EditorNodesPricingBlog

IMAGE UPSCALING

Bigger, and actually sharper

Generative upscaling adds detail rather than stretching the pixels you already have. Up to 16x with identity preservation, plus direct control over how much the model is allowed to invent.

Upscale an image

Why generative upscaling looks different

Traditional resizing has nothing to work with. Enlarging an image means spreading the existing pixels over a bigger area, and the result goes soft because no new information was added. Sharpening afterwards adds contrast at the edges without adding detail.

Clarity Pro generates the missing detail instead. It reads what the subject is and produces texture consistent with it, which is why fabric still looks like fabric and skin still looks like skin at factors where interpolation produces mush.

The trade off is that invented detail is invented. On a face, on text, or on a logo, the model can produce something plausible that is not what was there. That is what the resemblance control exists to restrain.

Upscaling in the Image Enhance form

Upscaling shares its form with background removal, so finding it starts with the operation selector.

Generation type dropdown with Image Enhance selected in the image workspace

Upscaling lives under Image Enhance in the generation type dropdown.

1

Select Image Enhance

Upscaling is not a separate generation type. It sits inside Image Enhance, which also holds background removal, so the first step is choosing Image Enhance and then picking the operation.

This matters when you are looking for it in the interface, since there is no entry labelled upscaling in the type list itself.

2

Choose the upscaling operation

Image Enhance covers two operations. Background removal cuts the subject out, and upscaling raises resolution. Selecting upscaling brings up the factor and quality controls.

The two operations run separately. If a shot needs both, remove the background first and upscale the result, since upscaling a large image you are about to cut down wastes the extra pixels.

Operation type selector offering background removal and image upscaling

The operation selector. Upscaling and background removal share this form.

Model selector showing Clarity Pro as the upscaling model

Clarity Pro handles upscaling, with identity preservation across the factor range.

3

Clarity Pro does the upscaling

Clarity Pro is the upscaling model, and it goes up to 16x with identity preservation, meaning faces and fine structure survive the enlargement rather than being smoothed into approximations.

This is generative upscaling rather than interpolation. The model adds plausible detail rather than stretching existing pixels, which is why it holds up at factors where traditional resizing falls apart.

4

Seed for reproducibility

Because generative upscaling invents detail, two runs at the same factor can differ. Fixing the seed makes a result reproducible, which matters when several frames from one sequence need to be treated identically.

Leave it empty when you want to try again for a different interpretation of the fine detail.

Seed field in the image enhance form for reproducible upscaling results

The seed field, which makes an upscale reproducible.

Settings reference

Every control in the upscaling form and what it changes.

Source image

The image to enlarge. Upload a file or select one of your previous completed tasks from the workspace.

Upload or previous result

Operation type

Selects upscaling rather than background removal. Both operations live under Image Enhance and share one form.

Upscaling or background removal

Upscale factor

How much larger the result is. Credit cost scales with the factor, so pick the one your delivery actually needs rather than the largest available.

1, 4, 8, 16, 25 or 64

Creativity

How much new detail the model is allowed to invent. Higher values produce more texture but drift further from the source.

Adjustable

Resemblance

How closely the result must track the original. Raise it when the upscale is changing things you wanted preserved.

Adjustable

Seed

Makes the upscale reproducible. Useful when several frames from one sequence need identical treatment.

Integer, optional

Picking a factor

Work backwards from delivery. A 1024 pixel frame going to a 4K timeline needs roughly 4x. The same frame going to a print at A3 needs more. Choosing 64x because it is available costs credits proportionally and gives you a file nobody asked for.

Creativity and resemblance are the two dials worth learning. Raise creativity when the source is clean and you want richer texture. Raise resemblance when the upscale is changing a face, altering text, or drifting from something that needs to stay recognisable.

For a sequence of frames that will sit next to each other, fix the seed. Without it, each frame gets its own interpretation of the fine detail, and the inconsistency shows once they play in order.

Related tutorials

Coverage of enhancement work on the blog.

BEN2 background removal

The other operation under Image Enhance, and how it pairs with upscaling.

Real time video upscaling

How upscaling behaves on moving footage rather than a single frame.

Upscaling questions

How much can an image be upscaled?

The factor options are 1, 4, 8, 16, 25 and 64. Clarity Pro provides photorealistic upscaling up to 16x with identity preservation. The higher factors exist for specific cases, but the sensible choice is the smallest factor that reaches your delivery resolution.

Yes. Traditional resizing stretches the pixels that already exist, which is why enlarged images go soft. Generative upscaling adds plausible new detail based on what the model understands the subject to be, so texture and edges stay defined at factors where resizing fails.

Creativity governs how much new detail the model invents, and resemblance governs how tightly it must track the original. They pull against each other. If an upscale is inventing detail you did not want, raise resemblance and lower creativity.

After. Upscaling is a delivery step. Running it on drafts spends credits on frames you will discard, and editing an upscaled image means every subsequent pass works on a larger file for no benefit.

Yes. Credit cost scales with the upscale factor, and the exact figure appears in the form before you submit. This is the main reason to pick the factor your output needs rather than the largest one offered.

Because the model generates the added detail rather than computing it. Fix the seed to make the result reproducible, which is what you want when several frames from one sequence must be treated the same way.

Yes, and that is the right order. Remove the background first, then upscale the result, so the enlargement is not spent on pixels you are about to cut away.

Other ways to generate images

Image Upscaling is one of seven generation types in the image workspace.

Text to Image

Write a prompt and generate images from scratch. Pick the model, aspect ratio and output format, and add a LoRA style if you have one.

Image to Image

Start from an image you already have and describe the change. Useful for style transfer, edits and mood changes on an existing frame.

Character Generation

Upload one to three reference photos and keep the same character consistent across many generations.

Background Remover

Cut the background out of an image with edge detection precise enough for compositing work.

Model Training

Train a custom FLUX LoRA on your own images so you can generate in your own style or with a specific subject.

Draw to Edit

Sketch an idea on the built in canvas, add a prompt, and turn the drawing into a finished image.

    © 2026 - AI FILMS LLC