The brief was a premium activewear commercial: six women racing through a sunlit shopping street, with a camera moving from the front of the pack to its side. The women needed to look like competitors rather than a group out for a jog. Their faces, outfits, pace, light and the street all had to hold while the camera moved.
The first two takes belong to an early vertical approach built around one close starting image. The next four use a wider, more realistic opening frame and a separate identity sheet for the six runners. Each take also changes the model, the direction or both. That makes this a production comparison: the useful question is how the reference, prompt and model work together to make the requested shot.
Group 1: a weak reference, then two different directions
This vertical source frame has detail, but it is crowded and visibly synthetic before anything moves. For an activewear ad, that leaves the model starting from an image the viewer already questions. The two takes then make different choices: Seedance 2.5 is asked for a side-tracking race, while H3 Max is asked for a locked-off run-to-camera shot. Their outcomes show why the reference frame cannot be separated from the prompt and the model that must execute it.
Take 01
Dreamina Seedance 2.5
4.1 sec · 9:16
The group and street remain readable, but the action plays as a slow run rather than a competitive race. The prompt asks the camera to match the runners’ speed, which removes the background movement that would sell the pace.
Take 02
H3 Max
1.6 sec · 9:16
This is the stronger-looking result in the first group, but it is a different brief: a locked-off run-to-camera shot. Its prompt never asks for the side track, so image quality alone does not solve the requested ad shot.
Group 2: a stronger reference, then model and prompt choices
The second opening frame gives the shot an actual street, believable late-afternoon light, a clear route for the pack and space for a camera move. The identity sheet adds the six faces and outfits without asking the opening image to do every job at once. Takes 03-06 share those materials, but their models and prompts differ. That is where the comparison shows whether the generator can hold the scene and whether the prompt gives it a camera move and a race to perform.


Take 03
H3 Turbo
5.2 sec · 16:9
The crane-shot prompt specifies the front-to-side move, but the output loses the runners and their anatomy. The better reference cannot compensate when the model fails to execute the planned motion.
Take 04
H3 Max
5.2 sec · 16:9
The image is sharper and holds the group well, but this prompt only asks for a camera car tracking backward from the front. No sideward move is requested, so the runners still read as jogging.
Take 05
Dreamina Seedance 2.0 Fast
5.1 sec · 16:9
Its timed prompt and identity sheet bring the camera move and two lead runners closer to the brief. The motion stays too smooth, though, and the energy still feels more like jogging than a race.
Take 06
Dreamina Seedance 2.5
5.0 sec · 480p · selected take
This is the strongest result in the set. The detailed timed direction and Seedance 2.5 make the sideward movement, facial effort and rivalry read more clearly while the reference, identities and late light stay together.
| Take | What the inputs made possible | What still stopped the final ad shot |
|---|---|---|
| 01 · Seedance 2.5 | Early reference and side-tracking direction keep the street and group readable | Camera-matched pacing makes the race slow |
| 02 · H3 Max | Locked-off prompt produces a clear run-to-camera image | It is not the requested side-tracking shot |
| 03 · H3 Turbo | Better reference and a fully specified crane move | Low-quality motion and framing |
| 04 · H3 Max | Sharper image and group detail | Prompt never asks for the camera change; energy reads as jogging |
| 05 · Seedance 2.0 Fast | Timed side track and identity reference | Overly smooth movement; weak race energy |
| 06 · Seedance 2.5 | Reference, timed direction, camera, expressions and rivalry | 480p delivery resolution |
The reference, prompt and model are one production decision
The reference frame establishes what the model has to continue: the street, sunlight, racing formation, cast and depth of the shot. The identity sheet then lets the prompt protect the six women without importing its grey studio background into the street. That stronger foundation is why Group 2 begins with a believable race rather than a scene the viewer already doubts.
The prompt turns that foundation into a sequence. Take 04 is sharp but has no side track because its prompt never asks for one. Takes 03, 05 and 06 describe a timed front-to-side move, the competition between the two leaders and the changing light. The model then determines how faithfully it can keep the faces, bodies, pace and camera plan together. The selected Seedance 2.5 take is the best combination in this set, not evidence that any one of those inputs can carry the result alone.
Start with the right production inputs
See the selected sportswear take in my portfolio, alongside product films, cinematic stories and educational explainers. Bring a brief and we can discuss the visual direction for your own production.
Explore video production ↗