modes as long as each final file is recorded.
Common mistakes
- Creative on a face you need to keep. Even with high resemblance, Creative can change features. Use the portrait presets, keep creativity low, and compare at 100%.
- Upscaling before editing. Retouch, crop and colour-correct first. Upscale once, at the end.
- Huge factors in one step. Several modest passes are easier to judge than one 16x jump, and you can stop when the print size is reached.
- Forgetting the record. Note the mode, factor, engine, preset and slider values. Matching a series later is impossible without them.
Upscaling costs are based on the output size, so larger factors cost more. Check the current pricing page rather than older tutorials. For the background to the product names, see Freepik is now Magnific.
Try it
close-up portrait, natural skin texture with fine pores, soft window light, linen shirt with visible weave, realistic photograph
Use it as the Creative prompt on a soft AI portrait at 2x with a portrait preset and resemblance raised. Then upscale the same source in Precision with default settings. Compare the eyes and the shirt at 100%: if the face changed in Creative, lower creativity or stay with Precision.