MagnificUpscaling & exportParameters & settings
Magnific Upscaler Settings Explained: Creativity to HDR
Magnific upscales and enhances images, and its sliders decide how much it invents: creativity adds new detail, resemblance holds the image to the original, HDR sharpens and lifts contrast, fractality controls how much the prompt shapes fine detail, and the engine and optimised-for choice set the overall character. Start low on creativity, keep resemblance moderate and raise one slider at a time.
Magnific (now magnific.com; the company formerly called Freepik relaunched under the Magnific name in April 2026) is an upscaler that does more than enlarge. Depending on how you set it, it will add pores, fabric weave, foliage and texture that were never in the source, or it will quietly clean and sharpen while leaving the picture alone. Every control is a dial between those two behaviours. This guide explains each setting, how they interact and where to begin for the kinds of images working creatives upscale most.
How Magnific works, briefly
Rather than stretching pixels, Magnific regenerates the image at a larger size using a diffusion process guided by the original. The original tells it what should be where; the settings tell it how freely to reinterpret. That is why the result can look better than the source, and why it can also drift: a face can gain detail that changes its likeness, text can become gibberish, and a soft illustration can turn photographic.
The settings
- Scale factor. How much larger the output is than the input, in steps such as 2x, 4x and higher. Bigger steps cost more credits and give the model more room to invent. For a clean result, upscale in two moderate passes rather than one large jump.
- Creativity. The headline control. At low values the upscaler respects the source and adds only plausible fine detail. At high values it hallucinates freely: new textures, re-drawn features, extra elements. For anything where likeness matters, stay low.
- HDR. Increases local contrast, sharpness and definition. A little makes the image crisp; a lot gives a hyper-detailed, over-processed look with dark halos around edges. Portraits usually want less than landscapes or renders.
- Resemblance. How strongly the output must match the original’s structure and colours. Raise it when the upscaler changes things you want kept (faces, logos, layouts). Lower it when you want it to repair a weak source. High resemblance with high creativity fight each other and produce muddy detail.
- Fractality. How much the prompt influences detail at small scales. Higher values repeat the prompt’s described textures across the image at finer and finer levels, which can be striking on abstract art or fabric and distracting on faces. Keep it low unless you are deliberately texturing.
- Engine. The underlying model variant. At the time of writing Magnific offers several, each with a different balance of fidelity and invention; the names and line-up change, so test the same image on two engines before settling.
- Optimised for. A preset that tunes the engine for a type of image: portraits, art and illustration, nature and landscapes, films and photography, 3D renders, game assets and so on. Pick the one closest to your source; it changes how skin, foliage and hard edges are treated.
- Prompt. A short description of what the image shows. It steers the detail the upscaler adds. Describe what is there, not what you wish were there, unless you are using creativity on purpose.
How the sliders interact
Creativity and resemblance are the main pair. Think of creativity as how much to invent and resemblance as how tightly to stay on the original’s rails. Low creativity plus moderate resemblance is the safe zone for client work. Higher creativity needs higher resemblance to stay recognisable, but push both too far and the result loses coherence. HDR sits on top: it does not change what is drawn, only how hard the edges and contrast are. Fractality is the one to leave alone until the rest is right, because it multiplies whatever the prompt says.
Pick the optimised-for preset before touching any slider; it shifts the behaviour of all of them. Then fix the seed, change one slider, compare at 100% zoom on the part that matters (eyes, type, product edges).
Settings at a glance
| Setting | What it does | Start with |
|---|---|---|
| Scale | Output size relative to input | 2x, then a second 2x pass if needed |
| Creativity | How much new detail is invented | Low (slightly above zero); very low for faces and logos |
| HDR | Local contrast and sharpness | Low for portraits; low-medium for landscapes and renders |
| Resemblance | Fidelity to the original structure and colour | Moderate; raise if anything changes that must not |
| Fractality | How strongly the prompt shapes fine texture | Zero or very low |
| Engine | Model variant | The default; test one alternative on the same seed |
| Optimised for | Preset tuned to the image type | Closest match to your source |
| Prompt | Describes the image to guide detail | One factual sentence |
| Seed | Reproducibility | Fixed while comparing |
Credit cost rises with scale and output size; check the platform’s current pricing page.
Starting points by image type
- Portraits and people. Optimised for portraits, creativity very low, resemblance moderate to high, HDR low. Check eyes, teeth and hairline at 100%. If the likeness shifts, lower creativity before anything else.
- Product shots and packaging. Creativity very low, resemblance high, HDR low-medium. Text and logos are the weak point; if they degrade, mask them out or composite the original type back over the upscale in your editor.
- Illustration and concept art. Optimised for art and illustration, creativity low-medium, HDR low. Raise fractality only if you want the brushwork or texture pushed.
- 3D renders. Optimised for renders, creativity low-medium to add material detail, HDR medium for crisp edges. Watch reflections and straight lines.
For a comparison with other upscalers and when to pick each, see best AI image upscalers compared. If you also want to change the lighting or look of an image, the Relight and Style Transfer guide covers Magnific’s other tools.
Try it
Image: a 1024px AI-generated portrait of a woman in a green wool coat, soft overcast light.
Prompt: close-up portrait of a woman in a green wool coat, natural skin texture, soft overcast daylight, shallow depth of field.
Scale: 2x. Optimised for: portraits. Engine: default.
Creativity: low (just above zero). HDR: low. Resemblance: moderate. Fractality: 0. Seed: fixed.
Run it, then run again with only creativity raised one notch and compare the eyes and lips at 100% zoom; the moment the likeness changes is your ceiling for this image, and most portraits hit it sooner than you expect.