# Luminous Worlds · 明彩世界

An authoring toolkit for graphic science films with luminous colour fields,
dense illustrated environments, legible mechanisms and independent character
performance. It supports a hybrid of original illustration plates and crisp
procedural geometry. It is not an automatic scene generator or a certified
replica of any reference film.

## Shape, colour and space

Establish a small set of colour roles per sequence: deep indigo shadows, cyan or
teal separation, warm focal shapes, and pale highlights. Alternate bright inhabited
environments and dark explanatory spaces. Match value hierarchy before hue.
Saturation alone does not establish a visual match.

Assign lighting by the role of the shot. A bright microscopic field, a quiet
dark astronomical field and a luminous mechanism need different value structures.
Keep sharp object contours and soften selected peripheral layers to establish
depth. Place emission at an actual rim or active site, with a wider low-opacity
halo and a narrow crisp highlight. Background streaks may support a transition;
they should enter and leave with that event instead of becoming a universal
surface treatment. Recheck title contrast and moving-particle intersections
after changing the background value.

Draw silhouettes with intentional Bézier curves and a few hard shading planes.
Reserve outlines for facial features, internal mechanisms and contact edges.
Use gradients for broad illumination or local emission; retain clean contours.
Avoid random surface noise, photographic textures and universal drop shadows.

Choose material treatment by the visual role and the reference subject. A soft
highlight may describe a translucent or smooth object, while a broad flat shadow
can clarify a solid diagram. Do not remove every gradient or add shine to every
actor under one global rule. Preserve the silhouette, facial anchors and contact
geometry when changing the fill treatment. Compare the result at close-up and
small prop scales, then verify the cut between contexts so the material change
reads as lighting or representation rather than a different object.

Environment plates need a foreground opening, a readable action plane and a
background. Generate or draw them without the protagonist or action props, then
animate those separately. Inspect plates at their actual render scale. A plate
that looks sharp at thumbnail size may have painterly edges, invented details
or inconsistent perspective. Simplify or regenerate those regions before use.
Do not treat a slow pan of a completed illustration as character animation.

When an illustration already has useful staging and depth, simplify its material
treatment without discarding that structure. Compare a replacement against the
whole composition: cleaner edges can still be a regression if silhouettes become
repetitive or the environment loses depth. Keep the action opening, occlusion
order and scale anchors while reducing incidental detail.

Compose dense worlds and quieter explanatory close-ups deliberately. Foreground
objects must have correct ground contact, occlusion, contact shadows and lighting
relative to the illustration. Screen-space labels and world-space leaders need
explicit ownership. Keep labels within safe bounds throughout camera travel.

For a broad landscape in an oblique plan view, establish one continuous ground
plane before placing inhabitants. Derive a river and its banks from the same
curve, and keep a raised terrace's top, side and ground shadow attached to the
same contour. Anchor growth at the roots and sway around those anchors. Group
cultivated and wild areas with different spacing and silhouettes; a repeated
background plate with oversized foreground props does not establish a new
location. Check the entire camera move for accidental crop boundaries and for
small animals disappearing against larger colour fields. Compare the area occupied
by dark vegetation, bright ground and active subjects, not just their individual
colours. Vary silhouettes and cluster sizes before adding incidental detail; a
larger readable animal action can carry more meaning than many tiny oscillators.

When an airborne actor lands, fold its wings, extend its legs and reduce its
shadow displacement continuously into the settled pose. Do not switch between
two incompatible drawings on a threshold frame. For slowing cyclic motion,
integrate speed into phase so deceleration does not reverse or reset the cycle.
Use sound cues for visible action beats without making every actor repeat the
same motion or timing.

For a rotating cutaway, attach each leader to a named local feature and project
that anchor through the specimen transform. Check the settled view and the whole
reveal: a line can remain inside the silhouette while pointing to the wrong
structure. Route leaders so they do not cross one another, and reserve quiet
space for labels before enlarging the specimen. Soft peripheral forms should
establish depth without weakening the focal contour.

Give a long anatomical explanation successive attention windows. Use the spoken
phrase to time a selective highlight, a small reframe or a change of representation,
then reduce the prominence of previous labels. Keep schematic emphasis distinct
from physical activity: an editorial trace does not justify depicting growth,
transport or metabolism. Clip all internal texture and small shapes to the actual
section boundary. Inspect that boundary during the reveal as well as in the held
view; subordinate detail can escape even when the principal silhouette is correct.

For a microscopic close-up, preserve the focal shape and colour identity from
the wider view. Reduce competing labels and peripheral decoration before adding
detail. A single enlarged mechanism can explain more than evenly scattered
symbols. Attach markings to sampled positions on the visible curve: Bézier
control points usually lie away from that curve. Inspect the markings at both
scales and during deformation. Keep illustrative molecular shapes and slowing
gestures distinct from measured structures, trajectories or reaction rates.

For an oblique room, define one ground-plane projection and derive furniture
corners, equipment footprints and contact shadows from it. Let the work surface
cross the frame diagonally, with actors at different depths and a cropped
foreground layer when useful. Keep the active hand and its target in a clear
opening. Decorative perspective lines cannot compensate for a frontal row of
objects. Check both the wide view and its cut to the detail view; preserve the
action's timing and object identity when changing the room around it.

Compare the size of the principal silhouette and its facial features within
equivalent shot types. Matching colour while leaving the character much smaller
changes the visual hierarchy and weakens the performance. When enlarging an
actor, restage companions, shadows and callouts together; check their swept
bounds during the whole action. A gentle push can emphasize recognition, but
its focus and timing should follow the reaction rather than run independently.

When several symbols resolve into one thought or choice, contract the competing
symbols before enlarging the selected field. Give the character room to turn
and react. Move the field, its pointer and its contents through one transform,
and compute clearance from the moving actor rather than a fixed screen point.
Check small animated contents at the largest scale: a droplet or a highlight
that fits the initial symbol may escape the expanded field. A thought image
expresses a possibility or cue; it does not establish that the event occurred.

## Performance and camera

Give the principal action anticipation, transfer, contact and a readable result.
A vehicle wheel rotates from travelled distance and radius. A lid moves from an
open position to the actual vessel rim; remove its static duplicate. A held
object follows the hand rather than a separate screen-space oscillator. When a
travelling prop becomes part of a settled collection, end the travelling copy
on the handoff frame; otherwise duplicate geometry can accumulate unnoticed
in a thumbnail.

A comparison can unfold as one case, several cases and a final focal case.
Stage those changes around the narration rather than holding the whole grid
from the first frame. Give each subject a readable reaction and preserve a
common scale during the comparison. When enlarging the final subject, check
its complete bounds against the associated symbol, labels and base. Distinguish
illustrative rates and character metaphors from measured quantities; animation
speed alone must not imply a quantitative scientific result.

When the reference uses paired fields, match the field division, subject scale,
base geometry and label hierarchy before adding secondary decoration. Keep each
case's label close to its subject; redundant header and footer sentences can
turn a character comparison into a presentation slide. A change of case may be
a direct graphic cut at a stable anchor. Inspect both adjacent frames and avoid
an extended transit that leaves the focal base empty. Place wear, cracks and
other state marks outside the eyes, brows and mouth so the reaction stays clear.

For divided environments, build each panel from its own foreground, subject and
background planes. Keep the divider and panel clipping in screen coordinates;
apply subject motion inside the panel. A lateral track can introduce another
case while preserving the previous case as a visual anchor. Change meaningful
silhouette features before using colour to distinguish cases.

For rooted organic subjects, deform the stem from a fixed contact point and
attach leaves to evaluated curve positions. Orient the tip from the curve's
actual tangent, with the renderer's axis convention made explicit. Let changes
propagate with staggered timing and interpolate material colours continuously;
a threshold palette swap can flash during a gradual reaction. Paired states
are illustrative comparisons unless the production has evidence for measured
response rates or a causal experiment.

For an opening or growth action, preserve the contact point while the body
changes size. Separate the outer surfaces from the interior and let an opening
become visible before a new structure emerges through it. Clip material planes
to both the moving part and its outer contour; clipping only to one side of a
seam can expose rectangular paint bounds. Draw branches from evaluated points
on the main curve, and delay them until the growing tip has passed those points.
Inspect the opening onset, the first visible emergence and the final tip within
the camera bounds. Connect small impacts to existing timeline events when their
sound is already authored. Such compressed actions are explanatory illustrations
unless the production supplies a validated physical or biological model.

Carry the subject's established state across a change of camera or scale:
opened surfaces, connected structures and held objects should not reset at the
cut. Transform the related parts together, then continue their action. When a
growing tip reaches an attachment point, ramp the attached part's size and
opening continuously from zero; a binary visibility gate can reveal a part
already at full size. Inspect both sides of the cut and the first visible
attachment frames.

Construct an asymmetric curved shape in one local orientation, then mirror the
whole shape when reversing its lean. Flipping only some control points can
create a crossing or an unintended needle. Let material marks follow the
surface curvature. Inspect background tangencies at action onsets: a separate
blade or contour touching the subject can imply an attachment or growth that
has not occurred yet. Clear that overlap without flattening the whole scene.

In a comparison, let gaze anticipate the explanatory event before posture settles
into a distinct reaction. Keep the associated symbol and the actor at different
value levels so both remain legible without competing. A darker face may need
simpler eyes, but that change also changes how a blink must be drawn: compress
solid eye shapes continuously instead of clipping them into rectangular slits,
and avoid adding a second closed-eye mark over a still-visible eye. Inspect the
intermediate blink frames and the cut back to the ordinary lighting treatment.

For reach actions, solve both arm segments to a single hand target and choose
the elbow side deliberately. Check the resulting joint lengths and whether the
target is actually reachable. A clamped reach is not proof of contact. Derive a
held tool and its payload from the hand transform. Use the same world position,
scale and orientation at pickup and placement; remove the source copy at pickup
and replace the travelling copy with its settled representation at placement.
Keep labels, glass fronts and other foreground surfaces above the moving prop
when the scene requires that occlusion.

When a connected action is hard to read, first distinguish size from contrast.
A large subject can still disappear against a similar-value background. Adjust
the environment before enlarging the subject unnecessarily. If the group does
need enlargement, transform the actor, held prop, emitted stream, receiving
surface and contact shadow together. Keep fixed screen labels outside that group.

Mirroring a vehicle changes wheel-rotation handedness. Derive wheel angle from
world travel, then account for the reflected transform. Set each axle height
from its own radius and a shared ground line. A front-facing silhouette and a
correct trajectory still do not establish rolling contact by themselves.

Secondary motion may include drift, particles, environmental sway and blinks.
These should support the action, not substitute for it. Metric differences can
reveal a motion deficit, but adding arbitrary particles to increase a score is
not a remedy for weak acting.

Change scale when a new relationship becomes visible. Track a moving subject;
reveal a mechanism through a cutaway; pull back to connect a local event to its
context. Author inserts when a long explanation otherwise becomes a fixed
composition. Keep a common world-to-screen transform for connected objects.
For an inhabited landscape, an establishing view, a closer action and a return
to the larger setting can make a small event readable. Place the closer cut
before contact so the viewer can follow the approach and settlement. Keep the
actor and its ground shadow in one coordinate system through the scale change.
Check the two frames around the cut, the contact interval and the whole return
move; a larger frame-change value alone does not establish better pacing.
Choose transitions from the relationship between shots. A clear cut may be
enough; repeating a decorative wipe at every boundary weakens that relationship.
When switching between a wide view and an insert, retain the same material and
shape treatment. Inspect consecutive frames around contact, release and cut
boundaries, then inspect the encoded sequence at playback speed.

Narration segments do not have to end with a picture cut. When a subject connects
two explanations, carry its screen position, scale, angle and expression across
the boundary, drawing it once while the surrounding environments change on their
own clocks. Reveal internal structure only after incompatible facial or surface
features have cleared. Move the associated sound cues to the visible action,
even when that action begins before the narration boundary. Comparing adjacent
rendered poses can detect an accidental jump; it does not establish artistic
equivalence to a reference.

For travel through an environment, keep objects at fixed world coordinates and
move the camera. Project all connected surfaces with the same focal length and
vanishing point. Clip polygons against the near plane before projecting them;
clamping vertices independently can make surfaces stretch across the frame.
Inspect the entire camera path for occlusion changes. Sorting by average depth
works only for suitably separated geometry; intersecting surfaces need splitting
or a depth buffer. Draw a room's enclosing background before the contained
objects when using an explicit painter's ordering. Repeated geometry should not
reset or wrap visibly as the camera passes it.

When a long surface spans an occluding structure, split it at that structure's
depth boundaries before painter sorting. A single average depth can place its
far portion over the nearer structure. Derive floor reflections by mirroring
the same geometry about the receiving plane, then clip and soften them within
that plane. Keep reflected detail subordinate to the direct view and check its
attachment throughout the camera move. Local light and broad value separation
should clarify depth before additional repeated labels or surface decoration.

Define a transport path once and use it both for the drawn route and the moving
object. A curved connector with a separately interpolated straight-line marker
does not show the same motion. Check the complete swept area, including the
object's extent above and below its anchor, against titles and foreground
surfaces. At arrival, place the object's contact edge on the receiving surface;
matching only its centre does not establish contact. Replace an empty receiver
with the arriving representation on one ownership boundary, without duplicate
containers during a fade.

For repeated objects entering a receiver, distinguish objects still held,
objects in transit and objects already settled. Use the same positions, sizes
and orientations at each handoff. Ground the first settled layer on the actual
interior floor and subsequent layers on the contents beneath; an evenly spaced
array can read as suspension even when every arrival is correctly timed.
Keep front rims, transparent walls and interior contents in a deliberate drawing
order, and inspect the complete release and closure sequence. A held closure
should enter continuously with the hand, settle on the opening, then remain in
place when the hand withdraws.

For a route assembled from several Bézier segments, align the incoming and
outgoing tangents at every join. Sample cumulative distance when a traveller
needs a consistent speed along bends. Check stroke width and cap geometry at
container openings: a rounded or oblique cap can protrude above a closed lid
even when the centreline ends correctly. Keep warning symbols and identity
labels in separate areas so the change of state remains legible.

When a mechanism splits or joins moving elements, preserve identity across the
boundary: derive the last joined state and first separated state from the same
world coordinates, sizes and orientation. Do not apply a parent scale to only
one side of the handoff. During a shape morph, clip highlights and internal
creases to the current contour, not the original silhouette. Separate the
observable result from decorative motion, and document which scientific shapes
are metaphors rather than simulations or measured structures.

For an object entering from offscreen, place its entire initial silhouette beyond
the visible bounds after applying the world-to-screen transform. An anchor that
looks remote in local coordinates may still be onscreen. Inspect the first
visible frame of every repeated entrance; a correct exit does not validate the
next entrance.

Give a character explicit attention targets and interpolate gaze, eyelids,
brows and mouth continuously from the shot clock. Place a blink around a change
of attention when it helps the acting; do not cover a half-open pupil with a
closed-eye stroke. Compute ground support from the deformed, rotated silhouette
so squash and lean preserve contact. Multiply nested opacity by inherited
opacity: replacing it can leave a child visible after its parent disappears.
Fade cyclic particles at their reset boundary or reset them fully offscreen.

Check the camera's complete visible footprint, including its first and last
frames. Ground polygons and tint layers need enough overscan to cover that area;
a plate extending beyond the viewport does not make its overlays cover it.
For a growing stroke, omit drawing at zero progress and interpolate its control
points as well as its width. Some drawing APIs ignore a zero line width and keep
the previous value. Place emerging geometry behind its enclosing object when
the attachment is meant to be occluded.

## Shared timing and sound

Use integer frames and half-open shot intervals. Derive state directly from a
frame request so random access and offline rendering agree. Voice timing comes
from the actual generated waveform. A change to a visual action must update its
named audio event on the same timeline.

Write a musical motif with harmonic movement, changing orchestration and a
resolution. Add environmental sound and material-specific effects. Keep stems
separate, duck the score from real voice activity, and use silence selectively.
Record the provenance of TTS, instruments, generated images and supplied assets
in the external production. Never put credentials or provider jobs in the library.

Vary the musical role of each passage: sparse questions, a moving accompaniment,
a quieter warning and a resolved ending need different phrasing and rests.
Adding instruments to a constant ostinato does not create that development.
Write melody against the actual harmony, retain common tones across changes,
and let sustained notes and reverberation finish naturally. Check MIDI note
lifetimes so a repeated pitch is not cut short by an earlier note-off.

Use continuous gain curves for musical entrances, dramatic dips and exits.
Avoid rectangular attenuation windows and ducking that follows each syllable;
leave enough release time to connect words into phrases. Ride the levels of
sparse and full ensembles separately, preserving narration headroom when the
orchestration grows. Retain instrumental stems for revision. Signal checks can
find clipping, discontinuities and scheduling faults, but do not replace a
listening review of phrasing, timbre and the balance with speech.

## Compare, revise and describe evidence

1. Decode the entire reference and candidate in presentation order. Record frame
   counts, rate, timestamps and descriptors for every frame. Preserve the
   reference's full coverage even when advertisements or credits are excluded
   from the style sample.
2. Declare comparable visual roles and reference windows before measuring the
   candidate. Map shot phase within each role; do not nearest-neighbour search
   the entire reference for only flattering matches.
3. Compare value, saturation, edge/flat-area balance, composition occupancy and
   consecutive-frame change. These are diagnostics, not an artistic percentage.
   Changing subjects invalidates literal pixel-equivalence tests.
4. Inspect full-resolution stills, every transition and continuous action spans
   in the encoded film. A generated atlas has not been inspected until viewed.
   A thumbnail inspection has not established full-resolution quality.
5. Fix observed causes, preserve before/after evidence externally, re-render and
   invalidate reviews affected by the change. Measure the final encoded audio
   and video; play through the end and test seeking.

Report automated all-frame coverage, inspected images, motion playback,
subjective listening and unresolved differences separately. A successful render,
ASR match or close feature distribution cannot certify “100% style equivalence”.
A requested acceptance target remains unmet until supported by the relevant
review; do not silently lower it.

### Curved objects and surface detail

When an object must read as a sphere, draw its surface markings in object space.
Project coastlines, patches, clouds or panels through the same surface mapping,
so their widths compress toward the silhouette. A flat path clipped by a circle
does not establish spherical perspective. Keep the illumination normal field
shared across every material; a light/dark boundary must continue across a
surface marking rather than stop at its edge.

Use restrained surface rotation to reveal volume while keeping the composition
readable. Derive orientation directly from the requested frame. Review the
smallest and largest rendered scales, the silhouette, polar texture seams,
occlusions and transitions. Atmospheric emission belongs outside the volume;
keep surface highlights and shadows coherent with one light direction.

For rasterized procedural surfaces, preserve transparent coverage at the edge,
use a texture resolution appropriate to the largest displayed radius, and
inspect the encoded result for crawl or aliasing during rotation. Shading or
geometric correctness alone does not establish resemblance to a reference.

Compare the illuminated fraction as well as the surface projection. A correct
sphere can still look unlike a reference if its visible day side is too broad
or its night-side ocean is too bright. Inspect reference, prior render and new
render together under the same color conversion before accepting the lighting
change. Treat that visual decision separately from a successful technical
render or frame-difference audit.

### Environment scale and shot differentiation

Build related environments around each shot's camera height and subject size.
A ground-level close-up, a growth demonstration and a human action need different
background openings and foreground silhouettes, even when they share a location.
Do not repeat a complete background plate and rely only on a dark overlay to
make it serve all three roles.

Separate distant shapes, the action plane and near vegetation. Use lower
contrast and smaller parallax in the distance; keep the contact chain visible
on the action plane. For a physical action, scale the person, tool, stream and
receiving surface together. Foreground leaves can frame the subject but must
not hide the gesture or its result. Test the first frame, maximum growth or
reach, and final frame before encoding; a midpoint can miss a clipped leaf,
a zero-length rounded stroke or an object entering behind foreground foliage.

Judge scene differentiation and reference resemblance separately. Brighter
colors and new geometry may fix repetition while still leaving the environment
too sparse or the silhouettes too regular. Keep a fixed reference comparison
alongside the prior and revised picture, and retain unresolved differences.

### Character construction without losing contact

Compare the head-to-body proportion, cheek contour, eye treatment, hand shape
and sleeve silhouette within equivalent shot sizes. Changing a thin profile
into a broader three-quarter face requires redrawing the outer cheek and nose
together; retaining the old profile projection can accidentally depict two
noses. Keep the new cast original rather than carrying over reference identities.

Keep the rig's shoulder, hand, held-object and floor anchors explicit while
redrawing clothing and limbs. Curved sleeve contours can soften an angular arm
without changing the tool tip or grip location. Add finger and thumb silhouettes
at the hand anchor, then inspect the actual grip at pickup, lift and release.
Interpolate expression shapes through their onset; switching abruptly between a
closed-mouth stroke and an open smile can create a visible pop even when gaze
and head rotation are smooth. Check the same actor at a distant working scale
and a close interaction scale before accepting the new construction.

### Camera coverage beyond the frame

Extend opaque environment layers beyond the entire animated camera footprint.
An off-centre camera can expose transparent edges even when a scene fills the
nominal canvas. Inspect the first, last and extremal camera frames at native
resolution and examine source alpha separately from encoded black pixels.
Continue wall/floor boundaries into the overscan area; preserve acting and
contact anchors when repairing coverage. Recheck the encoded sequence, because
a source alpha check alone does not validate the delivered picture.

### Planted feet and directed gestures

Define a foot's stance and swing phases independently of torso bob and rotation.
During stance, subtract the body's travelled distance from the local foot
position so the world-space sole remains planted. During swing, return the foot
smoothly and lift it above the floor. Derive root travel and gait phase from the
same progress variable, including acceleration and deceleration. Keep rounded
leg caps above the sole; a mathematically correct foot anchor can still leave a
visible nub below the ground.

Facing belongs to the entire character rig. Mirror limb reach, gaze and travel
consistently when the actor addresses a subject on the other side. Check the
start, stopping pose, gesture apex and exit against foreground occlusion. A
larger, more readable actor may need a lateral adjustment to keep antennae or
hands clear. Use a numerical stance check to find sliding, then inspect decoded
step sequences and practical-size playback. Neither check replaces the other.

### Dense vegetation and atmospheric joins

Arrange vegetation in overlapping clusters with different heights, spacing,
curvature and leaf widths. Give stems and leaves a small number of consistent
light and shade planes. Reduce contrast toward the distance while retaining a
clear opening around the action; adding more identical stalks increases density
without creating natural variety.

Treat mist as a compositing layer with enough opacity to conceal hidden geometry
ends. A translucent wash can reveal aligned stem cutoffs as a straight line.
Inspect the join at native resolution and during camera movement, then check
that the correction does not erase foreground silhouettes. Round or overlap
growing stems where they meet their base, and verify the earliest visible growth
as well as the mature shape. Compare the encoded result alongside both the
reference and prior revision at equivalent scales; improved density alone does
not establish a complete style match.

### Oblique workspaces and inspection inserts

Choose the room projection before placing actors and tools. Use asymmetric
equipment groups, visible top and side planes, and foreground overlap to make
the working surface belong to the room. Continue the floor and walls beyond
the camera footprint. A shallow diagonal camera drift can reinforce depth
without displacing the hand-to-tool relationship.

Derive a held tool's grip from its actual world-space pose, then transform that
anchor into the actor's local coordinates, including scale and facing. Check
the full action for reach limits and silhouette clearance. A numerically valid
arm can still fold across the face or overlap an unrelated instrument; move
the actor or instrument and inspect the revised gesture. Allow the far arm's
bend direction to differ when holding a board or another broad object.

Stage attention through the head, torso and feet together. A side-facing worker
and a rear-facing observer can direct the viewer toward an action without both
characters presenting their faces to the camera. Draw distinct side and back
silhouettes, including coat seams, ears and the visible cheek; do not merely
remove one eye from a frontal head. Keep facial strokes within the turning
contour. Preserve the tool's contact anchors when changing the body drawing.
If a broad prop narrows in perspective, apply that same transform to both hand
targets so the grip follows the visible edges throughout the gesture.

An inspection insert must preserve the object's state at the cut. Begin its
transformation only after that state is established, or explicitly motivate a
time jump. For a magnifying lens, redraw the complete underlying scene inside
the clipped, scaled lens region. Filling that region with a flat surface colour
fails when the lens crosses the surface edge. Check the lens entering, scanning
and withdrawing at native resolution and across consecutive encoded frames.

### Foreground actions and shared trajectories

Stage a near actor and a distant activity at clearly different scales, with
their supporting surfaces visible. Reassess the prop and receiving object when
enlarging the actor; enlarging every element equally can preserve an implausible
relationship. Keep tree branches covered by their foliage or visibly continued
beyond the frame, and check secondary work surfaces for actual support.

Derive both hands from the held object's transform when the object tilts. Verify
the direction of the tilt from its visible outlet and intended action, not only
from the sign of an angle. Draw a fluid stream and its moving drops from the same
curve; independently interpolated drops can leave the stream. Keep the receiving
surface fixed, and synchronize onset and settling with the existing sound cues.

For a close facial expression, morph a single mouth contour through the change.
Crossfading a closed stroke over an open shape may reveal two mouths in the
middle frames. Clip teeth and highlights to the changing contour, then inspect
the intermediate expression as well as its endpoints.

### Closing scale and foreground overlap

Build a closing image around a clear relationship between a near silhouette and
an offset focal object. Crop the near silhouette deliberately, while keeping its
eyes and other essential features inside the entire camera footprint. Reserve
a quiet area for the final title, and let the pictured action settle before the
title appears. A ring of repeated symbols can weaken this relationship even
when all its individual drawings are attractive.

Keep foreground curves attached to stable roots and derive attached leaves from
the curve's position and tangent. Separate growth along the curve from leaf
unfolding; a late sound accent can mark the latter without resetting the stem.
Begin a newly appearing leaf at zero area. A positive minimum scale makes the
first visible frame pop even when the easing curve itself starts smoothly;
inspect the frame before onset, onset and the immediately following frames.
Use distinct front and back layers around the focal object and inspect their
occlusion throughout the shot. Check coloured shadow treatments in the rendered
image: a cool translucent overlay on a warm surface can turn grey instead of
producing the intended saturated shade.

### Traveling habitat views

Put grounded actors, plants and trees into one depth order based on their ground
positions. Drawing every tree before every actor can place an animal visually
on top of a nearer canopy. Check both the ground point and the shadow against
riverbanks and raised terrain after moving an object.

Choose a contact shadow's colour from the surface receiving it. An actor crossing
from grass to warm soil needs a coherent shadow on each material; a fixed green
ellipse can look like a detached patch. Vary cluster silhouettes around stable
trunk and root anchors so the variation does not break ground contact. Check the
result at both the widest and closest camera positions.

Reserve clearance for an actor's entire moving silhouette, including wings,
antennae and its shadow. Keeping decoration away from the ground anchor alone
can still leave a stem crossing the body. Inspect the swept corridor through the
whole action before placing small foreground plants.

Schedule a small interaction for when the traveling camera has revealed it.
A correct action can happen entirely outside the view if its timing ignores
camera travel. Derive the receiving contact point from the actor's actual pose,
keep that point fixed in the environment, and change the receiving object only
after contact. Inspect approach, contact and withdrawal separately.