Fitness coach prompts for a trainer who looks the same every session
A fitness channel or app needs one recognizable coach across dozens of short clips. Reference images keep the face, build and outfit fixed; the harder part is the movement itself, because AI video models still get exercise form and rep counts wrong. These prompts focus on moves the models handle well, talking cues and atmosphere shots, with notes on when footage is fine for motivation and when it is not safe for instruction.
What makes fitness prompts work
Shoot references of the coach in their workout clothes, full body from front, side and back, plus a face close-up. That is four of the five images RefVid takes per character. Body proportions matter more here than in most use cases, because every exercise shows the whole body.
Pick simple, slow, symmetrical moves: squats, a plank hold, a standing stretch, a lunge. Fast or complex moves such as burpees, kettlebell swings or anything with crossing limbs tend to break form or merge limbs.
Describe tempo and count loosely. "Two slow squats" is more reliable than "ten squats"; models do not count well. Use a side view for leg exercises so knee and hip angles are visible.
The examples call the coach Jordan; use the name you saved yours under. Clip length and aspect ratio are settings, shown here in brackets.
Exercise demonstrations
Prompt (6 s, 16:9): Jordan performs two slow bodyweight squats in a bright, minimal studio, arms extended forward for balance, feet shoulder-width apart. Side view, static camera at hip height, full body in frame. Even daylight from tall windows, light wood floor.
Why it works: a side view shows depth and back angle, and two reps fit comfortably in six seconds. Check the knees on every take before using it.
Prompt (7 s, 16:9): Jordan holds a forearm plank on a grey mat, body in a straight line, breathing steadily. The camera slowly moves from the side toward the head at floor level. Cool studio light, dark background.
Why it works: a static hold means only the camera moves, which keeps the body stable. This is a good format for any isometric exercise.
Prompt (6 s, 16:9): Jordan steps forward into a slow lunge, lowers until both knees are bent, pauses and steps back to standing. Three-quarter front view, static camera, full body. Bright gym with mirrors blurred behind.
Why it works: one rep with a pause is easy to read. If the back leg bends the wrong way, switch to a pure side view.
Coach talking to camera
Prompt (6 s, 9:16): Jordan stands in a gym holding a water bottle, looks into the camera and says: "Three more rounds. Slow on the way down, breathe on the way up." Medium shot, slight handheld movement, warm overhead lights, equipment out of focus.
Why it works: short cues are what coaches actually say between sets. With sound on, the model generates the voice with the picture, and a short line keeps voice and mouth in step. RefVid has no voice cloning, so the voice can vary between clips.
Prompt (8 s, 9:16): Jordan sits on a bench after a workout, towel over one shoulder, and talks calmly to the camera about recovery, nodding while speaking. Static camera, soft late-afternoon light through a window.
Why it works: a calm talking shot without a written line suits a voiceover you record yourself, which keeps one voice across a whole program. Keep sound off and add your audio in an editing app.
Outdoor sessions and atmosphere
Prompt (7 s, 16:9): Jordan jogs along a riverside path at sunrise, steady pace, the camera tracking alongside at chest height. Mist on the water, low golden sun behind, soft breathing sounds.
Why it works: a steady jog in profile is a reliable motion and a good channel intro. Keep the pace easy; sprinting adds blur and limb errors.
Prompt (8 s, 16:9): Jordan stands on a mat on a wooden deck overlooking the sea and moves slowly from a standing reach into a forward fold, holding the stretch. Static wide shot. Calm early-morning light, gentle wind.
Why it works: slow stretches suit the model's strengths. Simple poses are reliable; advanced balances often drift.
Clip length and form
RefVid clips run 3 to 15 seconds on Kling 3.0 Omni. Five to eight seconds fits one or two slow reps; a longer clip fits a hold or a short sequence, but more time also means more chances for form to slip. For comparison, Veo 3.1 reference clips are always 8 seconds, according to Google's documentation as of October 2026.
Plan a workout as many short clips rather than one long one, and keep the coach's reference set identical across all of them. RefVid keeps the coach as one saved character, so every clip in a program uses the same face and build.
Frequently asked questions
- Is AI-generated exercise footage accurate enough to teach form?
- Not reliably. Joint angles and rep counts in AI video can be wrong. Use AI clips for motivation and atmosphere, and have a qualified coach review any clip used for instruction.
- Which exercises work best in AI fitness video?
- Slow, simple, symmetrical moves: squats, lunges, planks and stretches. Fast or complex moves with crossing limbs, such as burpees or kettlebell swings, break most often.
- How do I keep an AI fitness coach's body consistent?
- Use full-body references from front, side and back in the same workout clothes, plus a face close-up, and keep that set fixed for every clip.