Skill: POV Prompt Anchoring
What
Every panel prompt in a first-person POV sequence must explicitly state the camera identity, height, and angle — not just the first panel.
Why
Image generation models default to cinematic third-person camera. Saying "First-person POV" once in panel 01 and relying on chaining to carry it forward does NOT work. The model forgets POV by panel 3-4 and reverts to floating observer camera. Frame chaining carries visual style and environment, but NOT camera identity.
How
Every prompt must include:
- "First-person POV" — literal text, every panel
- Camera height — "5'10" eye height", "adult standing height", "looking down from standing"
- Camera angle — "looking forward", "looking DOWN", "looking STRAIGHT DOWN", "steep downward angle"
- What's at frame edges — doorframe, own hands, floor, bed edge (sells the embodiment)
Panel 01 (establishing shot):
"First-person POV photograph taken through the eyes of a tall adult male
standing in a hospital doorway. The camera is at 5'10" eye height.
The left and right edges of the wooden doorframe are visible at the sides..."
Chained panels (02+):
"Same scene. First-person POV, now stepped two paces forward through
the doorway into the room. The doorframe is no longer visible — we are inside..."
Body/hand panels:
"Same scene. First-person POV, the viewer glances down at their OWN body.
Camera points steeply downward. This is NOT a third-person shot — it is
what you see when you look down at your own legs while standing."
The "NOT a third-person shot" trick:
For panels showing the POV character's own body, explicitly say what it ISN'T. The model responds to negative instruction when the risk of misinterpretation is high.
Subject direction relative to camera:
- Characters moving AWAY from the viewer → "running AWAY from the camera, back partially visible"
- Characters looking AT the viewer → "looking UP toward the camera, tilted face, eyes meeting the viewer"
- Objects getting closer → "the bed is LARGER in frame now — we are closer to it"
Common Mistakes
- Stating "POV" only in panel 01 and hoping chain carries it (it doesn't)
- Using camera-language: "close-up shot", "dolly in", "camera tracks" (use body-language instead)
- Describing subjects running "toward the camera" when they should be running AWAY (POV is behind them)
- Body panels shown as third-person full-body shots instead of foreshortened own-body view
- Not specifying what's at frame edges (hands, doorframe, floor) — these are what sell first-person