skills/pixverse-fashion-video/references/production.md

# Try-On Production

Read `../../../skills-shared/ugc-production.md` with this recipe. This is an eight-beat
creator try-on, not a silent editorial model walk. Explicit user direction can request
that different treatment; otherwise preserve the complete structure below.

## Identity, Garment And Board

Use one accepted creator and the same canonical wearable description everywhere.
Lock silhouette, color, print/lettering, collar/neckline, closures, pockets, sleeves,
hem, hardware and realistic drape. Record known front/side/back evidence. Do not
claim a requested rear detail was shown by avoiding the turn; obtain a usable rear
reference when exact unseen construction matters. Accessories keep realistic weight,
scale and attachment; do not apply clothing-only handling to shoes, jewelry or bags.

For a new creator, make one product-free 3:4 reference with the requested appearance,
natural face proportions and non-distracting wardrobe. Resolve setting/wardrobe variety
once, not by regenerating identity in every shot. An existing accepted portrait is reused.

Each board: eight active portrait panels in one row; product reference first, creator
second, cleaned previous board third when applicable. The board is 21:9 with the
contain-fit geometry described in the shared contract. No generated captions or labels.
Run the required realism refinement before using it for motion.

## Board Progression

| Board | Role | Continuity and new information |
| --- | --- | --- |
| 1 | Canonical try-on | Before outfit → worn item → fit, fabric, turn, style, final look |
| 2 | Home tour | First shot continues board 1's last room; explore other rooms in the same home |
| 3 | Outdoor | Dry start; light rain from slot 2; droplets/surface sheen, dry hair; no invented waterproof claim |
| 4 | Home reflection | Seated return home, specific observations and settled conversational warmth |
| 5+ | Continuing loop | Alternate compatible established/new settings and real new observations |

Preserve explicit setting, weather and tone overrides. Rain changes surface appearance,
not fabric opacity, garment color, branding or unsupported performance claims. Board 4
can have fuller observations at the same speaking pace; never slow speech to fill it.
After board 1, the pre-wear outfit and shopping bag never return.

## Canonical Eight Slots

| Slot | Required beat | Action, camera and sound |
| --- | --- | --- |
| 1 | PRE_WEAR | Muted home base outfit; one closed plain kraft bag by handle or upright on a surface; item completely hidden. Selfie default, mid-event, visible speaking mouth. No opening, peeking, extraction or dressing. |
| 2 | WEARING | Already fully dressed after the hard cut; bag gone. Fixed full-body/three-quarter framing, one brief natural twirl revealing the back and returning front. Pause lip articulation only while back faces camera, resume immediately on return. |
| 3 | FRONT_POSE | Read the complete outfit front-on from head to foot, alive weight/pose change; distance or POV clearly differs from slot 2; on-camera dialogue. |
| 4 | TEXTURE_CLOSEUP | Fixed hand-free macro of fabric/cut. One passive breath, drape settle or light response; hands outside the crop and off the garment; same voice as VO, no speaking mouth. |
| 5 | TURN | Side/back fit in a distinct compact rotation with face returning toward lens; clear visible speech when face is seen. Do not repeat the full twirl. |
| 6 | DETAIL | Second fixed hand-free macro on a different feature: collar, hem, seam, sleeve or hardware. Single passive movement; VO only. |
| 7 | STYLE_POSE | Different room of the same home, complete outfit, purposeful pose/step and visible speech; no bag. |
| 8 | FINAL_LOOK | Active satisfied wrap in that room/continuation, visible speech, complete thought, no appended CTA. |

Every panel opens during its event, not before the actor starts. Aim for clear changes in
POV, distance and action between neighbors; tight/mid/wide should each appear at least
twice. Keep the full-fit and macro anchors when the generic alternation examples conflict.
One hero garment; no duplicate jacket behind a wearing creator. Explicitly assign both
hands in each slot. A phone arm consumes a hand; two-handed adjustments use a fixed camera.

For boards 2+, retain two different hand-free macros at 4/6 and visible speech at
1/2/3/5/7/8 while adapting the other beats to that board's role. No repeated introduction,
bag, on-screen costume change or board-1 twirl. Keep face, hair and item exact.

## Timing

| Clip | C1 | C2 | C3 | C4 | C5 | C6 | C7 | C8 |
| --- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| 4s | .5 | .5 | .5 | .5 | .5 | .5 | .5 | .5 |
| 6s | .75 | 1 | .75 | .5 | .75 | .5 | 1 | .75 |
| 8s | 1 | 1.25 | 1 | .75 | 1 | .75 | 1.25 | 1 |
| 10s | 1.25 | 1.5 | 1.25 | 1 | 1.25 | 1 | 1.5 | 1.25 |
| 12s | 1.5 | 1.75 | 1.5 | 1.25 | 1.5 | 1.25 | 1.75 | 1.5 |
| 15s | 1.9 | 2.3 | 1.9 | 1.4 | 1.9 | 1.4 | 2.3 | 1.9 |

Write cumulative ranges summing exactly to D. Small .25–.5s redistribution may protect
an action or phrase; every shot remains at least .5s. The 15s source weights above total
15s. For other durations interpolate the role weights, then correct the final sum. Later
boards are more even, retaining shorter macros and enough time for their real words.

Write all eight `Cut` sections and seven `Hard cut to.` markers. The 12s boundaries are
0, 1.5, 3.25, 4.75, 6, 7.5, 8.75, 10.5, 12. Do not reduce this to three body movements
inside one uncut shot. A locked camera still contains a moving, speaking person.

## Speech And Motion Prompt

Use the shared prompt structure and four-to-ten-sentence moving descriptions. Carry
the personal-want/arrival mini-story only as a fictional creative premise or genuine
user-supplied experience, not a fabricated real testimonial. Follow with specific fit,
fabric and room observations; don't fill six lip-sync shots with one vague slogan.
In the Audio section explicitly allocate 1/2/3/5/7/8 to on-camera native dialogue,
4/6 to the same voice off camera, and write each exact phrase once.

At the opening of board 1, one short natural laugh/gasp can precede the first meaningful
words, within the .4s onset. No theatrical scream unless asked. The main peak is the
new outfit; a second can be the style landing. One recovered glance or small self-correction
and a small playful beat may occur away from peaks. Micro-behaviors are bodily changes
inside a coherent action, not a request to perform five separate actions in 1.5 seconds.
In macros, show multiple observable consequences of one breath/settle rather than adding
touching, rubbing or operator hands to satisfy a generic micro-beat count.

Keep normal real-time phone capture, natural neutral light and deep focus. No golden-hour
editorial treatment, cinematic camera movement, beauty smoothing or slow motion by default.
Specify accurate skin and cloth detail without slimming faces or inventing materials.

## Transitions And Ending

The pre-wear → worn boundary always hard-cuts; never animate changing or bag extraction.
On the closing board, if C6→C7 is fixed macro → selfie, there is no state/location jump,
and C7 lasts at least 3s, the source-style default is one Pick-Up transition there:
reach past the lens, lift the frame to arm's length over about 1.5s, voice uninterrupted,
lip-sync engages as the face arrives. Otherwise keep the hard cut. A user request for
classic cuts wins. Other boards use this device only for an explicit one-take treatment.
Never add both Pick-Up and Set-Down or spend the dialogue's whole time on the move.

The last board lands the resolution and ends alive, with no CTA tail. For a single
board, frame-match the opener's framing/gesture while keeping the worn outfit and no
bag. The source also offers a mid-phrase timing loop: use only when the brief permits
an intentional unfinished loop; a complete-dialogue requirement wins. For multiple
boards, end mid-gesture after the exact final word when complete speech is required.

## Required Checks And Prompt-Only Output

Check every macro, full fit, turn and 2–3 speaking frames. Verify actual cuts, identity,
hair, garment features, two hands, bag disappearance, no duplicate item, no unintended
text and the six/two speech split. A face-free macro is not a failed lip-sync shot.
Keep accepted clips and repair only the failed dependency through a new paid preflight.

If a rendered macro introduces hands or a speaking face despite a clean board,
do not mark the six/two speech split as passed. Crop the accepted board's actual
face-free, hand-free detail into an isolated control image; generate only that
macro with the crop as its visual input, without a portrait reference that can
reintroduce the creator. Preserve the full film's accepted native audio under
the replacement detail as VO. Keep frame-accurate boundaries and match color,
material and playback speed; never time-stretch speech or cover a required
on-camera line. Inspect the replacement and both joins. Do not regenerate the
whole film when only these detail pictures failed.

For prompt-only work deliver the complete eight-section video prompt, exact dialogue,
reference-role map and corresponding eight-slot text plan. Label a board that does not
yet exist as planned; do not generate it, invent an ID or claim it was visually inspected.