Shot metadata is structured information that describes how a shot looks and feels. It covers camera position, lens choices, framing, motion, lighting, time, and narrative intent. For AI workflows, these fields guide prompt construction, enforce consistency across panels or cuts, and make revisions measurable.

Use short, consistent terms. Keep a master glossary so teams and prompts match.

Recommended fields:

• Shot type: EWS, WS, MS, MCU, CU, ECU, OTS, POV, Two-shot, Crowd
• Camera angle: eye-level, low-angle, high-angle, top-down, dutch/tilt, worm’s-eye
• Camera distance: extreme-wide, wide, medium, close, macro
• Lens/focal: 18mm ultra-wide, 24mm, 35mm, 50mm, 85mm, 135mm; macro; orthographic
• Aperture/DoF: f/1.8 shallow, f/2.8, f/4, deep focus; bokeh quality
• Movement: static, pan-left, pan-right, tilt-up, tilt-down, push-in, pull-out, dolly, crane, handheld, track, orbit
• Framing/composition: rule-of-thirds, centered, symmetrical, leading-lines, golden-angle, headroom-tight, negative-space-left/right
• Subject/blocking: 1p, 2p, 3p; stage-left/center/right; foreground/midground/background elements
• Lighting: key style (soft/hard), ratio (2:1, 4:1), direction (top/back/rim), practicals, volumetrics
• Color/time: warm 3200K, neutral 4500K, cool 6500K; dawn, golden hour, noon, dusk, night, neon
• Mood/tone: serene, tense, melancholic, heroic, ominous, whimsical
• Aspect & scale: 1:1, 3:2, 4:3, 16:9, 21:9; resolution target
• Duration & rhythm (video): 48f @24fps (2s), hold, whip, match-cut
• IDs: scene, shot, panel, take, version

Store as short tokens for reuse and sorting.

• Single images: emphasize shot type, angle, lens, framing, lighting, mood, aspect.
• Video shots: add movement, duration, transitions, continuity notes, seed/latents strategy.
• Comic panels: add page/panel IDs, gutter/bleed, reading order, speech balloon real estate, eye-line continuity.

Keep a shared schema so teams can move concepts between formats without rework.

Attach metadata terms directly in your positive prompt and keep style cues separate from content.

Example (anime still):

• Content: teenage swordswoman on rooftop at dusk
• Shot metadata: MS, low-angle, 35mm, push-in feel, rule-of-thirds, rim light, 4500K cool-warm mix, wind in hair, heroic mood, 16:9
• Style controls: cel-shaded, clean linework, soft gradients, limited palette

Combined prompt: "teenage swordswoman on a city rooftop at dusk; MS, low-angle, 35mm, rule-of-thirds, rim light, 4500K cool-warm mix, heroic mood, 16:9; cel-shaded, clean linework, soft gradients, limited palette"

Tips:

• Put shot tokens early to influence layout.
• Use consistent abbreviations across a project.
• Keep movement words in video-specific prompts (pan, dolly) or in guidance tracks if supported.

Use negative prompts to protect framing and lens intent:

• Avoid: fisheye, extreme distortion, dutch tilt, blown highlights, clipped blacks, busy background
• Layout guards: no cropping face, no cut-off hands, preserve headroom, avoid tangent overlaps
• Style guards: no hyper-real pores, avoid photoreal grain (for cel-look)

These reduce unwanted drift when the model improvises.

Keep metadata portable:

• Image EXIF/XMP: write a ShotMetadata block (JSON string) into XMP dc:description or a custom namespace.
• Sidecar JSON/YAML: filename-shot.json with keys: scene, shot, panel, tokens, negatives, seed, sampler, aspect, color_temp, key_light, notes.
• Project sheets: a table for sequence planning and versioning.

Always store human-readable tokens plus machine fields (e.g., focal_mm: 35).

For multi-shot sequences:

• IDs: Use scene S01, shot SH003, panel P05, version v3.
• Seed strategy: lock per character or per shot depending on continuity needs.
• Look anchors: keep palette refs, model/Lora versions, texture refs, line weight notes.
• Continuity fields: eye-line direction, screen-left/right, costume state, prop positions.

Document changes: if lens shifts 35mm→50mm, note the motivation (tension ramp).

Create a single source of truth:

• Taxonomy file: lists allowed tokens and synonyms.
• Prompt template: content | shot tokens | style tokens | negatives.
• Validation: simple lints to flag off-list terms or missing fields.

This speeds onboarding and reduces mismatched prompts between artists.

Run this quick pass:

• Framing: headroom correct, no tangents, subject clear.
• Lens/angle: matches intent and continuity.
• Lighting: key direction and ratio are consistent.
• Color/time: matches scene plan.
• Aspect: set as required; safe areas observed.
• Metadata saved: IDs, tokens, negatives, seed, version.
• Notes: any deviations justified and recorded.

{ "scene": "S01", "shot": "SH003", "panel": null, "content": "teenage swordswoman on city rooftop at dusk", "tokens": ["MS","low-angle","35mm","rule-of-thirds","rim light","4500K","heroic","16:9"], "style": ["cel-shaded","clean linework","soft gradients"], "negatives": ["fisheye","dutch tilt","busy background"], "seed": 123456789, "model": "anime-diffusion-x", "lora": ["lineweight_v2"], "notes": "wind in hair; skyline bokeh" }