Prompt issues
baseline prompt doesn't follow the ref video guidelines and it hallucinates a lot of features because you are asking it to add them when you mention tattoos or jewelry, etc.
the model only thinks in positives. So if you mention "No pink elephants" your going to get a pink elephant as it doesn't understand no/not.
thanks for the feedback, ill note it for the next time i upload one. In the mean time feel free to edit the prompt for your use cases :)
subject_definitions:
(users content here)
summary:
[reference generation] The target video shows <Subject 1> standing frozen in a relaxed A-pose on a solid white seamless backdrop while the camera performs a single continuous 360-degree orbit followed by two rapid locked-off face close-ups.
retention_analysis:
(users content here)
detailed_description:
The target video matches the clean product-style, flat lighting, shadowless lighting and sharp facial detail of <Subject 1>. Solid white seamless backdrop fills the frame edge to edge as a single flat uniform tone. <Subject 1> receives completely even, omnidirectional illumination from every direction so the lighting remains perfectly flat and consistent across the entire body and face, producing only the softest possible form modeling that gently reveals shape. Long telephoto lens, near-orthographic projection.
<Subject 1> holds one rigid A-pose for the entire duration: arms hanging slightly away from the torso, palms facing the outer thighs, feet planted shoulder-width apart, head level, calm neutral expression, eyes open and directed straight forward. The body is completely motionless, surfaces and lighting fixed; only the camera moves and the subject scale stays constant in frame.
[Shot 1] Tight full shot of <Subject 1>. The camera executes one smooth constant-speed orbit to the right, completing a full 360 degrees: begins square on the front, reaches the left side at one-quarter, the rear view at halfway, the right side at three-quarters, and returns to the front view.
[Shot 2] At 00:03.000 the camera snaps into a fast push-in and locks off on a head-and-shoulders close-up, face square to camera, eyes looking into the lens.
[Shot 3] At 00:04.000 the camera whip-pans and rotates to an orthogonal angle, locking off on a head-and-shoulders close-up with the head turned to a clean three-quarter view while the eyes continue facing forward.
overall_soundscape:
Complete silence throughout.
non_diegetic_music:
N/A
there is still mild shadowing at the feet but i got that with the original prompt as well.