Skip to content

1173206772/photo-skills / skills/hk-multiverse-photo

hk-multiverse-photo

Verified

Transform supplied Hong Kong, city, street, vehicle, skyline, or harbor photographs into bold multiverse-collision still images that preserve source geometry while combining real photography, flat 2D animation, and dimensional 3D/cel rendering. Use for Spider-Verse-inspired city edits, cyberpunk universe collisions, distinct style fields, phase-wheel sectors, vehicle timeline portals, tram interchange scenes, Douyin/9:16 batches, selective frame revisions, coherent multi-image series, or original HK//VERSE archive annotations. Generate an original visual system without copying characters, official logos, or film frames. Do not use for video unless the user explicitly asks for motion.

FileKindSize
SKILL.mdskill md11.7 KB
agents/openai.yamlconfig260 B
references/annotation-system.mdmarkdown13.7 KB
references/collision-layouts.mdmarkdown30.3 KB
references/universe-atlas.mdmarkdown9.3 KB
scripts/make_annotation_svg.pyexecutablescript14.0 KB
scripts/validate_image.pyexecutablescript8.1 KB

references/collision-layouts.md

View on GitHub
# Collision Layouts: Source-Hook Decision Matrix

Use this reference to choose the collision geometry for a Hong Kong city, street, skyline, harbor, vehicle, or tram photograph. The source photograph is the composition. Multiverse effects must grow from a visible source hook—perspective, transport direction, wheel phase, rails, wires, waterline, reflection, or skyline rhythm—rather than being pasted over it.

## Non-negotiable composition contract

Budget the readable scene by **role**, not by medium. The three role budgets are mutually exclusive and add to 100%:

- **PRIMARY universe — 45–60%:** one unmistakable world owns the largest connected share of foreground and background. Its palette, material, atmosphere, and rendering logic must read as the governing reality rather than a local effect.
- **PHOTO anchor — 20–35%:** an intentionally natural photographic region with the source exposure, weather, lens character, grain, reflections, shadows, and local color intact. It proves what the place and subject really look like.
- **INCOMING universes — 15–30% total:** two to four unequal fields, each with a distinct silhouette and rendering system. Their combined coverage, not each field, must remain within the range.

The percentages are target visible-area budgets, not opacity values. Each matrix row gives an exact total of 100%. If fields overlap, count the topmost readable state; count a seam or shard inside the role that visually owns it rather than as extra coverage. Do not trade away the hierarchy: PRIMARY remains largest, PHOTO remains a clear anchor, and all incoming fields together remain subordinate to PRIMARY.

**Source geometry is 100% locked across PRIMARY, PHOTO, and every INCOMING field.** Coverage outside PHOTO may be re-rendered, re-lit, phase-shifted, or given new material and atmosphere, but it may not move or redesign the photographed structure. Lock:

- the dominant subject's silhouette, pose, facing direction, and placement;
- faces, hands, readable route identifiers, vehicle identity, and distinctive signage;
- the original horizon, waterline, curb line, rail gauge, and primary vanishing point;
- camera height, focal-length cues, vanishing geometry, depth order, and occlusion order;
- at least one uninterrupted route from foreground to background so the original place remains navigable.

Every frame must visibly contain all three media languages: **natural PHOTO**, **flat 2D**, and **dimensional 3D/cel**. If PRIMARY is flat 2D, at least one incoming universe must be dimensional 3D/cel; if PRIMARY is dimensional 3D/cel, at least one incoming universe must be decisively flat 2D. Do not recolor PHOTO to “match” the universes. The collision gains force from their contrast.

A stylized photographic universe such as U02 NOIR counts as PRIMARY or INCOMING, never as PHOTO. The PHOTO budget means source-natural optical rendering without an alternate-universe material pass.

## Source-hook decision matrix

Choose the first row whose source hook is genuinely visible. Do not invent a Ferris wheel, rails, wires, water, or a vehicle merely to qualify for a layout.

| Source class | Qualifying visible hook | Primary collision layout | PRIMARY / PHOTO / INCOMING | Incoming count and mandatory media contrast | Required correspondence | Preferred seams | Reject the layout when |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Dense street canyon | Two façade walls or a façade-plus-elevated edge form a corridor around a readable street floor | **Street-canyon fields**: let one governing world span a façade, the upper canyon, and background while the road remains the recognition route | **52 / 25 / 23** | **3 incoming:** 9% flat 2D, 8% dimensional 3D/cel, 6% distinct material/render pass | Field boundaries follow façade planes, balconies, awnings, curb edges, and occlusion; all source planes stay registered | Cell fractures on wall planes; render-pass seams under awnings; a few angular space shards in the upper canyon | The image is an open plaza, a flat elevation, or has no stable corridor |
| Street, alley, tunnel, or arcade | One strong vanishing point plus clear near/mid/far occlusion layers | **Vanishing-point depth slices**: PRIMARY spans multiple depths; subordinate slices converge on the photographed vanishing point | **50 / 28 / 22** | **3 incoming:** 8% flat 2D, 7% dimensional 3D/cel, 7% contrasting render state | Slice widths shrink with perspective; people, poles, and vehicles retain exact depth and silhouette across passes | Perspective-tapered render-pass seams; sparse cell fractures; one deep volumetric breach near a mid/far transition | The perspective is orthographic, the vanishing point is ambiguous, or the subject depends on an unbroken flat façade |
| Harbor with Ferris wheel | Wheel hub, rim, spokes, sky, and some water are visible | **Ferris-wheel sector/phase collision**: PRIMARY phase owns the wheel/sky/sea system; incoming sectors remain angularly matched | **48 / 24 / 28** | **2–4 incoming:** default 11% flat 2D, 9% 3D/cel, 8% contrasting time/material state; split one field only when a fourth is useful | Every universe sector maps by angle to a matching sky ray and by reflected angle to a corresponding water wedge; hub, rim geometry, supports, shoreline, and waterline stay locked | Radial cell fractures; spoke-aligned render-pass seams; angular sky shards; shallow volumetric ellipses centered on the hub | The wheel is tiny, heavily occluded, cropped without a readable hub, or no sky/water counterpart exists |
| Road vehicle, taxi, bus, ferry, or train in motion context | A dominant vehicle has a legible longitudinal travel axis and enough surrounding space to cross-section it | **Vehicle-as-time-axis**: PRIMARY owns the continuous vehicle/world timeline; incoming states appear through uneven perpendicular wormhole apertures | **50 / 20 / 30** | **4 incoming:** uneven apertures of 5%, 7%, 8%, and 10%; at least one flat 2D and one dimensional 3D/cel | Apertures reveal visibly different states while vehicle silhouette, wheels/hull, windows, route marks, position, and shadow/wake stay registered | Uneven volumetric ellipses; local render-pass seams on body panels; a small number of angular shards at aperture exits | Vehicle identity is too small, there is no readable travel axis, or apertures would erase the PHOTO anchor |
| Hong Kong tram | Tram plus at least two of these: catenary wires, rails, switch points, poles, paired tracks | **Tram interchange**: PRIMARY governs the surrounding transit world; wires and rails route incoming universes through the source-locked tram | **50 / 25 / 25** | **3 incoming:** 10% flat 2D wire-route field, 8% dimensional rail junction, 7% contrasting cel/material branch | Wire junctions pair with rail junctions; routes pass through real attachment points without moving or redesigning the tram | Wire-following cell fractures; rail-aligned render-pass seams; compact volumetric ellipses only at true junctions | Wires or rails are absent, invented topology would dominate, or the tram lacks a clear photographic anchor |
| Skyline plus water | A readable horizon or seawall separates a building rhythm from water/reflections | **Skyline/water mapping**: PRIMARY joins skyline, sky, and water; incoming cluster cells project above and reflect below | **52 / 23 / 25** | **3 incoming:** 10% flat 2D sky/water map, 8% dimensional 3D/cel cluster, 7% contrasting volumetric reflection state | Each selected tower cluster maps to one sky region and one water region; source tower contours and horizon stay fixed while wave offsets govern reflection distortion | Vertical render-pass seams hidden at tower gaps; water cell fractures; sparse angular sky shards; low-profile volumetric ellipses on water | Water is only a sliver, the skyline is blocked, or no stable horizon/waterline can be preserved |
| Broad city view or plaza | No stronger specialized hook, but three or more distinct architectural masses and open negative space are visible | **Mass-to-field city collision**: PRIMARY spans the largest connected mass and its background; incoming worlds inherit other masses and voids | **55 / 25 / 20** | **2 incoming:** 9% flat 2D and 11% dimensional 3D/cel | Collision cells inherit rooflines, setbacks, street edges, and foreground occluders; every mass retains exact placement and silhouette | Irregular cell fractures and a few angular space shards; render-pass seams at mass boundaries | The only available division is a repeated vertical split or the result would become a global filter |

When two hooks are equally strong, use one as the primary layout driver and the other only as a secondary echo. For example, a tram beside a harbor uses the rail/wire interchange as the structure; water may echo its phase colors, but it must not introduce a second independent skyline grid.

## Layout recipes

### 1. Street-canyon fields

Use this when the street is framed by vertical walls and the negative space between them reads as a canyon.

**Frame allocation**

- Give **52% PRIMARY** to one connected world spanning **30%** of the dominant façade and **22%** of the upper canyon/background. Re-render those source planes without shifting a single edge, window, sign, or depth relationship.
- Keep **25% PHOTO**: a **15%** continuous ground-and-subject route from the lower edge toward depth, plus **10%** of source-natural people, signs, street furniture, and façade detail.
- Add **23% INCOMING** as three unequal universes: **9% flat 2D** on the opposing façade, **8% dimensional 3D/cel** at one depth break, and **6%** in the overhead slot using a third distinct render/material system.

**Construction**

1. Trace the source planes: left wall, right wall, street floor, overhead sky, and any elevated bridge or awning.
2. Make PRIMARY visibly govern both a near plane and the background. The 25% PHOTO route remains the truth anchor, not the dominant coverage.
3. Let balconies, drainpipes, signs, windows, and awnings cross or occlude seams naturally. Do not cut every object at the boundary.
4. Give the 2D incoming field flat value blocks and decisive drawn contours. Give the 3D/cel incoming field volume, modeled thickness, and lighting that visibly occupies space while its supporting façade geometry stays registered.
5. Preserve the camera's original light and cast shadows in PHOTO. Universe fields may have their own light, but causal spill and seam contact must respect the locked source planes.

Keep the hierarchy readable: one 52% PRIMARY, one 25% PHOTO route, and three small incoming fields. A sequence of equally wide façade strips reads as a template rather than a collision.

### 2. Vanishing-point depth slices

Use depth, not screen width, to decide the slices.

**Frame allocation**

- Give **50% PRIMARY** to one governing render state joined by a continuous corridor through near, mid, and far depth. It must own the atmosphere and background, not appear as isolated bands.
- Keep **28% PHOTO**: approximately **16%** in a continuous foreground-to-vanishing route and **12%** on people, poles, vehicles, signage, or thresholds that prove the original depth.
- Add **22% INCOMING** as three irregular depth fields of **8% flat 2D**, **7% dimensional 3D/cel**, and **7%** in a third unmistakable material/render state.
- Use **four to six** irregular depth intervals overall; PRIMARY or PHOTO may recur across nonadjacent intervals, but the three incoming universes remain only 22% total.

**Construction**

- Project every seam toward the real vanishing point. Near seams may be broad; distant seams must compress.
- Derive depth changes from real thresholds: crosswalks, parked vehicles, arcade bays, lamp-post overlap, tunnel ribs, or façade setbacks.
- Use at least one partial occluder that stays PHOTO across two adjacent slices. This proves that the slices occupy depth rather than sitting on the image plane.
- Vary the slice grammar: one cell-fractured flat 2D pass, one dimensional 3D/cel pass fitted to the same depth geometry, one PHOTO pass, and the recurring PRIMARY. Do not alternate A/B/A/B mechanically or displace the photographed corridor.

### 3. Ferris-wheel sector and phase mapping

The wheel is a coordinate system, not a decorative portal.

**Frame allocation**

- Give **48% PRIMARY** to one phase system spanning approximately **14% wheel sectors**, **20% matching sky rays**, and **14% corresponding sea reflections.** It must read as the harbor's governing reality while every spoke, rim point, horizon, and reflection origin stays fixed.
- Keep **24% PHOTO** across the hub, selected rim arcs, supports, shoreline, waterline, and one coherent harbor/sky region. Preserve enough natural rim and support detail to prove the wheel's identity; re-rendered portions still follow the exact source circumference.
- Add **28% INCOMING** as two to four nonadjacent phase groups. Default to **11% flat 2D**, **9% dimensional 3D/cel**, and **8% contrasting time/material state**; split one group only when a fourth universe improves the scene. Each group's coverage includes its wheel sector, sky ray, and sea echo together.

**Sector rules**

1. Detect the hub and number the visible wheel sectors clockwise from the top. Use spoke angles from the source; do not replace the wheel with a perfect generic radial graphic.
2. Select **two to four unequal, nonadjacent incoming sector groups** within the 28% total. Keep at least one wide uninterrupted PHOTO sector between aggressive phase changes, and let PRIMARY occupy the large connected angular remainder.
3. Extend each selected sector outward as a matching sky ray. Its angular boundaries must continue the same spokes or their optically plausible extension.
4. Reflect that sector below the real waterline into a sea wedge. Reverse the vertical angle as a reflection would, then distort its edge and intensity with the photographed wave direction.
5. Keep phase identity consistent across all three zones: a cel-rendered wheel sector gets a cel-rendered sky ray and cel-like sea echo; a volumetric night-world sector gets spatial night sky and a luminous volumetric reflection.
6. Maintain the source phase relationship even when the wheel reflection is broken by waves. Do not assign unrelated colors or worlds to wheel, sky, and sea.

Record apparent image-plane angles in `phase_map`: set `0°` at 12 o'clock and increase clockwise. For hub `(hx, hy)` and point `(x, y)`, use `θ = atan2(x-hx, hy-y) mod 360°`. These are visible sector coordinates, not an inference about the wheel's physical rotation. Mirror the hub across the waterline for reflection mapping and use `θr = (180°-θ) mod 360°`, then distort only the reflected edge and intensity with the photographed waves.

When wheel, skyline, and water are all prominent, the wheel is the sole phase coordinate system. Assign each building cluster to the existing universe sector containing its hub-relative centroid. Its source-vertical water reflection inherits that universe and merges into the matching sea wedge. Building assignments do not create new incoming fields or independent vertical phase grids.

Use `palette_guardrails` to cap any overrepresented family. “Less pink-blue,” for example, means shifting PRIMARY or one incoming field toward warm, neutral, or graphite construction while keeping U03 pink/violet as a small edge accent rather than deleting medium contrast.

Prefer radial cell fractures and angular space shards where sectors meet. A single shallow volumetric ellipse may echo the hub, but concentric full-frame rings turn the scene into a logo.

### 4. Vehicle as a time axis

The vehicle's travel direction defines the longitudinal time axis. Wormhole apertures cross that axis approximately perpendicularly, like uneven temporal cross-sections.

**Frame allocation**

- Give **50% PRIMARY** to the continuous vehicle/world timeline between and around the apertures. Carry its palette, material, weather, and dimensional 3D/cel logic through both vehicle and setting while keeping every source edge registered.
- Keep **20% PHOTO** as a continuous recognition anchor containing the vehicle front or rear, an uninterrupted body run, at least one grounded wheel/hull contact, route information, and its source shadow or wake.
- Add **30% INCOMING** through four uneven apertures of **5%**, **7%**, **8%**, and **10%**. Make at least one state decisively flat 2D and at least one dimensional 3D/cel; the other two require different time, material, weather, or motion logic.
- Treat PRIMARY as the baseline vehicle state visible between apertures. The four incoming apertures then show four additional states of that same source-locked vehicle.

**Aperture rules**

- Use **four** apertures, never identical rings. Their visible areas are **5%, 7%, 8%, and 10%** of the frame, totaling the allowed 30% INCOMING coverage.
- Orient each aperture plane roughly **70–110 degrees** to the vehicle's travel axis, adjusting for perspective. Foreshorten the ellipse according to camera angle.
- Vary center height and spacing using real body features such as doors, windows, axles, deck levels, or hull ribs. Avoid a regular bead-string sequence.
- Make every aperture spatially convincing: near lip, thickness, internal parallax, occlusion, and a far exit or receding environment. The flat 2D world stays flat inside that volumetric opening. Retain local reflections and body curvature where each rim meets the source.
- Keep the PHOTO anchor uninterrupted across at least **25% of the vehicle's visible length**. Across all other roles, preserve 100% of the original silhouette, axle/window/door positions, perspective, and contact geometry.

**Vehicle states must visibly differ**

Show distinct, immediately readable states of the same vehicle or route, not minor color grades. Useful contrasts include:

- pristine/new versus aged/worn/weathered surface treatment;
- daytime/dry versus neon-night/rain;
- dark cabin versus fully illuminated cabin while preserving every photographed occupant and window position;
- historical livery versus present livery versus speculative material treatment on the same locked body design;
- intact road or harbor versus source-registered flooded surfaces, overgrowth, low-gravity particles, or a partially dematerialized material pass;
- stationary boarding state versus motion-streaked transit state.

Use at least **three different state dimensions** across the four apertures—for example era, weather, and material finish. Vary rendering and condition without changing the vehicle's outline or source position. Repeating the same vehicle with four hue shifts is a timid filter, not time travel.

### 5. Tram interchange through wires and rails

Hong Kong tram images already contain a topological diagram: catenary wires above, rails below, and the tram between them. Use that real network as the interchange.

**Frame allocation**

- Give **50% PRIMARY** to the governing transit world across connected street, façade, upper-wire, and background regions. Its material and atmospheric logic must continue behind the tram while every rail, wire, building, and tram contour remains registered.
- Keep **25% PHOTO**: approximately **17%** on an unmistakable natural tram/body region and **8%** on route number, people, wheel/rail contact, wire attachment, and nearby street evidence.
- Add **25% INCOMING** as three routed universes: **10% flat 2D** carried by the wire network, **8% dimensional 3D/cel** at a rail junction, and **7%** in a third cel/material branch.

**Interchange rules**

1. Trace actual visible rails, switch points, catenary wires, support arms, trolley pole/pantograph contact, and utility poles.
2. Select one authentic upper junction and one lower junction. Pair them by a path that crosses or touches the tram at a real functional point.
3. Let overhead wires carry 2D schematic energy or route diagrams; let rails carry grounded 3D/cel depth, light, or material change. This maintains a clear 2D/3D contrast without shifting either network.
4. Place a compact volumetric ellipse only where a real switch, crossing, contact point, or occluded continuation supports it. Its orientation must follow the local junction plane.
5. Preserve a clear 25% PHOTO anchor on tram and transit context. Other tram regions may carry PRIMARY or incoming rendering, but must retain the exact body silhouette, windows, doors, wheels, route marks, and source position; never replace it with a generic illustration.
6. Keep rail gauge, wheel placement, perspective, and wire attachment credible. Do not draw a decorative circuit that disconnects from the source infrastructure.

The result should read as one transit system momentarily routing through incompatible worlds, not as glowing lines added over a tram photo.

### 6. Skyline and water mapping

Map architecture upward into sky and downward into water using the source horizon as the hinge.

**Frame allocation**

- Give **52% PRIMARY** to one connected world spanning approximately **30% skyline/sky** and **22% corresponding water**. The universe may change material, atmosphere, and render pass but may not move a tower edge, horizon, seawall, boat, or reflection origin.
- Keep **23% PHOTO** as a paired anchor: approximately **12%** of natural skyline/horizon and **11%** of its directly associated water, boats, or seawall.
- Add **25% INCOMING** as three mapped universes: **10% flat 2D** sky/water cells, **8% dimensional 3D/cel** tower cluster, and **7%** volumetric reflection state.

**Mapping rules**

- Group the skyline into **three to six unequal clusters** using real gaps, roof heights, color blocks, and setback rhythm. Never divide it into equal columns.
- For each transformed cluster, project its horizontal span upward into a sky cell and downward from the true waterline into a water cell.
- Preserve the vertical association between tower and reflection, but allow water cells to shear, widen, break, and fade according to wave direction, boat wake, and exposure time.
- Keep at least one major skyline cluster and its water below entirely photographic. This paired anchor makes the altered mappings legible.
- Use clean graphic sky regions for 2D contrast; reserve volumetric fog, modeled surface depth, space revealed behind existing tower gaps, or luminous water cavities for 3D/cel contrast.
- Hide render-pass seams in tower gaps, roof setbacks, cloud edges, and reflection breaks. If a seam crosses a building, align it with a real structural edge rather than slicing windows arbitrarily.

### 7. Broad city mass-to-field collision

Use this fallback only when no stronger wheel, water, rail, wire, vehicle, or vanishing-point hook controls the frame.

**Frame allocation**

- Give **55% PRIMARY** to the largest connected architectural mass and its adjacent street or sky background. Carry one rendering logic across both depth zones so it reads as the governing city.
- Keep **25% PHOTO** on a recognizable landmark or façade cluster, source-natural people/vehicles/signs, and one continuous foreground-to-background route.
- Add **20% INCOMING** as two unequal universes: **9% flat 2D** fitted to one secondary mass or void and **11% dimensional 3D/cel** fitted to another.

**Construction**

- Derive all three role boundaries from actual rooflines, setbacks, façade returns, street edges, shadows, and foreground occlusion.
- Keep every building footprint, roof contour, window grid, subject, and camera relationship in exact source registration across PRIMARY and both incoming worlds.
- Let PRIMARY occupy both foreground and background. A large isolated foreground patch is not a governing world.
- Make the PHOTO anchor contiguous enough to identify the place at thumbnail size.
- Do not subdivide the two incoming universes further. Equal columns, repeated diagonals, and checkerboards defeat the source-mass logic.

## Seam grammar

Use seams as evidence that incompatible spaces meet. Select **one dominant seam family** and at most **two supporting families** per image.

### Cell fractures

Irregular polygonal boundaries that inherit source geometry. Best for façade planes, water patches, wheel sectors, and sky cells.

- Scale cells with perspective: larger in foreground, tighter at depth.
- Let some cell edges disappear behind real occluders.
- Keep angles and sizes uneven; avoid a honeycomb or stained-glass blanket.
- Use hard edge contrast where realities collide, then allow a few local dissolves or source-aligned space shards without moving the photographed objects.

### Render-pass seams

Boundaries between photographic, cel-rendered, wireframe, material-pass, night, historical, or speculative render states.

- Place them on real depth thresholds, panel joints, building edges, spoke lines, rails, or water reflection breaks.
- Make the state change decisive: lighting model, texture behavior, spatial depth, and material response should change together.
- Preserve continuity of silhouette and perspective across the seam so it reads as the same subject in another pass.

### Angular space shards

Sparse wedges or splinters of displaced space. Best as accents in sky slots, around rapid exits, between wheel sectors, or above skyline gaps.

- Limit them to roughly **3–8% of frame area**.
- Aim shard direction from the source hook: wheel spokes, vehicle velocity, façade perspective, or skyline verticals.
- Give 3D shards thickness, edge light, and occlusion; give 2D shards flat fills and drawn outlines. Do not blur the two languages together.

### Volumetric ellipses

Foreshortened apertures with thickness and internal depth. Best for vehicle cross-sections, rail/wire junctions, wheel-hub echoes, and water cavities.

- Ellipse orientation must be justified by a real axis or plane.
- Vary size, tilt, rim thickness, depth, and placement.
- Show an entry lip, interior atmosphere or environment, and an occluded far side.
- Never use a row of identical circles or a centered generic “portal” unrelated to the photograph.

## Contrast calibration

A convincing collision contains three legible image languages:

| Language | What must remain visible | What makes it distinct |
| --- | --- | --- |
| PHOTO anchor | Sensor texture, source lighting, natural reflections, optical depth of field, weather, real signage and wear | Natural rendering across 20–35%; no global recolor or synthetic repaint |
| 2D world | Flat value grouping, assertive contour, simplified texture, graphic color boundaries, intentionally shallow space | Hard graphic decisions; not merely lowered detail or boosted saturation |
| 3D/cel world | Added volume, modeled thickness, internal occlusion, spatial particles/fog, physically coherent local light | Occupies measurable depth while the source shell stays registered; not a glow halo or embossed filter |

At least one direct boundary should place flat 2D against PHOTO, and another should place dimensional 3D/cel against either PHOTO or the 2D field. If all transitions are softened with haze, the three languages collapse into one cautious composite.

## Anti-patterns and corrections

### Paper tears, peeled edges, and scrapbook openings

**Reject:** torn paper fibers, curled corners, ripped posters, tape, collage scraps, or a paper hole revealing another scene. These describe a craft material, not a spatial collision.

**Correct with:** source-aligned cell fractures, render-pass discontinuities, angular space shards, or volumetric apertures that inherit the photographed geometry.

### Timid full-frame filter

**Reject:** one color grade, soft bloom, chromatic aberration, generic cyberpunk neon, or a translucent texture over the whole photo while all geometry remains unchanged.

**Correct with:** let one PRIMARY universe govern 45–60%, retain a 20–35% PHOTO anchor, add two to four incoming universes totaling 15–30%, and make flat 2D and dimensional 3D/cel fields structurally explicit.

### Repeated rigid splits

**Reject:** equal vertical thirds, checkerboards, repeated diagonal slashes, four identical quadrants, evenly spaced depth bands, or a row of cloned portal rings.

**Correct with:** unequal regions derived from façade masses, perspective depth, wheel sectors, axle/door spacing, rail/wire junctions, skyline gaps, or wave-distorted reflections.

### Unmotivated portal placement

**Reject:** a centered circle placed wherever there is empty space, an aperture floating without a source axis, or a wormhole that clips the subject only for spectacle.

**Correct with:** identify the real axis, hinge, hub, junction, or reflection line first; orient and foreshorten the aperture from that structure.

### Broken source identity

**Reject:** replacing the whole tram or vehicle, changing its silhouette between seams, losing all wheels or contact shadow, scrambling signage, shifting the horizon, or moving reflections away from their sources.

**Correct with:** restore a clear PHOTO recognition anchor and return every subject, landmark, vehicle component, horizon, rail, wire, and reflection origin to exact source registration across all roles.

### Equal-strength chaos

**Reject:** every region using a different world, every seam glowing, and all effects having equal size and contrast.

**Correct with:** one PRIMARY universe at **45–60%**, one PHOTO anchor at **20–35%**, and **two to four** unequal INCOMING universes totaling **15–30%**. Make each incoming field identifiable but visibly subordinate to PRIMARY.

## Pre-generation checklist

Before writing or executing the image-generation prompt, verify:

- [ ] The chosen layout is triggered by a visible source feature.
- [ ] PRIMARY, PHOTO, and INCOMING budgets add to 100%; PRIMARY is 45–60%, PHOTO is 20–35%, and two to four incoming fields total 15–30%.
- [ ] Source geometry is 100% locked across every role: camera, crop, subject silhouette and position, landmarks, horizon/waterline, perspective, vehicle structure/direction, rails/wires, signs, and occlusion order.
- [ ] The PHOTO anchor remains naturally photographic and clearly identifies the place or subject at thumbnail size.
- [ ] Flat 2D and dimensional 3D/cel regions are both visible and differ structurally, not just by color.
- [ ] Each seam follows a real plane, axis, depth cue, sector, junction, or reflection.
- [ ] Ferris-wheel sectors, sky rays, and water reflections share one angular phase map when that layout is used.
- [ ] Vehicle apertures are uneven, perpendicular to travel in perspective, volumetric, and reveal visibly different states.
- [ ] Tram wire routes and rail routes connect through real junctions and attachment points.
- [ ] Skyline-to-water regions maintain tower/reflection correspondence while respecting waves.
- [ ] There are no paper tears, timid global filters, repeated rigid splits, cloned apertures, or decorative geometry without a source hook.
slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."Stars: 198kUpdated Claude CodeCodexVerifiedInstall
ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report.Stars: 196kUpdated Claude CodeCodexVerifiedInstall
ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt.Stars: 196kUpdated Claude CodeCodexVerifiedInstall