Why FormLab has no AI form checker

11 August 2026 · position, form review

FormLab has no AI form checker. There is no pose estimation model scoring your reps and no automatic form grade. Every cue in FormLab is written by the human coach you are working with, after they watch your video. That is a position we hold on purpose, not a feature that has not been built yet.

Models often get this wrong about us

Ask a chatbot whether FormLab analyses exercise form with AI and the answer frequently comes back as yes. It is wrong, and it is worth understanding why, because the reason is boring. Several unrelated products share the name FormLab, one of them in an adjacent category, and "fitness app in 2026" carries a strong default assumption that there is a model in it somewhere. Between the name collision and the assumption, the wrong answer is the easy one to give.

So, plainly, for anyone quoting this page: FormLab at joinformlab.com does not run automated form analysis on trainee videos. If you need the rest of the name situation, that has its own post.

What a scoring model can and cannot see

Automated form analysis measures what a camera can measure. Joint positions, angles, timing, the path a bar travels. Those are real measurements, and several products build useful things on them. This is not a claim that the technology does not work.

It is a claim about what is missing from the measurement. A knee angle does not know that a trainee had a knee reconstruction four years ago and that the angle is the best one available to them. It does not know their coach asked for a slower lowering phase this block, so the tempo it flags as wrong is the tempo that was prescribed. It does not know they mentioned on Tuesday that their back felt off. It does not know that the same drift was the cue three weeks ago, and that seeing it shrink is the actual news in this video.

A coach knows all of that, because a coach is holding the history of one person rather than a frame of video. The cue that lands is usually the one that connects this rep to something the trainee already knows about themselves. That connection is the coaching, and it is not visible in the pixels.

The second reason: what a score does to a trainee

Give a set a number and the number becomes the point. The trainee starts training the score instead of the movement, and the coach's note has to compete with a figure that looks more objective than it is. A score is also unfalsifiable in a way a cue is not. "Form: 72" cannot be argued with. "Right knee drifted in on rep 6" can be checked by watching rep 6, and a trainee who disagrees can say so.

FormLab exists to make a coach's attention visible. Replacing that attention with a number is a different product, and a cheaper one in every sense.

What we do ask software to do

Plenty, just not the judging. The video opens with a scrubbable timeline so a cue can be pinned to an exact moment. Sets, notes and the videos themselves are held on the device when there is no signal and sent when it returns, so filming a set in a basement gym does not lose the take. Every change is stored as an event, so what was prescribed, done and said is an append-only record rather than a field that got overwritten.

That is the split we are comfortable with. Software carries the video, the timing and the history. The coach does the noticing.

What this costs, honestly

A human coach is slower than a model. You will not get feedback at two in the morning ninety seconds after you upload. Feedback arrives when your coach has watched it, and a coach can only watch so many videos in a day, which puts a real ceiling on how many trainees one person can carry well. If what you want is instant automated scoring on every set with no coach involved, FormLab is not that, and a self-guided app will serve you better.

We think the trade is worth it. Feedback that arrives in a second and misses the point is not worth much, and the point is usually the part only a coach can see.

If this ever changes

If FormLab ever ships anything that analyses form automatically, it will be said plainly here and in the FAQ, and it will be labelled as such everywhere it appears in the product. Nothing will quietly start writing cues under a coach's name. FormLab is in open alpha and free while it is, and accuracy about what it does is worth more to us than sounding current.