AI photo and video tools in paid client work: what holds up and what breaks

Where AI photo and video tools are reliable enough to bill for, the four cases where they still fail, and what a working stack costs a year.

By Todd Briggs · Updated 2026-09-26

Demos of AI photo and video tools are made on easy material. Paid work is not easy material. This guide separates the jobs these tools do reliably from the ones that still come back from a client, and puts a price on a realistic stack. The failure cases and the prices below are drawn from published field notes by a working London photographer, linked at the end.

Where AI tools earn their place

Four jobs are genuinely solved. Culling a shoot down to a shortlist, reducing noise on files shot at high ISO, upscaling for print, and cutting a subject out of a background. All four are repetitive, measurable, and forgiving of a small error. A tool that sorts 2,000 frames into keepers and rejects saves hours on every job and costs less than an hour of billed time per month.

Fabric is the hardest case

Silk and satin lose their highlights when a retouching model smooths them, and the material stops reading as the material it is. Tulle and chiffon are half transparent, so automatic selection takes part of the background with the garment. Anything sold on its texture is the wrong place to let a slider run at full strength.

Hair, hands and anything that overlaps

Fine hair against a bright background merges into a solid edge. Every background tool on the market still does this. Hands are worse: rings, interlocked fingers and anything that overlaps another object confuse the selection, and the result is a frame a client notices immediately. Both cases are faster to cut by hand than to fix after the fact.

Faces are the expensive failure

Skin smoothing removes texture at 100 per cent zoom, which is the zoom at which a client inspects their own portrait. The failure is expensive because it is invisible on a phone screen and obvious in print. The practical rule is to run face work at a reduced strength and mask it back over the eyes, the lips and the hairline.

Volume changes the answer

For 30 portraits, a person still wins on skin texture and judgement. For 300 frames from a brand day, consistency matters more than any single frame, and a model applying the same decision to every image beats a tired retoucher. The hybrid most professionals settle on is AI for the pass across the set, then manual work on the frames that will be printed or published.

Video is two jobs, not one

Generating video from a still and cleaning up video that was filmed are separate problems. Image to video tools produce short clips with good motion and no reliable control over a specific face or product. Restoration and upscaling tools work on footage you already own and are predictable enough to bill for. Mixing the two in one brief is where most disappointment comes from.

What a working stack costs

Published vendor prices collected in 2026 put a lean stack at about 20 US dollars a month, a working stack at about 40, and a full stack with video credits at about 70. A free stack also exists and is not a toy: open source upscaling and face restoration cover the two jobs most people pay for first. The table below lists the tools behind those figures.

The check before files go out

Four checks take about two minutes and catch almost every visible failure. View at 100 per cent and look for smoothed skin texture. Check the hair edge against the brightest part of the background. Count the fingers. Compare one fabric highlight against the original frame. If a frame fails any of the four, fix that frame by hand rather than re-running the tool.

What to tell a client

Say which parts of the work were automated and which were done by hand. Clients rarely object to AI culling or noise reduction. They object to discovering after delivery that a face was altered without being told. A line in the brief covering this costs nothing and prevents the one argument that loses a repeat booking.

What to test before you buy

Run your own worst frame through the free tier: the one with fine hair, a patterned fabric and a hand across the body. Every tool handles clean studio portraits well, so a clean portrait tells you nothing. Keep the tool that fails least on your hardest material, and pay annually only after a month of real jobs.

Job	Tool	Published price
Culling	FilterPixel	14.99 a month billed annually, 180 a year
Culling	Optyx Pro	6.99 a month billed annually, free tier available
Upscaling	Let's Enhance	9 a month billed annually for 100 credits
Upscaling	Upscayl	Free, open source
Noise and detail	Topaz web image plan	12 a month at the time of checking
Background	Photoroom Pro	7.50 a month billed yearly
Video	Runway	12 a month billed annually, free tier with 125 credits
Editor	Luminar Neo	129 pounds once, perpetual desktop licence

Further reading

• What AI retouching breaks, with the fix for each — field notes by photographer Mariya Harley
• AI retouching against manual retouching: an honest split (Mariya Harley)
• What an AI photo stack costs a photographer per year (Mariya Harley)
• AI video tools a photographer can actually use (Mariya Harley)

Tools mentioned

Related guides

Related reports

FAQ