ChatGPT’s latest image-editing capabilities got us curious.

AI has quickly become part of the photography workflow. Adobe Lightroom and Photoshop now include AI-assisted masking, blemish removal, distraction removal, generative tools, culling features, and more. Dedicated platforms such as Imagen, Aftershoot, and Evoto are going even further, promising to edit hundreds or thousands of photographs using AI.

So we wanted to try something different.

Instead of using software built specifically for photographers, what happens if you simply upload your photographs into ChatGPT, describe the edits you want, and let it do the work?

Could the ChatGPT interface realistically be used for photo retouching? Would increasing ChatGPT's reasoning level produce better edits? How fast would it be? And could you actually batch edit a photography session this way?

We put it to the test.

Photo Editing Has Always Used Computation

There is an important distinction to make before calling everything photographers have been doing for the past 30 years "AI."

Digital photography has always depended heavily on computation.

Exposure adjustments, sharpening, noise reduction, color correction, healing brushes, clone tools, masks, and countless other Photoshop and Lightroom features use mathematical algorithms to manipulate image data.

But a Clone Stamp and today's AI tools are not the same thing.

A traditional Clone Stamp essentially asks the photographer to identify the source pixels and tell the software where to place them. A Healing Brush goes further by analyzing texture, color, and surrounding information to make that replacement blend naturally.

Modern AI tools increasingly make those decisions themselves.

Adobe's current Lightroom tools can automatically identify blemishes and people, while Generative Remove analyzes an area and creates replacement imagery that attempts to blend with the surrounding photograph. Photoshop's Remove tool can similarly use generative AI to reconstruct unwanted areas.

The biggest change is that photographers increasingly don't have to tell the software how to perform the edit.

They can tell it what they want.

That is what makes ChatGPT interesting.

Instead of opening a healing tool and clicking individual blemishes, we could upload a portrait and say something like:

Remove minor skin blemishes while maintaining realistic skin texture and facial features. Add moderate contrast and warmth while keeping the portrait natural.

ChatGPT interprets the instruction and determines how to create the finished image.

OpenAI says its current ChatGPT Images system is designed to edit uploaded images while better retaining details such as the person's appearance, lighting, and composition. ChatGPT Images can also now work with additional reasoning through "Images with thinking."

That gave us an interesting opportunity.

What happens when we give ChatGPT more time to think about the exact same type of photo edit?

Testing Different Levels of ChatGPT Intelligence

GPT-5.6 Sol now gives eligible ChatGPT users a reasoning slider.

OpenAI currently describes its main levels as Instant, Medium, High, and Extra High, with increasing amounts of reasoning applied to a request.

For our test, we concentrated on progressively higher levels and repeatedly gave ChatGPT similar portrait-retouching instructions.

The goal wasn't to transform the photographs.

We wanted something a working photographer might actually ask an editor to do:

Clean up temporary blemishes, maintain natural skin texture, preserve facial characteristics, introduce some warmth, improve the overall color and contrast, and avoid making the subject look obviously retouched.

And there was absolutely a difference in the results.

Lower Reasoning: Usable, But Look Closely

At the lower level, the edits were surprisingly usable.

If someone saw the finished photograph casually on social media or a phone screen, there was a good chance they wouldn't immediately recognize anything wrong with it.

For an untrained eye, many of the photographs probably looked completely acceptable.

But once we started examining faces closely, the weaknesses became apparent.

The biggest problem was softness.

Areas where blemishes had been removed could become overly smooth. Skin texture sometimes disappeared around the correction instead of naturally blending with the surrounding skin.

It felt less like professional retouching and more like someone relatively new to photo editing had discovered the healing and smoothing tools and used them a little too heavily.

The photograph wasn't necessarily bad.

It just looked processed.

That distinction became increasingly important as we moved through the reasoning levels.

Medium Reasoning: More Realistic, But Still Inconsistent

Medium was arguably a significant improvement.

The excessive softness was reduced.

Skin generally retained more texture and the edits started feeling considerably more believable.

Instead of simply smoothing away imperfections, ChatGPT appeared more willing to preserve details in the face.

But Medium introduced a different problem.

Sometimes it left behind small details or transitions that made it possible to tell where something had been edited.

The photograph might look more realistic overall, yet an area around a blemish, edge, shadow, or piece of skin texture could still draw attention once you knew where to look.

In other words:

Lower sometimes hid the edit by smoothing too much. Medium preserved more detail but occasionally exposed the edit by not blending everything perfectly.

It was better.

But it still didn't consistently feel like professional portrait retouching.

High Reasoning: This Was the Turning Point

High produced our strongest results.

This was the first level where the editing consistently began feeling less like an automated cleanup tool and more like something a photographer or experienced retoucher might actually deliver.

The biggest improvement wasn't simply blemish removal.

It was restraint.

Facial characteristics were retained.

Natural skin texture remained visible.

The person still looked like the person in the original photograph.

The system seemed less interested in creating a technically "perfect" face and more interested in deciding which imperfections should actually be removed.

That is exactly what good portrait retouching should do.

A professional photographer usually isn't trying to erase every pore, wrinkle, freckle, line, or characteristic that makes someone recognizable.

Temporary blemishes may disappear.

Permanent characteristics often stay.

That philosophy was much more apparent in the High results.

Of the three levels we tested, High came closest to feeling professionally retouched rather than AI retouched.

Interestingly, that preference matches what photographers themselves say they want from AI.

A 2026 survey of 363 photographers reported by PetaPixel found that 87% valued natural-looking retouching, while 61% specifically checked whether AI preserved image detail. Seventy-eight percent said they would prefer AI to handle roughly 70–80% of retouching rather than surrender the entire creative process to automation.

That is essentially what we were seeing.

The best edit wasn't the one that removed the most.

It was the one that knew what not to remove.

Then We Measured Speed

Quality is only half of the equation for a working photographer.

Time matters.

We measured from the arrival of the uploaded photograph to the finished image being returned.

There is an important problem with the first number.

During the lower-level testing there were several occasions where we stopped to provide additional feedback or refine the request. Those interruptions increased the average.

When those pauses were excluded, the typical completion time was closer to:

52 seconds per photograph.

We then looked specifically at the actual rendering portion of three of our most recent High-level edits.

Those averaged approximately:

39 seconds per image.

That produced one of the most surprising findings from the entire experiment.

High reasoning didn't appear to make photo editing meaningfully slower.

In our testing, it actually produced our best combination of quality and speed.

This shouldn't be treated as a laboratory benchmark. Cloud workloads, image complexity, file sizes, network conditions, and differences between photographs can all affect processing time.

OpenAI itself says ChatGPT image generation can take from seconds to a few minutes depending on the request.

But for our practical test, roughly 40 seconds of rendering for a carefully retouched portrait was impressive.

Can ChatGPT Batch Edit Photos?

This was the test we were most interested in.

A professional photographer rarely needs to edit one photograph.

They might need to edit 50.

Or 500.

Or 2,000.

We uploaded seven photographs simultaneously and requested the same general level of editing:

Blemish cleanup, noticeable warmth, color correction, and general portrait polish.

The seven photographs were completed in approximately:

1 minute.

That works out to an effective throughput of roughly:

8–10 seconds per photograph.

That doesn't mean ChatGPT suddenly began creating each image in eight seconds.

Multiple images were being handled within the same overall workflow.

That distinction is important because our individual High-level render measurements were still closer to approximately 39 seconds.

But from the photographer's perspective, throughput is ultimately what matters.

If seven photographs enter and seven finished photographs come back approximately one minute later, you effectively processed seven photographs per minute.

Using that result as a rough projection gives us:

For easy planning, we would call that approximately:

15 minutes per 100 images

and

30 minutes per 200 images.

There is a huge disclaimer attached to those numbers.

We tested seven images.

We did not successfully run 100 or 200 photographs through ChatGPT simultaneously.

Those larger numbers are extrapolations based on the throughput we observed. ChatGPT could encounter upload limits, processing limits, changing server demand, or other bottlenecks as the batch becomes larger.

OpenAI also states that the number of images you can include depends on factors such as image size and accompanying text, rather than promising a fixed professional batch limit.

So while ChatGPT demonstrated that multiple-image editing is possible, this is not yet the same thing as a dedicated Lightroom batch-processing workflow.

How Does That Compare With Actual Photography Software?

This is where ChatGPT runs into some very serious competition.

The tools aren't doing exactly the same job, so raw speed comparisons need some context.

ChatGPT is interpreting a natural-language request and performing a generative edit.

Lightroom, Aftershoot, Imagen, and similar platforms are primarily designed around photo-development workflows. They can analyze a RAW image and intelligently adjust exposure, white balance, color, crop, masking, and other settings without necessarily rebuilding parts of the image.

That difference makes dedicated photography applications extremely efficient at scale.

Adobe Lightroom

Lightroom has been designed around large photography libraries for years.

You can edit one representative photograph, copy those adjustments, and batch-apply them across multiple selected photographs. Lightroom also allows presets to be applied across entire groups of photographs.

Adobe has now layered AI on top of that workflow.

In 2026, Lightroom added one-tap AI-powered blemish removal on mobile, along with increasingly sophisticated distraction-removal, masking, culling, and generative tools.

There isn't one meaningful "Lightroom edits 100 photos in X seconds" number because performance varies dramatically based on hardware and what is actually being processed.

Copying an exposure preset across 100 photographs is very different from running AI Denoise, complex masks, or Generative Remove 100 times.

But Lightroom's workflow advantage is enormous.

You're still working inside the catalog.

You're still working with your RAW files.

Your edits remain adjustable.

And you can change one slider rather than asking the system to regenerate the result.

Aftershoot

Aftershoot is built specifically around high-volume photographer workflows.

It can learn a photographer's editing style and then make individual exposure, color, white balance, crop, masking, and other decisions across a session.

Independent testing by Shotkit in late 2025 reported approximately three minutes to process 156 RAW photographsusing Aftershoot on a local SSD. The reviewer noted that skin tones remained natural and that the edits adjusted intelligently from image to image rather than simply applying an identical preset.

That translates to roughly 1.15 seconds per photograph in that particular test.

An even larger real-world Fstoppers experiment in April 2026 ran approximately 2,600 wedding RAW files through an AI-heavy Aftershoot workflow. On a MacBook Pro, the automated workflow finished in about 52 minutes, with only around 30 minutes of actual hands-on photographer time required for reviewing and refinement.

Those tests are not identical to our ChatGPT blemish-removal test, but they demonstrate how aggressively dedicated software is optimized for volume.

Imagen

Imagen approaches the problem somewhat differently.

Rather than giving every photograph the same preset, it learns patterns from a photographer's previously edited work and attempts to make similar decisions based on each new photograph.

A 2026 Shotkit evaluation described processing at approximately 0.33 seconds per image, putting a 1,000-image wedding gallery at roughly five and a half minutes of processing.

Imagen itself currently advertises processing 1,000 photographs in minutes and says its automation can reduce editing time by up to 96%. Those are company claims rather than controlled independent benchmarks, but they illustrate exactly what the product is designed to accomplish.

Again, Imagen isn't necessarily performing the same generative facial retouch that we asked ChatGPT to perform.

But for the bread-and-butter work of making 1,000 wedding photographs look consistent, it is operating on an entirely different scale.

Evoto

Evoto may actually be one of the more interesting comparisons to our ChatGPT experiment because its tools are specifically designed around portrait retouching.

Its portrait system includes blemish removal, skin retouching, eyes, hair, facial adjustments, teeth, makeup, and other targeted controls.

Those adjustments can then be synchronized across multiple portraits.

That is much closer to the actual work we asked ChatGPT to perform.

The difference is workflow.

Evoto gives the photographer sliders, presets, masks, subject detection, synchronization controls, and batch tools.

ChatGPT gives the photographer a text box.

And strangely enough, that text box may be ChatGPT's biggest advantage.

The Speed Comparison

If we reduce everything to rough throughput, the difference becomes obvious.

That table makes ChatGPT look slow.

And compared strictly with professional batch software, it is.

Imagen's roughly 0.33-second processing figure is dramatically faster than our 8–10-second effective ChatGPT batch result.

Aftershoot was also substantially faster.

But this still isn't an entirely fair race.

Imagen and Aftershoot are making intelligent photographic development decisions.

ChatGPT was removing actual blemishes, interpreting which facial details should remain, rebuilding those areas, adjusting warmth, and producing a new finished image based on a conversational request.

It is a fundamentally different kind of edit.

Where ChatGPT Still Falls Behind

The biggest weakness isn't actually speed.

It's control and consistency at scale.

A professional photography application lets you know exactly what happened.

Exposure increased by 0.4.

Highlights dropped 20%.

Temperature increased 300K.

A mask was placed over the subject.

Skin smoothing was set to 15%.

Those decisions can be adjusted afterward.

With ChatGPT, the instruction might simply be:

Make the portrait warmer but keep the skin natural.

The result may look excellent.

But you don't receive a Lightroom Develop panel showing every decision that created it.

There is also the generative nature of the workflow to consider.

OpenAI specifically warns that selections are not always precise and that image edits can extend beyond the region you've selected.

That is exactly why preservation matters so much in photography.

A professional photographer may intentionally leave a mole, wrinkle, freckle, scar, birthmark, strand of hair, or subtle shadow untouched.

An AI model may decide that the feature looks like something that should be corrected.

Our High setting was much better about this.

But dedicated photographic tools still give the photographer considerably more direct control.

Where ChatGPT Got Hung Up

The testing was not completely smooth.

One of the more unexpected problems came from ChatGPT's content safeguards. In one portrait, the subject was wearing a top that the system appeared to interpret as potentially exposing too much skin. Even though the photograph itself was not nude, ChatGPT refused to perform the requested edit.

For a photographer, that creates an important limitation. A completely legitimate portrait, boudoir image, fashion photograph, maternity session, or even certain types of clothing could potentially trigger a safeguard and prevent an otherwise simple retouch.

We also ran into some strange behavior when attempting batch edits at the lower reasoning levels.

Instead of returning each photograph as its own edited image, ChatGPT occasionally combined multiple uploaded photographs into a single image. That obviously is not useful if you are trying to process a group of client photographs individually.

There were also instances where ChatGPT changed things we never asked it to change.

One of our original photographs was black and white, but during the batch-editing process ChatGPT added color to it. The request was simply to clean up blemishes and add warmth where appropriate, not to reinterpret a black-and-white photograph as a color image.

These issues became less common as we moved to the higher reasoning level, but they highlight an important difference between ChatGPT and dedicated photography software.

Lightroom expects a black-and-white photograph to remain black and white unless you tell it otherwise. A batch preset generally treats each photograph as an individual file. ChatGPT is interpreting the entire request, the images, and what it believes you want as an end result.

Sometimes that interpretation is impressive.

Sometimes it makes a decision the photographer never asked it to make.

For professional work, that means the results still need to be reviewed carefully. At this stage, we would not upload a large client gallery, walk away, and assume every photograph came back exactly as intended.

The technology is capable, but it still needs supervision.

Where ChatGPT Is Surprisingly Good

There is another side to this.

Most professional software expects photographers to understand the language of photo editing.

• Exposure.
• Contrast.
• Clarity.
• Texture.
• Frequency separation.
• Mask feathering.
• Healing.
• Curves.
• HSL.

ChatGPT doesn't require that knowledge.

You can simply say:

Keep her freckles, remove the temporary blemishes, warm the image slightly, keep the background the same, and don't change her face.

That is an entirely different user experience. Like these photos in this segement, the model has a bandage over his eye, we ask for chat gpt to remove the bandage then carry out the prompt and it did it easily.

OpenAI actually recommends this type of specific instruction when editing photographs, telling users to clearly identify both what should change and what should remain untouched.

For someone who isn't a professional retoucher, that may be incredibly powerful.

And even for a professional photographer, there are times when explaining the desired outcome might be faster than manually building the edit.

So, Would We Edit a Wedding in ChatGPT?

No.

At least not yet.

If we came home with 3,000 RAW photographs from a wedding, Lightroom combined with a dedicated AI workflow such as Aftershoot or Imagen makes far more sense.

Those platforms were built for that job.

They offer catalog management, RAW processing, synchronized adjustments, culling, presets, personal editing profiles, metadata, consistent exports, and workflows capable of handling thousands of photographs.

ChatGPT currently feels much better suited to selected images rather than entire shoots.

A photographer could finish the bulk of a wedding in Lightroom and then bring five hero portraits into ChatGPT for a very specific cleanup.

A portrait photographer could retouch ten client selections.

A photographer could clean a distracting element from a website hero image.

A small business owner who doesn't know Photoshop could repair a photograph simply by explaining what looks wrong.

That is where the interface becomes incredibly compelling.

Our Final Results

If we judged only the three ChatGPT reasoning levels we tested, the winner was easy.

Lower was usable.

The edits would probably fool an untrained eye, particularly at smaller viewing sizes, but excessive smoothing and weaker blending gave some photographs a noticeably amateur retouching quality when examined closely.

Medium was better.

Texture retention improved considerably and the results generally looked more realistic, but small inconsistencies occasionally revealed where an edit had taken place.

High was the clear winner.

It provided the strongest blending, the best retention of natural facial details, and the most restraint.

It didn't simply make the face cleaner.

It made better decisions about what actually needed cleaning.

That was the setting that most consistently felt like someone with photography experience had edited the image.

And that's an important distinction.

Is ChatGPT Worth Using for Photo Editing?

Yes.

But we wouldn't cancel Lightroom.

ChatGPT is not currently the fastest option for processing hundreds or thousands of photographs.

Dedicated AI platforms are substantially faster, have mature batch-processing systems, work within established photography workflows, and provide photographers significantly more control.

But ChatGPT showed us something those speed comparisons don't completely capture.

Conversational photo editing actually works.

We uploaded a photograph.

We explained what we wanted in normal language.

And at the High reasoning level, the finished result was often good enough that it felt closer to professional retouching than to an obvious AI filter.

That's impressive.

The bigger story may therefore not be whether ChatGPT replaces Lightroom.

It probably shouldn't.

The bigger story is that the interface between photographers and editing software is changing.

For decades, photographers learned how to tell software which tools to use.

Move this slider.

Create this mask.

Clone these pixels.

Heal this blemish.

Select this subject.

Now we're beginning to simply tell the computer what the photograph should look like.

Keep her skin natural.

Remove the temporary blemishes.

Don't touch her freckles.

Warm the photograph slightly.

Leave everything else alone.

Dedicated photography software still wins on speed, scale, consistency, and control.

But ChatGPT is beginning to win at something completely different:

understanding what you meant.

And after testing it at multiple reasoning levels, High was the first time we could genuinely see how that might become part of a photographer's professional workflow.