The hard rule first: never generate an exercise demonstration. Not as a shortcut, not as a placeholder, not for a thumbnail. AI video models produce anatomically plausible but mechanically wrong movement with total confidence: a deadlift with a rounded lumbar spine, a squat with knees collapsing inward, a press with the shoulder in a position that causes injury. Someone will copy it. Form content is the one category in fitness where being wrong hurts a person.
Everything else is fair game, and there is a lot of everything else.
TL;DR
- Never generate: form demonstrations, technique, anything a viewer might copy with weight in their hands
- Film: yourself, your clients (with permission), your gym. This is also what converts
- Generate: challenge and program graphics, class schedule visuals, abstract motion backgrounds, ad variants
- A month of that: roughly 500 credits, about $5
- The thing that actually grows a fitness business is a real coach visible on camera, repeatedly
Why the form rule is absolute
An AI model has learned what a squat looks like from video, not what a squat does to a joint. It will reproduce the silhouette and get the mechanics subtly wrong in ways that read as fine to a general viewer and as alarming to anyone who coaches. Rounded backs under load, valgus knees, hyperextended elbows, a spotter standing in a position that could not actually catch the bar.
Publishing that under a coaching brand is both a liability and a credibility problem. Anyone qualified who sees it stops taking you seriously, and anyone unqualified might copy it.
Film demonstrations. A phone on a tripod, good side angle, one clean rep. It costs nothing and it is the content people actually save.
What to generate instead
Program and challenge graphics. Six-week challenge cards, phase headers, progression charts. Seedream 5 at 6 credits per image.
Class schedules and announcements. Typographic, on brand, updated weekly. Same 6-credit route.
Abstract motion backgrounds. Behind text: light through a window, dust in air, a slow drift. Wan 3.0 at 480p, 24 credits per 5 seconds, and it never needs to be sharper.
Ad creative variants. One real clip of you coaching, ten different treatments around it. This is where volume pays.
Transformation-adjacent graphics, carefully. Charts, timelines and data visuals are fine. Generated before-and-after bodies are not, and are explicitly against most platforms' ad policies.
What a month costs
Canon 8frame prices, a credit is $0.01 at pack rate:
| Item | Route | Credits |
|---|---|---|
| 30 schedule and program graphics | Seedream 5 at 6 | 180 |
| 6 background plates | Wan 3.0 480p at 24 | 144 |
| 3 promo clips with audio | Veo 3.1 Lite 8s at 54 | 162 |
| Music bed for reels, 1 minute | ElevenLabs Music at 81/min | 81 |
| Total | ~567 credits, about $5.67 |
Comfortably inside the $19 Starter plan's 1,000 credits.
The honest competitive point
In fitness the product is trust in a specific person's judgement about your body. Synthetic presenters and generated bodies work directly against that. The coaches growing fastest in 2026 are the ones posting unpolished footage of real sessions, consistently, with their face in it.
Use generation for the design layer that surrounds that footage, which is the part that makes a feed look like a business instead of a camera roll.
FAQ
Can I use AI for exercise demos if I check them? Checking each frame of generated movement for mechanical correctness takes longer than filming a rep. Film the rep.
What about generated before-and-after images? No. They misrepresent results and most ad platforms prohibit them outright.
Do I need video to sell coaching? You need to be visible and consistent. Video is the cheapest way to be both.
Film the movement, design everything around it. The 8frame canvas is free and unlimited, and generation is paid from $19/month.