# Vivid Science · 鲜彩科学动画 Original science explanations told through saturated vector worlds, expressive characters, scale changes and diagrams that grow out of the action. ## Visual grammar Use a small family of saturated hues per shot: deep indigo or plum, a cool cyan/teal field, one warm coral/orange focal subject, and pale cream highlights. Alternate dark cinematic fields with bright, open explanatory fields. Match value separation and colour roles before tuning exact hues. Avoid applying a global saturation filter to compensate for weak drawing. Build silhouettes from smooth, purposeful Bézier contours. Most forms have a base fill, a darker coloured rim or shadow plane, a lighter facet and a few large material marks. Dark lines belong mainly to facial details and internal structure; an all-purpose black outline changes the look. Gradients may shape large fields or a local glow, but must preserve crisp flat edges. Avoid grain, photographic textures and indiscriminate blur. Shape broad reflections to follow the body's contour instead of placing a generic polygon on top. Keep them inside the same local transform as the body. When revising a shared character, inspect every shot that uses it, including small, enlarged and mirrored poses; a softer contour can change apparent contact or clearance even when its animation is unchanged. Preserve purposeful points and anatomical distinctions rather than rounding every edge uniformly. For flexible appendages, vary width along a curved centreline and carry the highlight through the same deformation. Extend the root beneath the body fill; replacing a round-capped stroke with a filled contour can otherwise reveal an attachment gap. Inspect the compressed, extended and turning poses. Keep jointed limbs visibly articulated rather than making every appendage a flexible ribbon. Design supporting forms at their actual delivered size. Preserve a distinctive outer contour, one broad shadow and one useful highlight before adding small internal marks. Derive shade and highlight colours from each local palette so a recoloured subject retains the same structure. Inspect the lightest and darkest variants on their real backgrounds. If detail is reduced with scale, fade it over a short range rather than switching it at a hard threshold; check shrinking and growing objects for visible popping. Compose in foreground, action and background layers. A close-up may crop a subject intentionally; labels, important gestures and causal relationships must remain visible. Background density varies with the narrative: populated worlds can be dense, explanatory diagrams need empty space. Repeated decorative symbols must not replace actual subject drawings. For an environment, reduce contrast and contour detail with distance. Use a few cropped near shapes, a readable action plane and a quiet far silhouette instead of stacking equally detailed horizontal bands. Give reflections and clouds unequal widths and gaps; keep their direction consistent with the light source. Place broad illumination behind the acting subject and reserve a quiet title area throughout the camera move. Vary a crowd's spacing, size and depth without allowing random placement to merge adjacent silhouettes. Build a cutout as real geometry or a clipped path when its background contains a gradient or moving layers. Painting a hole with one sampled background colour can expose a disc or seam during a camera move. Derive a boundary highlight from the same path as the filled shape so the two cannot drift apart. For coordinated group movement, share the main travel phase but assign unequal, deliberately spaced positions within the group. Check the full appendage bounds, not just body centres; a regular grid can look artificial even without overlaps. Use continuous value changes for gradual depth and brackets or leaders for measurement ranges. Hard colour boundaries should imply real divisions only. Keep a foreground object's path continuous for the whole shot. A modulo-based loop is safe only when the entire silhouette has left the camera's visible area before it wraps. Check the frames immediately before and after every reset, including tentacles and other long appendages. Evaluate foreground placement after the camera transform: an object that frames the source canvas may be completely cropped away in the delivered shot. Use different travel distances for near and far layers while keeping the acting subjects easy to follow. Check each moving body's value against the entire background path, including light beams and gradients. A colour that separates at one depth can disappear at another while its bright eyes remain visible. Adjust the body palette or shadow plane and reinspect the worst crossing at delivery resolution. Use quiet coloured shapes to group a diagram when they clarify its relationships. Reveal the groups in causal order and reserve label space beyond the full moving silhouettes. In a dense microscopic world, vary contours, scale and overlap deliberately: a few large cropped foreground forms, readable middle-distance subjects and lower-contrast distant forms establish depth more clearly than an even ring of identical bodies. Keep important internal marks legible at the smallest delivered size. Judge density from the gaps and occlusion between forms, not only their count. When the reference presents a packed world, let background masses interlock and overlap behind the focal silhouette instead of surrounding it with isolated icons. Use broad highlights, varied internal contours and a distinct focal colour to preserve hierarchy. Limited softness can separate a nearby cropped layer; retain crisp edges on the explanatory subject. Borrow the spatial principle without importing anatomy or mechanisms that the new subject lacks. Compare typography by letter proportions, terminal shape, stroke weight and spacing at the final display size. A heavy rounded face can shift a scientific diagram toward a different visual tone even when its palette matches. Retain a licensed font file with the production, verify that the renderer actually loads it, and recheck every visible label after changing the font or its weight. Protect labels from moving particles, markers and bright background shapes. Give important words a quiet field or a coloured capsule, and inspect the whole interval in which a label is visible, including its entrance and exit. Size capsules from the rendered word width with generous side padding, rather than shrinking the lettering to fit a fixed box. Place explanatory symbols beside the objects they describe so a nearby light source, arrow or decoration does not accidentally imply the wrong causal relationship. If a capsule has a small connector, orient it toward its subject, including labels placed below a group. Reserve enough margin for both the capsule and its connector. When lighting changes over a shot, check the intermediate palette as well as its endpoints. Interpolating both lettering and background between light and dark can make them meet at the same value halfway through. Use a stable contrasting field for essential labels, and keep new foreground layers clear of their full animated bounds. Give populations distinct near, middle and distant scales, with quieter colour and detail in the distance; vary spacing without losing the main subject's reading space. Faces use readable eye whites, dark pupils, sparse highlights, coordinated brows and a mouth shape that communicates the current intention. Author a model sheet and expressions for each new production. Keep anatomy and explanatory metaphors consistent with its subject. Do not carry a previous production's cast into the next film. Judge facial marks by their role and scale rather than adding decorative rings to every eye. A gaze change and a pupil-size response can communicate attention with few shapes. Clip the pupil to a changing eyelid aperture during a blink; flattening every eye layer together makes the pupil appear to deform. Inspect consecutive frames at native pixel scale, including the nearly closed interval. Keep incidental particles behind the acting face so they do not resemble an extra pupil highlight or an unintended mark. Use continuous expression controls for a sustained eye closure or a change from smiling to alarm; replacing the complete face at a mood threshold can create a single-frame jump. Blend brow height, aperture, pupil size and mouth contour at different rates around the same intention. Check raised brows against the deformed head boundary. When a subject receives another object, derive the contact point from its actual rotation, breathing and pose, and use intentional occlusion to complete the transfer instead of shrinking the object in open space. ## Motion and camera Give each character an intention, anticipation, action and response. Separate primary movement from delayed secondary motion. Use short eased accelerations, controlled squash/stretch and a settling hold when they help the action read. A continuous sine drift is secondary animation, not the action itself. Establish the trigger before the response, through visible action or a clearly timed sound cue. For an offscreen sound, let the gaze lead its visual reveal. A useful sequence is stimulus, gaze change, preparation, release and recovery. Place facial and limb changes around those beats instead of running every movement at the same frequency. Inspect the full deformed silhouettes at closest approach: separated character centres can still leave a nose, hand, foot or trailing appendage intersecting. Give interacting characters separate attention and blink events. A reaction may briefly close one character’s eyes while the other keeps tracking its target. Ease gaze changes across the preparation and release, and vary secondary paths with restrained delay and amplitude; copying one vertical path onto every body can make an interaction look mechanically linked. Inspect the closest approach again after changing those paths. Carry a response through the whole pose: a brief recoil, a change in stroke strength, a revised facial intention and a settling hold should read together. Keep interpolated mouth shapes inside the face throughout the turn, including the intermediate shapes between a smile, an open mouth and a frown. Inspect those transitions in consecutive native-pixel crops as well as full frames. Clip interior details to the same animated aperture used for the mouth fill. Protruding teeth must follow the changing lip and retract behind it; fading stationary teeth can leave translucent fragments outside an open mouth. When refining an antagonist, retain its distinct proportions and purposeful points while smoothing the surrounding curves and broad material reflections. Place a small reaction sound in an available speech gap and bind it to the same timeline event; changing the sound mix requires new encoded-audio checks. For a shared departure, let individuals notice and prepare at slightly different times, then recover into a common travel direction. Keep the offsets bounded by the spacing available throughout the path. Drive cyclic strokes with a continuous phase; multiplying absolute time by a changing frequency can create a visible jump. Tie the main release to a named timeline event, with the gaze and preparation leading it, so a picture revision preserves the sound and speech relationship. Alternate landscape, medium, close and explanatory views. Push in to reveal a mechanism, track to follow a transfer, and pull back to show the larger relation. Maintain one world-to-screen transform for characters, leaders and masks. Recompose a close shot when a simple camera crop makes an unrelated foreground body dominate the frame. Reserve space around the acting face and appendages, reduce background contrast and keep large warm shapes from competing with the focal subject. Give a crowd enough spacing to read as separate silhouettes and keep its complete motion paths outside the title's reading area. Place ambient particles behind faces unless their foreground crossing serves a visible action. Build a multi-part explanation from illustrated groups with a clear actor, mechanism and label. Introduce one group first, then move and rescale the existing groups as the next one arrives; keep each label attached to its own group. Reserve room for the complete animated silhouette during that reflow. Quiet organic fields can separate ideas without implying that each field is a literal physical compartment. Draw particles from the subject’s material properties, with varied fragments and uneven outlines rather than generic decorative icons. Check opacity through nested drawing helpers and inspect an approaching object’s full path around the face. Retiming a reveal also requires moving its sound cue. Anchor a shape or colour across transitions. Match cuts, expanding shapes and organic wipes can carry a change of scale; reserve full-frame covers for brief transitions and never use them to conceal an unfinished scene. For an inset that expands into a new world, retain the same focal object and screen position through the expansion. Let the incoming background extend beyond every transformed viewport edge. Fade the outgoing label before the incoming label becomes readable, then hold the new composition long enough to understand it. Check the beginning, middle and end of the move for exposed background wedges, doubled labels and an apparent change of object identity. Derive a magnification leader's attachment from the host's actual deformation, including breathing, rotation and scale. Choose a visible edge of the marked region and route the complete leader outside the face and other reading areas. Keep one focal-coordinate function for the inset and the expanded view. In a dense microscopic composition, distinguish a broad membrane from its quieter interior, vary the scale and contrast of surrounding layers, and reserve strong local colour for the subject being explained. Reveal symbolic annotations with the explanation without implying that their appearance creates the material. Separate preparation, movement and settling when a scale transition calls for anticipation. Apply a brief retraction to both the boundary and its contents, then expand them with the same focal anchor. Keep position and scale continuous at each phase boundary. Compare the full motion sequence, including the smallest inset and the first frame of the surrounding world; endpoint agreement alone does not establish the same transition. For a schematic cross-section, separate the overview from the enlarged local mechanism so the drawing does not imply a literal physical scale. Keep the locator on the intended material or region. Build depth from a common curved contour, a darker side face and a lit rim; check complete animated appendages against every foreground mask. Stage energy input, ingestion and transport as distinct actions, and make matter markers originate in matter rather than in light rays. Align the resulting action cues with the spoken keywords. When the explanation moves from overview to local detail, stage the handover: move the overview toward an edge, reveal the connected detail, allow a reading hold, then let the detail occupy the frame. Remove the leader as its context leaves. Decide whether a label belongs to the moving object or to the whole explanation and apply that choice consistently. Check the transformed contour at every viewport corner, and test the new palette at intermediate scale; enlargement can expose background gaps or weaken a small light source's contrast. For layered environments, assign distinct hue and value roles to distant terrain, open space and cropped foreground masses. Build organic blades from one changing centreline and width profile so highlights follow the same deformation as the outer shape. Vary bend, length and response delay while keeping their bases attached. Give a returning character a visible reason to glance or gesture, then let a companion respond after that gesture has begun. Check the whole interaction at small viewing size as well as the eye and appendage silhouettes. For translucent organic forms, separate a broad membrane, a few internal shapes and thin trailing details. Use local colour variation and a restrained rim to describe volume. Let attached parts respond with different delays, preserving their connection throughout deformation. Stage repeated forms with irregular spacing, scale and aspect; avoid solving an empty background by filling a grid. Keep the main action corridor clear and verify any foreground occlusion across the complete passage of the subject. For a causal diagram, reveal groups in the order of the explanation and vary their height and internal arrangement while keeping the reading direction clear. Draw group backplates before any object that crosses between groups. Keep a transferred marker attached to its carrier through translation, rotation and scale, and change ownership at a visible contact or ingestion event. Give remaining members room to settle after a focal member leaves. Inspect nested opacity helpers and the complete transfer corridor; a good endpoint does not prove that the marker remained visible along its path. Set the frame rate from the reference and intended delivery, then render every frame deterministically. A high container frame rate does not prove that a source animation changes on every frame. Measure repeat cadence separately. ## Narration, score and effects Narration should leave enough time to understand the visible action. Generate speech before locking timings, keep the actual waveform and natural speed, and use a shared integer-frame timeline for pictures, voice, labels and effects. Use English or the language requested for the new production. Create a coherent score with a motif, harmony, bass, rhythm and changes of instrumentation. Layer an environment bed and action-specific effects. Connect sounds to named visual events; loudness alone does not make a mix rich. Leave quiet intervals before important reveals, and duck music from the actual voice activity. Preserve separate voice, score, ambience and effects stems. ## Evidence and review Keep a source ledger for factual claims. A friendly diagram is not measurement or evidence; avoid changing scientific details just to preserve a convenient rig. For frame-by-frame comparison, decode both complete files in presentation order. Record frame index and timestamp, every-frame visual features, and consecutive frame atlases. Compare by visual role and shot phase when subjects differ. A nearest feature match is a diagnostic retrieval result, not a percentage of artistic equivalence. Report unmatched regions and inspect them. Record automated coverage, atlas inspection at its actual thumbnail resolution, enlarged still inspection, consecutive motion review, playback and listening separately. Generating an atlas is not reviewing it. No numeric aggregate or successful render establishes “100% identical style”. Repair observed defects, repeat affected checks, and report unresolved differences honestly. Bind review records to the encoded file hash. After a revision, compare decoded frames to locate the actual changed interval; encoder dependencies can extend changes beyond the edited scene. Recheck that interval, its boundaries and the following shot. Keep earlier-version review separate from latest-version review. Audio sample identity can preserve a previous transcription or signal check, but cannot create a listening review that never occurred. Inspect the encoded film, including the first/last frames and every transition. Verify raster encoder quality units, RGB-to-YUV colour handling, dimensions, frame cadence, decode count, audio streams and true peak. Review real playback. If another task is using ffmpeg, wait before starting another ffmpeg command. ## Package scope This is a reusable material and analysis toolkit with empty production inputs. It contains no film, narration, claims, scene composition or review record from a commissioned example. Author each new film in an external working directory. See [README](README.md) and [provenance](PROVENANCE.md).