CommunityImagegithub.com

vinyl-scene-poster

Transform a user photo into a 3:4 looping animated poster combining a vinyl record with a miniature scene: extract subjects, scenes, and colors from the photo, then build a design-driven composition using a circular scene record, center label, fixed tonearm, miniature characters, and localized sparkle effects. Ideal for record-style scene simulations, single-image animated posters, and AI + p5.js creative motion graphics—not for generic film filters or static record sleeve displays.

What is vinyl-scene-poster?

This skill converts a user-submitted photo into a visually striking, looping animated poster that merges the aesthetics of a vinyl record with a miniature diorama, outputting in a 3:4 aspect ratio. It begins by analyzing the photo to identify a central subject, environmental cues, up to three recognizable objects or shapes, and three to five usable colors, then formulates a creative concept like 'turn the [environment] in the photo into a [miniature world] on the record, with the [subject] moving along a [meaningful circular path].' The skill supports platforms like Claude Code, Cursor, or Codex and uses image generation models (e.g., DALL-E, Midjourney) to re-imagine a top-down view of the scene, followed by code (p5.js, Canvas 2D) to animate the record rotating while keeping the tonearm and background fixed. The final deliverable includes an MP4 video, a PNG cover image, and an executable source file; it intelligently handles edge cases like non-top-down photos by reconstructing the scene rather than simply clipping and rotating original elements, ensuring a cohesive and professional result.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/1NWZ/vinyl-scene-poster/tree/main

Installed? Explore more Image skills: steipete/songsee, affaan-m/frontend-design, mattpocock/improve-codebase-architecture · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Miniature World Inside the Record

Turn a user's photo into a dynamic poster that looks like "this moment is being played." The visual anchor is a giant overhead disc: the surface is both a record and a miniature scene related to the input photo; grooves simultaneously become ripples, paths, contour lines, or other semantically meaningful circular textures.

The default output is an animated finished piece. First, have the image model reconstruct a unified overhead scene based on the input, then have that scene rotate within the turntable; text, the tonearm, and the fixed background are drawn separately. It is not acceptable to simply clip slanted elements from the original photo and rotate them in a circle, nor to rotate the entire poster including the fixed tonearm. When the user only requests analysis, planning, or prompts, deliver within the scope they specify.

Input and Defaults

  • Required: A photo. First, actually look at the photo, then choose the scene. If no photo is provided, ask the user for one.
  • Optional: Theme, title, color palette, references, dimensions, duration, motion intensity, output format.
  • When not specified: 1080 × 1440, 3:4, 30 fps, 6-second loop; one concept; no audio track; deliver MP4, PNG cover, and a runnable animation source file. Append a GIF or interactive webpage only if the user requests it.
  • Actively determine the theme and layout from the image, without requiring the user to fill in a long parameter list. Only ask questions if the material is unclear, the main subject is unrecognizable, or key requirements conflict.
  • A "film look" can be a lightweight texture preference, but the structure of this technique is vinyl records. Add photographic film elements like light leaks, date stamps, or sprocket holes only when explicitly requested.

Start by Finding the Scene Reason from the Photo

Identify one main subject, one environmental cue, up to three recognizable objects or shapes, and three to five usable colors. Distinguish between what is actually visible in the photo and creative translations; do not fabricate shooting locations, dates, character relationships, or song information.

Form an internal concept: Turn the [environment] in the photo into a [miniature world] on the record, and have the [main subject] move along a [semantically meaningful circular path].

Image ClueRecord Surface WorldMeaning of Circular TextureMiniature Elements
Seaside, pool, summer figureWater surface recordRecord grooves doubling as pool ripplesFigure with clothing cues, swim ring, water splashes
Coffee, tabletop, breakfastCoffee or cream-toned tabletop worldCoffee ripples, stirring trailsCup, beans, small objects from the photo
Garden, countryside, travelCircular garden or terrain worldPaths, contour linesFigure, tree shadows, flowers, signpost
City, architecture, night sceneCircular neighborhood or neon water surfaceRing roads, light trailsBuilding silhouettes from the photo, vehicles, pedestrians
Pet or simple background portraitMiniature world designed around the subject's featuresA path formed by existing objects or shapesOne recognizable main subject and a few related props

This table is a starting point for association; do not force every photo into a pool scene. Prioritize preserving silhouettes, clothing color blocks, pet patterns, and iconic objects from the source image to ensure the final piece answers "why this belongs to that photo."

Use the photo as a reference for the subject, colors, and materials, then regenerate the geometric and occlusion relationships as seen from a direct overhead view. The person can retain their identity through hairstyle, clothing, accessories, and silhouette; food, objects, and pets maintain association through shape, color, pattern, and distinctive details. By default, use only one user protagonist; do not unreasonably duplicate the same person or pet into a circle. If the user explicitly requires the original image pixels to remain completely unchanged while the original photo is not a top-down view, explain the conflict between these two constraints, then proceed according to their clear choice; do not default to clipping and rotating from the original perspective.

Visual Skeleton

Default to a strictly top-down (90° nadir / orthographic top-down) unified scene language, rendered as a delicate illustration or miniature photorealistic scene. All objects and the ground share the same camera direction, scale, and lighting, integrated into the disc via contact shadows, waterlines, or environmental reflections. The recognizability of the reference layout comes from the scale contrast of "a large disc cropped in the frame + a small center label + a tilted fixed tonearm + miniature characters."

Using the canvas width W and height H as units, the following values are adjustable starting points:

  • Background uses warm white or a light tone derived from the original image; leave breathing room in the lower right for the tonearm base.
  • Disc center approximately (0.33W, 0.42H), outer radius about 0.67W. The disc breaks out at the top and left edges, and approaches the right edge of the canvas; do not shrink it to a small centered disc to show a full circle.
  • Center label radius approximately 0.23W, using a color clearly distinct from the disc surface. Add a small spindle hole and one or two thin outline strokes.
  • Elements on the disc are constrained between "label outer edge + safety margin" and "disc outer edge − safety margin." Check by the element's bounding radius, not just its center point.
  • By default, include four to seven miniature visual units, with the protagonist slightly larger or more prominent in color; balance density without equally spacing them all around the circle.
  • Tonearm base located approximately at (0.90W, 0.84H), with the stylus near the right edge of the label; forming a stable diagonal line from the lower right to the upper left.
  • Tonearm uses warm white, light gray, and a few dark construction lines, rendered above the disc surface, with consistent dimensions, orientation, pivot point, and shadow.

Positions can be adjusted for different inputs, but maintain a clear spatial relationship between the disc, label, miniature scene, and fixed tonearm. Follow the user if they explicitly request a different composition.

Color Palette and Text

Default to "large area scene key color + highly recognizable label color + neutral background color + a few accent colors." An approximate sample palette is water blue #008DE9, lemon yellow #FBED07, warm white #FAF7F2, orange-red #D95A13, dark #102A3A; this is an example, not a fixed filter for all photos.

Select the key color from the original image, adjusting brightness and saturation to suit the scene. Concentrate high-saturation key colors on the disc surface or label; text and characters must remain legible. Only use the blue-yellow combination directly when semantically matching summer, water, etc. Keep the image clean; do not default to adding heavy grain, vignettes, or dirt for a "vintage" effect.

Text should belong to record design, using real typography, not treating AI-generated pseudo-text as a finished product:

  • The main title captures the mood or scene of the photo, using two to four English words or four to eight Chinese characters; default to not copying SummerSwim from the references.
  • Arrange the title along the arc of the center label, accompany it with a small SIDE A and one or two lines of original phrase. English is not mandatory; follow the user's language preference.
  • Make the main title prominent and the small text restrained; use normal Unicode characters and an appropriate font, not fancy Unicode letters to simulate a typeface.
  • Label text can rotate with the disc. For the cover, choose a phase where the title is easy to read; when the user needs long text for constant reading, move it to a fixed white space.
  • Original phrases can mimic the format of track numbers, but must not fabricate real releases, real songs, or actual shooting information.

Motion Design: One Primary Motion, Limited Environmental Motion

  1. Primary motion: disc rotates continuously. Default to counterclockwise, one rotation every six seconds; the label, curved text, grooves, and miniature characters on the disc share the main rotation angle. Characters revolve with the disc and change orientation, creating the feel of a record playing. Do not rotate the entire canvas together.
  2. Stable reference: base and tonearm are fixed. Complete the disc coordinate transformation first, then draw the tonearm. Keep the pivot fixed; do not default to adding unwarranted mechanical wobble or camera push/pull.
  3. Environmental motion: choose one or two based on the scene. For water, use short waterlines, trails, and sparse four-corner highlights; for coffee, gentle ripples or localized steam; for gardens, slight leaf swings; for night scenes, localized light breathing. The motion must have a scene-based reason.
  4. Motion hierarchy: clear revolution, restrained local motion. Local displacement starts at 2%–5% of the character's size; do not alter facial structure, main silhouette, or cause objects to continuously morph. Water highlights fluctuate without flashing the entire screen white.
  5. Clear boundaries between moving and static elements. Assign scenes, characters, and text to distinct layers. Fixed text should not enter the disc rotation group; oscillation, orbital displacement, and orientation changes must not redundantly apply the same main rotation angle.

Not all circular lines need to move; concentric circles are difficult to see even when rotating, so rely on asymmetric characters, labels, short arcs, and highlights to convey rotation. All flashes, waterlines, and shadows are clipped within their own scene area and should not spill onto the warm white background.

Production Workflow

1. First Define the Static Poster and Material Layers

Begin with a single scene concept sentence, determine the retained features from the source image, theme, color palette, title, and layer relationships, then have the model reconstruct the static overhead scene. If the original image is already top-down, still check if it is suitable for a unified miniature scene; do not default to directly clipping out objects.

Use the image generation or editing tool available in the current environment, using the user image as an explicit subject reference. By default, generate a complete square overhead scene texture: the scene center coincides with the image center, reserve a center area for label coverage, distribute the subject and objects in the outer ring, and extend the same environment at the edges. The model generates the subject, environment, and contact relationships simultaneously; the image does not contain label text, spindle hole, tonearm, or external canvas background. Code will clip it into the disc and composite it with these independent layers.

Only generate separate overhead materials when the character truly needs independent motion; each material must still be reconstructed by the model and maintain the same top-down direction, scale, and lighting as the scene. A complete scene inside the disc rotating with the disc is a valid approach; do not break it into unrelated stickers for the sake of layering.

Write editing prompts based on the following structure:

Use the input photo only as a reference for the subject, colors, and materials; do not adopt its photographic angle. Rebuild a complete miniature scene for the [theme], using a strictly 90° vertical top-down view, orthographic or near-orthographic projection, with all objects viewed from directly above. Retain the [subject's features, colors, objects], reconstruct their top shapes and correct occlusion: the mouth of a round container should be nearly circular, not showing tall sides from the original image; flat objects should lie flat in the scene. The subject connects naturally with the [scene ground] through contact shadows, environmental reflections, or waterlines. Center the square image with a radius approximately 17%–19% of the frame width as an empty area for objects, place objects in the outer ring with sufficient edge space. The color palette is [source colors]. Do not generate text, labels, spindle holes, tonearms, watermarks, or sticker borders; do not use oblique top-down, isometric, or original image clipping effects.

Before animating, check the generated image's perspective: are the cup rims circular? Are tall cup fronts still visible? Do all objects have a consistent top orientation? Are contact shadows from the same environment? If a 3/4 perspective from the original photo appears, adjust the model prompt and regenerate; do not proceed directly to animation. Only change this criterion when the user explicitly chooses a different camera angle.

When you only have a flat composite image, do not pretend you have movable layers: remake the required layers, or switch to truly separable cutouts and procedural graphics.

2. Organize the Animation with Code

Use p5.js, Canvas 2D, or equivalent tools. First, check the runtime and export capabilities, then choose the implementation; do not rely on a video generation interface that may not exist. The repository includes a reusable Canvas example: scripts/scene.js handles layering and renderFrame(t), scripts/render.cjs handles export. When using it, pass your newly generated overhead scene to --scene, and modify the color palette, title, and composition to match the current image; do not keep the example's theme.

Default layer order: fixed background → complete top-down scene within the circular mask (including environment and subjects) → optional fine grooves and local light effects → center label and text → fixed tonearm and its shadow → fixed additional text. When using separate materials, break the scene into environment, subjects, and local effects; do not redraw subjects already contained in the model-generated scene.

Use a time-locatable renderFrame(t) consistently; preview and export share the same function:

T = 6
fps = 30
N = T * fps
u = (t mod T) / T
mainAngle = -2π * u

local oscillation = A * sin(2π * k * u + phase)
flash brightness = low + range * ((1 + sin(2π * k * u + phase)) / 2)^4

Use integer k; randomize positions, sizes, and initial phases only once with a fixed seed. Different objects can have different phases and integer frequencies. For a more natural main rotation speed, use periodic speed modulation, e.g., mainAngle = -2πu - a*sin(2πu) with 0 ≤ a < 1, maintaining same direction, position, and speed continuous at loop boundaries.

By default, place character positions in the disc's local coordinates, producing revolution via parent layer rotation. When explicitly requiring independent orbits, ensure they close within the same cycle using integer revolutions, and do not apply the parent's main rotation angle again. Reconstruct trails from the same periodic time function, not from infinitely accumulated previous frame afterimages.

In p5.js, rotate() rotates the coordinate system; use push()/pop() to isolate the disc from the fixed tonearm; by default, positive angles are clockwise, counterclockwise uses negative angles. When implementing, consult the official rotate documentation.

3. Export a Loopable Final Piece

Export frame numbers 0…N−1, corresponding to time n/fps, without duplicating the frame at t=T. When rendering frame by frame, first wait for images and fonts to load, then save the frame sequence and encode it as MP4; do not record browser dropped frames causing stutters into the final piece.

MP4 should prioritize H.264, yuv420p; the player loops the playback. Create a low-size preview separately for GIF if needed; do not use a blurry GIF as a substitute for the high-definition main file. Use relative paths for source files and package actual dependencies; if using p5.js, record the verified version and how to run it. Do not claim that pages requiring online scripts for fetching can run offline.

Do not directly use saveFrames() as a 30 fps high-definition exporter; its official documentation lists frame rate and duration limits. Use deterministic frame-by-frame capture or a reliable encoding method in the environment. See the official saveFrames documentation.

Acceptance and Delivery

  • Input association and perspective: The main subject or key objects remain recognizable; the theme and color palette are traceable back to the input photo. The scene has indeed been reconstructed by the model, with all objects in a unified vertical top-down perspective. Clipping and rotating from the original perspective fails this acceptance.
  • Static composition: The disc, label, and tonearm have clear layers; titles have no typos; the main subject is not persistently obscured by the label or tonearm.
  • Dynamic structure: The disc revolution is visible, the tonearm is stable; no drifting subjects, material flashing, whole flat image rotation, or unexpected text stretching.
  • Loop: Compare renderFrame(0) and renderFrame(T); states should match. Also play the last few frames followed by the first few frames, checking speed, particles, trails, and exposure continuity. The last frame and the first frame should be adjacent motion states; pixel-perfect equality is not required.
  • Files: Actually open the exported MP4, check dimensions, duration, frame rate, and playback; inspect at least the start, quarter, half, and three-quarter points, and verify the seam between end and beginning.
  • Normal delivery: poster.mp4, cover.png, animation source code and dependencies; a brief description of the theme extracted from the photo, retained features, and motion effects.
  • When the environment lacks a certain capability, deliver the completed and verifiable materials, source files, and accurate running steps, stating which step is missing; when no image model is available, clearly state that the overhead reconstruction is not yet complete, and do not substitute with clipping from the original perspective; do not present a static preview as a finished video.

For a closer look at the reference proportions and composition of this technique, read references/reference-analysis.md. This file is an observation record; when producing, still rely on the current user's image and requirements.

Related Skills