AI photo editing for brand teams

AI image editing tools can now remove a passer-by from a campaign photo, turn a landscape shot into a vertical story without cutting the product, or put a packshot on a new background in a few seconds. For a brand team the editing is rarely the hard part any more. The hard part is everything around it: starting from the approved file, respecting the licence and the model release, keeping the product exactly as it is, applying the logo correctly, getting the result approved, and being able to tell later which image was edited and how.

This guide covers what AI editing does reliably today, task by task, what each task gets wrong, the checks to run before an edited image goes out, and why it matters where the editing happens.

What AI editing does well, task by task

Extending an image to a new format (outpainting)

A campaign is shot in one format and published in a dozen: square and 4:5 feed posts, 9:16 stories, 16:9 thumbnails, wide banners. Cropping a landscape photo to 9:16 throws away two thirds of it and often cuts the product. Outpainting works the other way: the image is placed on a larger canvas and the AI draws the missing areas (more sky, more floor, more wall) so the whole original stays in the frame.

It works well when the edges of the photo are simple: sky, studio backdrop, plain wall, grass, sand, water. It works badly when the edge cuts through something with structure (a person, a shelf of products, text, a recognisable building), because the model has to invent the rest of it.

What to check: repeated patterns in the new area, objects that continue wrongly across the original edge, warped straight lines, and any new element that looks like a product, a logo or a word.

The pixel sizes for each platform are in our social media image sizes guide.

Removing people and objects

Removing a passer-by, a reflection, a stray cable or a competitor's product is the most requested edit, and the most useful one for reusing older photos: a shoot from three years ago becomes usable again once the person whose release has expired is gone.

Two techniques exist. Detection-based removal finds the people in the image, lets you select them, and fills the area behind them. Painted removal (inpainting) fills any area you paint over. Detection is faster and more precise on people in clear view. Painting is needed for small, distant or shadowed people, and for objects.

What to check: shadows and reflections of the removed person that stay behind, body parts left at the frame edge, repeated textures in the filled area, and, in group shots, hands on shoulders that now belong to nobody.

Replacing backgrounds

Background removal cuts the subject out on a transparent canvas. On product photos with clean edges it is now close to perfect. Hair, fur, glass, smoke and fine mesh are still the hard cases.

The second step places the subject on a new background: a flat colour for marketplaces, a gradient or marble surface for packshots, or a generated scene (kitchen counter, outdoor terrace, holiday table) for lifestyle use.

What to check: the product's contact shadow (a product without one looks pasted), lighting direction that matches the new scene, product scale against the generated props, and reflections on glossy products that still show the old studio.

Relighting, denoising, sharpening and upscaling

These edits change the quality of an image, not its content. Relighting corrects mixed lighting and colour casts in interior shots. Denoising removes grain from low-light or high-ISO photos and old scans. Enhancement sharpens slightly soft images. Upscaling enlarges an image beyond its native resolution by generating plausible detail rather than stretching pixels.

What to check: upscaled text and fine product details, which the model may redraw slightly wrong (the letters on a label, the stitching on a bag), and skin that looks smoothed or plastic after enhancement.

Changing perspective or style

Some tools re-render a scene from another camera angle or in another style (illustration, poster, film look). These are generative edits: the result is a new image that resembles the original. They are useful for mood boards, social content and concept work, and should not be used where the image must show the real product or the real place.

What AI editing gets wrong, and why it matters for a brand

The failure that matters most for a brand team is not a visible artefact. It is a small, plausible change to something that must not change:

The product. A generative edit can redraw a logo on the product, change the number of buttons, alter the colour slightly or smooth a texture. A product photo that no longer shows the real product is a consumer-protection problem, not a style issue.
Text. Labels, packaging copy, prices and legal lines are often redrawn as near-text. Check every word in the edited area.
People. Changing a person's expression, body, skin or hair is a different use from the one most model releases cover. Removing a person whose release has expired is usually fine; altering a person who is still in the picture needs a release that covers it.
The licence. Many stock and agency licences forbid modifications, or forbid AI processing of the image. An edit can breach a licence even when the result looks perfect.
Disclosure. Some platforms, markets and clients require AI-generated or AI-edited content to be labelled. Content Credentials record how an image was made, but only when the tools in the chain write and keep them.

A checklist before an AI-edited image is published

Source: the edit started from the current approved version of the asset, not from a download someone kept on their desktop.
Rights: the licence and any model or property release allow modification, and allow the use you plan.
Product: shape, colour, logo, label text and details match the real product.
People: no person was altered beyond what their release covers; removed people left no shadow, reflection or body part.
Text: every word in the image is spelled correctly and still says what it said.
Brand: logo size, clear space and colours follow the brand guidelines.
Artefacts: no repeated patterns, warped lines, extra fingers or objects that continue wrongly across an edge, checked at 100% zoom.
Record: the edited image is stored next to the original with the reason for the edit and the name of the person who approved it.

Where the editing happens matters

The same edit can be made in a general-purpose AI image editor or inside the asset library. The result can look identical. What differs is what happens around it.

	General-purpose AI editor	Editing inside the DAM
Starting file	Whatever the person downloaded	The current approved asset
Licence and release check	Up to the person	The asset's rights decide whether AI editing is allowed
Original	Often overwritten or duplicated	Never changed; edits are versions
Brand logo and colours	Uploaded by hand each time	Taken from the brand guide, with size and clear-space rules enforced
Approval	Outside the tool, by email or chat	The edited image waits for review before others can use it
Record	None, or a file name	Who edited, with which tool, why, and who approved
Cost	Usually a subscription per person	Per edit, visible to administrators
Where the file ends up	A new upload, if anyone remembers	Attached to the original asset

The library approach does not make the AI better. It makes the edit traceable and stops an image that should not be modified from being modified at all.

How freedam handles AI editing

freedam's Studio is an AI image editor built into the asset library. It works on images only.

Twelve built-in tools: Crop (with platform presets, and outpainting when the crop box goes past the image edge), Resize (with AI upscaling above 125%), Remove / Add, Remove People, Remove Background, Add Background (flat colour or a generated scene), Brand Overlay, Perspective, Daylight, Enhance, Denoise and Style Transfer.
Custom skills: administrators write a prompt once, for example "place this product in a showroom", attach a list of options, and the team picks an option with one click. The prompt stays on the server.
Non-destructive sessions: the original asset is never changed. Every run adds a version to the session, and any version can be the starting point of the next edit.
Rights and access first: Studio opens only when the access rule that applies to the person allows AI editing of the asset and the asset's rights policy does not forbid modification.
Brand rules enforced: Brand Overlay places logos, emblems, typefaces and colours from the brand guide, checks minimum size, maximum size and clear space as the logo is placed, and refuses to render an overlay that breaks a rule.
Approval: saving a version requires a written reason and attaches the image to the asset as a rendering that waits for review before others can use it.
Cost and audit: every run is recorded with its cost, and administrators can see which tool, option and prompt produced each result.

Details in the Studio documentation and custom skills.

Frequently asked questions

Can AI remove a person from a photo without leaving traces?

On people in clear view against a simple background, yes, in most cases. Check for leftover shadows, reflections and body parts at the frame edges, and zoom to 100% on the filled area. Small or shadowed people are often missed by detection and need to be painted over by hand.

Is outpainting good enough for campaign images?

For simple edges such as sky, studio backdrop, sand or a plain wall, usually yes. When the new area has to continue people, products, text or architecture, the result needs careful review and is often better replaced by a different crop or a reshoot.

Do we need to label AI-edited images?

It depends on the market, the platform and the client. Some platforms label AI content they detect, and some regulations and brand policies require disclosure of significant AI edits. Keep a record of which images were edited and how, so you can answer the question for each one.

Can stock photos be edited with AI?

Only if the licence allows it. Many licences restrict modifications, and some explicitly restrict AI processing. Check the licence of each image, or record the restriction on the asset so editing tools can refuse it.