Enhance and upscale images in Krea
Upscale to 4K+ and add detail with the AI enhancer.
What Is Enhancer?
Krea Enhancer is an AI-powered upscaling and refinement tool for images and videos. It is designed to improve resolution, recover lost detail, and sharpen clarity — whether you are working with an existing asset or an image previously generated in Krea. Common use cases include blowing up a low-resolution archive image for print, sharpening a photograph that is blurry or underexposed, or bringing a low-resolution AI output up to a quality suitable for larger formats.When to Use Enhancer
Getting Started
1. Upload or select an image
Upload the image you want to enhance, or select one from your previous Krea generations. The left sidebar gives quick access to previous enhancement jobs.2. Choose an enhancement model
The model selector is at the bottom left of the page. Choose your model before adjusting any settings, as each model exposes a different set of controls. See the model comparison table at the end of this page for guidance on which to use.3. Set an upscale factor
Four scaling options are available: 1x, 2x, 4x, and 8x. Use 1x if the original image has adequate resolution but needs sharpening or clarity improvements. Choose a higher multiplier to increase the output dimensions at the same time. Note that higher upscale values give the AI more room to interpret and fill in detail, which can introduce subtle modifications to textures and edges.4. Adjust the settings
Available settings vary by model. The table below describes each one.
A note on high detail settings. Strength, clarity, sharpness, and texture all sound like improvements in isolation, but higher values are not always better. Upscaling a painting with high texture will fill in additional brushstrokes that were not in the original, which can feel authentic up close but noisy when viewed at full size. With photography, high sharpness and clarity settings can produce an over-textured effect on smooth surfaces — adding visible pores and fine lines to skin that was originally clean.
The right approach is to start conservatively and increase settings only until you reach the result you want.
5. Add a prompt (optional)
A prompt describes the final image you want to see and helps the AI make decisions about how to fill in missing detail. This is particularly useful when the source image is ambiguous — for example, a silhouetted figure where the clothing and appearance are unclear.6. Scene Transfer (Krea Legacy only)
Scene Transfer allows you to modify the lighting, environment, and mood of an image. Upload a reference scene or enter a text-based scene description to guide the transformation. For example, applying a “cinematic blue lighting” description can shift an image toward a high-contrast, film-inspired look.7. Run the enhancement
Click Enhance to start processing. Generation time ranges from roughly 10 to 120 seconds depending on the model and upscale factor selected. Once complete, drag the comparison slider to view before and after versions side by side, or use the view controller at the bottom right to switch between layouts. Your result appears as a thumbnail at the bottom of the screen and is also saved to Assets. You can run multiple enhancements of the same image with different settings — each result saves as a separate thumbnail so you can compare them directly. Multiple enhancement jobs can also run simultaneously.Enhancement Presets (Krea Legacy)
In the Krea Legacy model, presets adjust the settings sliders automatically to target a particular aesthetic.Image Enhancement Models
A note on upscale dimensions
The upscale factor refers to the increase in width and height, not total file size. Because image dimensions scale in two directions, the overall size of the image grows as follows:
If an image has already been enhanced at 2x and you run a second 1x enhancement, the second pass operates on the 2x version, not the original. Compounding enhancements will compound the AI’s interpretations, so it is worth being deliberate about sequencing.
For high-resolution outputs such as large prints, upscaling in steps — 2x first, then 4x — tends to produce more controlled results than jumping directly to 8x.