A Team Page Fails on Mismatch, Not on Quality
Look at an about page that feels unprofessional and you will usually find that no single photo on it is bad. One was shot outdoors at golden hour, one under an office ceiling light, one is a cropped wedding photo, and one is a genuine studio portrait. Each is fine alone. Together they say the page was assembled from whatever people had, and that impression comes from the differences between the photos rather than from anything inside any one of them.
Use this without the search next time. Prathom Workbench puts Prathom's tools in your toolbar.
Add to Chrome — freeDrop an image here, or click to browse
Up to 25 MB. Uploaded temporarily, deleted after processing.
What it does
- One fixed spec applied to every upload
- Mid gray background with a gentle falloff
- Key light from the front left, eyes on the upper third
- The spec is published, so it can be repeated later
How to use AI Headshot Studio
- 1
Run every photo in the set through this page
The value is in applying one specification to all of them. A set where three were processed and one was not is still a mismatched set.
- 2
Keep the sources as similar as you can first
Framing, background, and light are handled here. Camera, focus, and resolution are not, and those differences survive the pass.
- 3
Compare the outputs against each other, not against their sources
Lay all the results side by side. The question is whether they look like one sitting, which is a question only the group can answer.
How it works
Each upload is scaled to about one and a fifth megapixels and repainted at a moderate-to-high denoise against one fixed instruction that does not vary by image: head and shoulders, eyes on the upper third, soft key light from the front left, mid gray background with a gentle falloff. Only framing, background, and light are named as changeable. The person is named as fixed.
The negative prompt is unusual for this catalog in that its first three entries are about the set rather than the picture — an inconsistent background, a different light direction, a changed crop. Those are only failures because another photo exists to disagree with.
Consistency is a property of the group
A single portrait cannot be inconsistent. This is obvious once stated and it changes how you evaluate the output.
Judging each result against its own source, the way you would for any other tool here, tells you almost nothing about whether this page worked. A run can be faithful to every individual upload and still produce twelve pictures that look like twelve different afternoons.
So the check is a group check. Put the outputs in a row and ask whether they look like one sitting. If one background is a shade darker or one key light comes from the other side, that is the defect, even though nothing is wrong with the picture on its own.
What converges and what does not
Being specific about this is more useful than a promise that the set will match.
Framing converges well, because head and shoulders with the eyes on the upper third is a geometric instruction with little room to interpret. Background converges well for the same reason: mid gray with a gentle falloff is close to a measurement. Light direction mostly converges, though a source lit hard from the right resists being told the key is on the left, and that resistance shows.
Skin and fabric rendering converge least. Those come from the source, and a set shot on six phones has six noise profiles, six color responses, and six degrees of sharpness. Nothing in an instruction reaches that. Expect a set that plainly belongs together and does not pretend to be one camera.
Why the spec is written down
Most tools in this category hold their style inside a preset, which is fine until you need the thirteenth photo.
Six months from now, someone joins, and their portrait has to sit next to eleven others without looking like an addition. That only works if the instruction is the same instruction. A preset that has since been improved will produce a better photo that does not belong.
Publishing the spec on the page is the commitment: mid gray, gentle falloff, key from the front left, head and shoulders, eyes on the upper third. If it ever changes, the sets made under the old one are on record, and you can decide whether to reprocess rather than discovering the drift on a live page.
Publication gate
This page ships once the workflow has been run against a set of at least six portraits from different cameras and lighting conditions, plus one added weeks later, with background tone sampled at matching points across every output and the key light direction confirmed identical in each.
Examples
Six staff photos from six different phones
The intended case. Background and framing converge tightly, and light direction converges well. Skin rendering still carries some of each camera's character.
One new joiner, months later
The reason the spec is written out on this page. The instruction does not drift between runs, so a photo processed later still belongs with the earlier ones.
Frequently asked questions
Will the outputs be identical in style?
Close, not identical. Each run is an independent sample with its own seed, so the gray behind one person will not be pixel-for-pixel the gray behind another. Background tone and framing converge tightly because they are stated plainly; the fine texture of skin and fabric varies more, because it comes from the source as much as from the instruction.
What if I need a genuinely identical backdrop?
Use the studio background page. It cuts the subject out and composites them onto a generated sweep with no model involved, so the same backdrop is byte-identical every time and your subject's pixels are untouched. It cannot restyle framing or light; this page can. That is the whole trade.
Why publish the specification on the page?
So it can be repeated. A house style is only useful if the twelfth photo matches the first, and that means the instruction has to be stable and knowable rather than a hidden preset that may be tuned later. The spec here is mid gray with a gentle falloff, soft key light from the front left, head and shoulders, eyes on the upper third.
Does this fix photos taken at very different resolutions?
It normalizes them to a common size, which is not the same as making them match. A sharp twelve-megapixel portrait and a soft crop from a group photo will end up the same dimensions and still read differently, because one has facial detail to preserve and the other does not.
Is the AI Headshot Studio tool free, and do I need an account?
Free, and deliberately account-free. Prathom has no login anywhere on the site, so there is no usage counter attached to you and no upgrade prompt waiting at the end of the job. The practical limits are the workflow's own: an image up to 25 MB going in, one queued GPU job at a time, and a result that stays available for thirty minutes.