Pocket Memory

Pocket Memory

Turn one photograph from a point in time into one felt moment:

ONE IMAGE → ONE TIME MODE → FOUR DEPENDENT FRAMINGS → ONE FELT MOMENT

Choose the source's real time mechanism before cropping:

MOTION TIME: ESTABLISH → CONTACT → TRAVEL → ARRIVAL / RESIDUE
ATTENTION TIME: FOUR SOURCE VIEWS → GRADIENT-FIRST ORDER

Motion Time follows a carrier that visibly moves through the photograph. Attention Time selects one near-original Source Anchor plus three connected source views, then orders all four by the strongest visual gradient. The Source Anchor is evidence, not a mandatory opening frame: it may enter, expand, turn, or settle the sequence. Adjacent frames change dominant roles while sharing enough source-visible material to prove the handoff. Layout, typography, softness, and finish may support either mode but must never manufacture its temporal logic.

Return:

1. one finished full-bleed 2×2 raster containing exactly four source-derived frames and, when appropriate, one subtitle block;
2. the exact line used in the artwork as selectable text outside the image;
3. one brief sentence explaining the factual anchor, visual path, and arrival point in the user's language unless requested otherwise.

When the source filename is known, save the finished raster as <source-basename>-pocket-memory.png, for example sunrise.jpg → sunrise-pocket-memory.png. If that path already exists and replacement was not explicitly requested, append -v2, -v3, and so on.

Prefer deterministic raster compositing because it preserves source pixels exactly. An image-editing model may render the layout only when it can preserve the supplied photograph as crop-only source material. Never ask a model to generate four scenes.

Runtime Capability Check

Before composing, verify that the host can inspect the source at useful detail and arrange faithful crops into one raster artifact. Do not require a particular vendor, model, API, or agent.

If the required capability is unavailable, state the limitation plainly. Offer an action-path crop plan and subtitle draft only as a fallback, and only when the user accepts it. The skill does not bundle a model, API key, external account, or paid service.

Source, Consent, and Privacy

• A supplied photo plus a transformation request is consent to use it with the required image-editing service; do not ask again.
• Inspect the image before composing. Use only the minimum required reference image(s).
• Do not browse, share, separately upload, save, or commit the source photo unless the user asks.
• Treat the user's account of the moment as authoritative for meaning and copy.
• Visual inference may choose where attention travels, but it must not add photographic events or content.

Decision Priority

When constraints conflict, resolve them in this order:

1. Preserve the user's remembered meaning and explicit facts.
2. Preserve the source's people, relationships, place, objects, geometry, light, weather, and action state.
3. Keep every photographic panel traceable to the supplied image.
4. Choose Motion Time only when a concrete carrier visibly travels; otherwise choose Attention Time.
5. Build either one action relay or one attention relay whose non-anchor views have different dominant subjects or relationships.
6. Give every adjacent pair in the final reading order a concrete shared anchor and an unbroken handoff.
7. Place the Source Anchor where tonal, spatial, scale, and information gradients make the sequence strongest; do not default it to Frame 1.
8. Make the chosen order impossible to shuffle without damage.
9. Use subtitle language to recognize the time relation without explaining the artwork.
10. Keep the 2×2 structure full-bleed and completely seamless.

The image sequence must still carry time after the subtitle is removed.

Build the Memory Card

Before writing or editing, resolve an internal Memory Card:

• Visible facts: people, objects, place cues, time of day, weather, light, colors, gestures, source text, and spatial relationships.
• Recalled facts: what the user explicitly says happened, felt, sounded like, or later changed.
• Emotional hinge: the smallest tension between what the moment felt like then and what it means now.
• Scene anchor: the near-original view that preserves the defining composition.
• Time mode: Motion Time for a genuinely traveling carrier; Attention Time for a static, simultaneous, diffuse, or contemplative scene.
• Visual-path candidates: source-visible directional structures such as light, gaze, gesture, road, shoreline, wave, shadow, body orientation, or visible movement traces.
• Path mechanism: decide whether the carrier is visibly traveling or diffusely filling the scene. A diffuse field must be narrowed to one source-visible breach, one directional transmission, and one unique receiver; do not treat every illuminated region as another step.
• Dominant visual path: the single route that connects a contact point to a meaningful arrival or residue without jumping across unrelated regions.
• Contact point: where two visible elements meet, align, touch, look toward one another, or begin a transmission.
• Travel carrier: the continuous light band, contour, road, gaze, gesture, wave, shadow, or texture that visibly carries the path onward.
• Arrival point: where that carrier reaches, settles, breaks, disperses, exits, or leaves a residue.
• Action relays: the shared source element and action verb passed between every adjacent pair.
• Semantic decay: in Motion Time, how Frames 3–4 become less like complete standalone photographs and more like movement and residue through crop choice. Attention Time instead changes dominant roles according to the selected ENTRY → CONTINUE → TURN / REVEAL → SETTLE order.
• Attention content set: one Source Anchor plus three source-grounded views A, B, and C, each with a different dominant subject or relationship.
• Sequence functions: after ordering, label the four positions ENTRY → CONTINUE → TURN / REVEAL → SETTLE. The Source Anchor may fill any one of these functions.
• Attention rectangles: all three derived views remain inside the Source Anchor's source extent; in the chosen order, adjacent views share only what is needed to carry a concrete handoff and may not be independent attractions.
• Depth relay: when the source itself places a sharp directional cue before a naturally soft destination, map one coherent source crop as sharp cue → physical threshold → soft field. Record the exact cue, threshold, and source-visible direction; softness is evidence from the lens, not an effect to add.
• Order candidates: compare plausible permutations of the four content views. Reject any with a broken handoff, then choose the one with the clearest tonal, scale, information-density, and spatial progression.
• Tonal route: map the source's available dark/cool, neutral/midtone, and warm/highlight zones. Choose a truthful route through at least two distinct tonal zones when the source contains them; never create contrast with separate panel grading.
• Salience override: identify any partial face, body, bridge, building, vehicle, hand, sign, or other identity-bearing form whose size, brightness, contrast, or position makes it visually dominant despite being intended as a handoff.
• Subtitle hinge: the concrete temporal relation the line may name—contact, travel, delay, arrival, persistence, or disappearance.
• Uncertainty boundary: what the source does not justify and must not be claimed or depicted.

Choose the time mode and its four dependent framings before choosing attractive crops. Read references/scene-grammar.md for detailed construction.

Time Mode Router

Motion Time

Use Motion Time only when the source already contains a carrier whose contact, route, and arrival can be pointed to without conceptual explanation. Waves, gestures, roads, gaze, narrow reflections, and directional movement traces often qualify.

ESTABLISH → CONTACT → TRAVEL → ARRIVAL / RESIDUE

Attention Time

Use Attention Time when the scene is static, simultaneous, diffuse, monumental, or better understood as the duration of looking. Do not invent a physical journey for light, fog, atmosphere, or color merely because several regions share it.

CONTENT: SOURCE ANCHOR + A + B + C
ORDER: ENTRY → CONTINUE → TURN / REVEAL → SETTLE

• The Source Anchor is the near-original view that proves the photograph's people, objects, light, and spatial relationships. It may occupy any frame.
• A, B, and C are different source-grounded subjects or relationships connected by a real contour, route, ray, reflection, gaze, gesture, road, river edge, or other visible handoff.
• Assign final sequence functions only after choosing the order: Entry invites the eye, Continue carries it, Turn/Reveal changes scale or context, and Settle provides the strongest endpoint.
• All derived crops lie inside the Source Anchor's source extent. Adjacent frames in the final order must share one small, unmistakable handoff zone.
• The dominant subject may appear in adjacent frames only when it changes from primary subject to a genuinely secondary handoff. It may not remain centered, similarly scaled, and semantically dominant twice.

Do not lock the Source Anchor to Frame 1. Try its plausible positions and compare the surviving orders. A nonstandard order such as B → A → SOURCE → C is valid when it produces a clearer gradient and every cut still has a factual handoff; it is not a reusable recipe.

Choose the order through this priority:

1. all three adjacent handoffs remain concrete and source-visible;
2. dominant subjects do not repeat;
3. hue, luminance, scale, information density, or spatial direction forms a smooth progression or one deliberate turn rather than a chaotic zigzag;
4. the final frame provides the strongest resolution.

Visual Path Contract

A valid path must pass four tests:

1. Continuity: the eye can follow it without jumping to an unrelated region.
2. Coverage: it supplies a contact, a visible route of travel, and an arrival or residue.
3. Specificity: it belongs to this photograph rather than a generic texture or shared color.
4. Meaning: it supports the user's remembered feeling without inventing facts.

If no candidate passes, do not force a finished sequence. Explain that the source lacks a reliable four-action path and offer the strongest partial crop plan.

Motion Time Contract

Create exactly one full-bleed 2×2 sequence, read top-left → top-right → bottom-left → bottom-right:

1. ESTABLISH — where: retain the full photograph or broadest defining crop. Preserve context and make the complete route discoverable.
2. CONTACT — connection: crop toward the path's contact point. Show a visible relationship beginning or becoming legible; do not merely make the main subject larger.
3. TRAVEL — transmission: follow the same carrier away from Contact. This is the most dynamic frame. Favor a crop whose direction, continuity, and reduced surrounding context make the carrier feel actively transmitted through the source.
4. ARRIVAL / RESIDUE — after contact: end where the carrier reaches, breaks, settles, disperses, or exits. Reduce standalone-photo completeness and let texture, fragments, or visible dissipation dominate. This frame must answer: “Where did the path arrive?”

The progression is:

complete → contact → transmission → arrival / dissolution

It must not collapse into:

wide shot → close-up → detail → detail

Before rendering, name the relay and verb for every adjacent pair:

Establish → Contact: [element] becomes connected
Contact → Travel: [same element] begins or continues moving
Travel → Arrival: [same element] reaches, breaks, settles, disperses, or exits

Similarity of color alone is not a relay. Each later frame must behave like the previous frame's second half.

Use exact source crop, enlargement, and translation. Apply one shared non-destructive tonal treatment at most. A very slight source-preserving reduction of clarity or directional softness may attenuate Travel or Arrival only when the crop sequence already passes without it; it must not move highlights, invent motion, create content, or become visible as an effect.

Do not mirror, rotate, repeat, redraw, outpaint, inpaint, erase objects, synthesize viewpoints, alter light or weather, advance gestures, or create another shutter time.

Attention Time Contract

Create one Source Anchor and three different source-grounded views before assigning them to the 2×2 reading order:

• SOURCE ANCHOR: full photograph or broadest defining crop. It preserves the factual composition but has no fixed frame number.
• VIEW A: one specific complete element or relationship plus a visible exit.
• VIEW B: a different dominant connector or relationship that inherits one handoff and exposes another.
• VIEW C: a different complete destination, residue, expansion, or contextual relation.

Now test plausible orders and label the chosen positions:

1. ENTRY: creates the clearest invitation into the visual path.
2. CONTINUE: inherits the first handoff and advances the gaze.
3. TURN / REVEAL: changes scale, context, direction, or tonal role without breaking continuity. The Source Anchor often works well here when the scene should gradually open.
4. SETTLE: supplies the strongest arrival, residue, or emotional resting point.

Hard rules:

• All derived views stay inside the Source Anchor's source extent. In the chosen order, every adjacent pair shares one visible handoff zone.
• The Source Anchor appears exactly once and may occupy Frame 1, 2, 3, or 4. Its position is chosen by continuity and gradient, not convention.
• Require different dominant subjects or relationships in adjacent frames. A repeated object is allowed only as a secondary handoff, never as the centered dominant subject twice.
• Apply the noun test to the final order: describe every adjacent pair with short noun phrases. If an adjacent pair receives the same phrase, rebuild or reorder.
• Apply the function test: Entry invites, Continue advances, Turn/Reveal changes the reading, and Settle resolves it. A medium bridge followed by a closer bridge fails even when crop scales differ.
• Shared color or atmosphere is not a handoff. Use one specific contour, object fragment, reflection line, ray, road, gesture, gaze, or edge.
• A naturally soft or negative-space Settle is valid only when the same source rectangle contains an inherited sharp cue, a real physical threshold, and a readable direction into the soft field. Empty blur alone is not a destination.
• Never detach, duplicate, reposition, extend, or redraw a source motif to make it “interact” with the background. Find one crop where foreground cue, threshold, and destination already coexist and align.
• Do not repeat the Source Anchor, and do not force a tighter crop when it would create a subjectless fragment.
• Do not mirror, rotate, regenerate, alter light, or simulate a later shutter time. The duration belongs to the gaze.

Gradient-first ordering gate

The original photograph is not automatically Frame 1. Compare viable permutations after crop selection:

• reject any order with an unrelated cut;
• prefer a smooth progression in at least two available dimensions: luminance/warmth, scale, information density, spatial direction, or semantic completeness;
• allow one deliberate turn, such as detail → source → whole reveal → residue, when it reads more clearly than a monotonic zoom;
• reject back-and-forth scale changes, warm-cool-warm noise, or a sequence whose best order is arbitrary;
• keep the ending stronger than a merely explanatory wide shot.

Run the order test: if another permutation preserves the same handoffs but produces a visibly smoother gradient and clearer landing, the current order fails.

Salience and subject-integrity gate

Intended hierarchy does not override what the eye actually sees. A partial identity-bearing form becomes the de facto subject when it is large, central, unusually bright, high-contrast, or structurally distinctive.

• If a partial face, body, bridge tower, building, vehicle, hand, sign, or similar form is visually dominant, either show it complete and treat it as that frame's primary subject, or crop it out far enough that it reads only as a small low-salience handoff.
• Never use a bright half bridge, half face, or half building as a “secondary” connector while another weaker subject is supposed to dominate.
• A handoff fragment should be edge-bound, subordinate in area and contrast, and unnecessary for identifying the frame's primary noun.
• When a complete bridge is the correct subject for one frame, show the full identity-bearing structure there; make the next frame's primary subject the deck, road, water, reflection, or another connected role rather than another bridge portrait.

Run the salience override test: ignore the written crop plan and name what the eye notices first. If that answer is a clipped object the plan called secondary, the crop fails.

Subject and tonal separation gate

Before rendering, write this private map in the chosen order:

Frame 1: [content role] → handoff 12
Frame 2: [content role] → handoff 23
Frame 3: [content role] → handoff 34
Frame 4: [content role / settlement]

Then assign each frame a source-derived tonal role. When the photograph contains distinct tonal zones, the sequence must use at least two of them: for example dark/cool structure, neutral midtone, and warm highlight. Do not place all derived views inside the same warm sky, same yellow haze, same blue wall, or same low-contrast field merely because the path passes through it.

Run the swatch test: mentally blur or reduce all four frames to tiny color-value blocks. If the derived views read as interchangeable swatches or the chosen order produces an avoidable tonal zigzag, reselect or reorder around source-visible tonal boundaries while preserving the handoffs. Never solve the failure by grading panels independently.

For a sharp-to-soft Settle, retain enough source-native foreground or midground material to bridge the prior frame's value and hue into the softer destination. Do not crop down to an isolated background swatch or grade the destination to match.

Truth Boundary

Use this evidence ladder:

clearly visible source action path > source-grounded attention relay > user-guided attention > crop/reframe only > omit

User recollection may guide copy and path selection. It does not authorize depicting a gesture finishing, light shifting, people leaving, weather changing, or another uncaptured state.

Without explicit user support, never claim death, breakup, illness, permanent separation, “the last time,” “never again,” or a named life event. Do not change who was present to intensify emotion.

Copy Direction: Subtitle Mode

Write only after the bare four-action sequence works. Read references/writing-guide.md whenever authoring or revising copy.

• Use one short subtitle sentence, normally 4–16 English words or 6–24 Chinese characters.
• Prefer a concrete temporal relation carried by the path: first, then, later, still, reach, arrive, remain, fade, or a natural equivalent.
• When the visible path implies travel or delay, let the sentence name that delay or arrival rather than merely describe the scene.
• Do not force delay language onto a path that expresses stillness, gaze, repetition, or persistence more accurately.
• In Attention Time, prefer copy about noticing, nearly missing, following, lingering, or looking again when those ideas are supported by the user's account. Do not force the subtitle to describe an object moving.
• The subtitle may arise from a user-supported retrospective feeling rather than literally narrating the carrier. In that mode, it should sound like a later voice over the scene, not an immediate description of what the image shows.
• Keep emotion one degree quieter than the first draft.
• Do not use last, goodbye, forever, never again, or equivalents unless supplied by the user.
• Do not add a title, frame label, editorial caption, or quotation treatment.
• Use bilingual subtitles only when the user asks for or supplies both languages. Place the main-language line above a smaller translation.
• Echo every exact text element outside the image so it remains selectable.

Visual Direction

Full-bleed 2×2

• The four rectangular panels fill the entire canvas.
• Use no outer matte, outer frame, individual panel stroke, rounded corners, overlap, bleed, scatter, tilt, stagger, decorative mask, or free-form layout.
• The viewer should first perceive one continuous photographic field and only then notice four acts of looking.

Seamless Cut

• Use a true 0px gap: no seam, gutter, divider, line, tonal strip, or background visible between panels.
• The four crops meet edge-to-edge at one exact central vertical cut and one exact central horizontal cut.
• Do not add exposure softening, glow, shadow, fade, blur, color transition, or any other treatment along the cuts.
• The boundary exists only because the source crops change at that coordinate.
• There is no outer margin or frame. The four crops themselves are the whole artwork.
• The cuts do not create time. Removing the subtitle must leave the action relay intact.

Subtitle placement

• Treat the whole 2×2 composite as one scene.
• Default placement: one stable subtitle block centered in the lower-middle of the overall canvas, usually around 68%–82% of canvas height, spanning the bottom pair when necessary.
• The subtitle may cross the shared vertical cut between Frames 3 and 4 because it belongs to the whole felt moment. This is not a free-form cross-panel effect: it remains one conventional, centered subtitle block.
• Do not place the subtitle wholly inside Frame 1 or in an upper corner.
• If the default placement obscures the arrival carrier or becomes unreadable, place it inside the quieter part of Frame 4 while keeping lower-third subtitle behavior.
• Use a thin, quiet white or warm-white handwritten script, with natural connected strokes, restrained slant, conventional horizontal setting, and a very slight 1–2px shadow or glow for legibility.
• For bilingual copy, make the main-language line larger and the translation smaller directly below it.
• The handwriting must feel personal and incidental, not calligraphic, ornate, brush-lettered, childish, or like a decorative poster title. No serif quotation, outline, title scale, curved or diagonal setting, text box, or background band.
• The subtitle is an emotional channel, not a graphic protagonist.

Read references/visual-art-direction.md before confirming the seamless cut or subtitle placement.

Editing Workflow

1. Inspect the source at useful detail and separate visible facts from user-stated meaning.
2. Decide Motion Time versus Attention Time. Do not prefer Motion Time by default.
3. For Motion Time, select and map one valid carrier. For Attention Time, select one Source Anchor plus three different connected views; do not assign frame numbers yet.
4. Audit partial identity-bearing objects with the salience override test. Complete or remove any dominant fragment before ordering.
5. Compare viable reading-order permutations. Reject broken handoffs, then choose the strongest gradient and settlement; place the Source Anchor accordingly.
6. Name the carrier/action or attention handoff between every adjacent pair in the final order.
7. Construct the full-bleed 2×2 with plain source crops only.
8. Hide all typography. Run the mode-specific continuity, order, shuffle, noun, function, salience, swatch, and independent-crop tests.
9. In Motion Time, reduce semantic completeness in Frames 3–4 through crop selection. In Attention Time, verify distinct dominant roles, three visible handoffs, and one intentional gradient. Preserve tonal hierarchy in every mode. Use slight source-preserving softness only if the sequence already works without it.
10. Draft three subtitle candidates: one about contact, one about travel or delay, and one about arrival or persistence. Keep the shortest accurate line.
11. Join the four panels at exact 0px central cuts with no visible separator or transition.
12. Place one subtitle block at the overall lower-middle; use Frame 4 only when necessary for legibility.
13. Compile one editing brief with exact source regions, chosen order, handoffs, settlement, seamless layout, subtitle, and prohibitions.
14. Composite from exact source pixels whenever possible; inspect at full size and thumbnail size.
15. Read references/generation-and-quality.md and retry at most once for a core failure.
16. Return the artwork, exact subtitle, and concise rationale. Do not expose hidden reasoning or the full editing prompt unless asked.

Self-Check

A. Source

• Is every panel traceable to the supplied photograph?
• Are people, objects, geometry, waves, reflections, gestures, light, and weather unchanged?
• Did any uncaptured scene or hidden background appear?

B. Mode and continuity

• Was the mode chosen from source evidence rather than convenience?
• In Motion Time, do Contact, Travel, and Arrival carry one visible element through concrete actions?
• In Attention Time, does the chosen order contain one Source Anchor and three different dominant views, with a source-visible handoff between every adjacent pair?
• Was the Source Anchor placed by gradient and continuity rather than automatically assigned to Frame 1?
• Is every identity-bearing primary subject or relationship complete and recognizable rather than an arbitrary fragment?
• Does any bright or large partial identity-bearing form override the intended subject hierarchy?
• Does each frame perform a different function rather than merely increasing focal length?
• If Settle is naturally soft or spatial rather than object-led, does one source rectangle still show the inherited sharp cue, the physical threshold, and the direction into that field?

C. Time

• With the subtitle removed, does the sequence read as one action relay or one intentional entry → continue → turn/reveal → settle progression?
• Would exchanging any two frames damage the experience?
• Are the three derived views dependent steps rather than attractive independent crops?
• In Motion Time, do Frames 3–4 progressively lose standalone-photo completeness? In Attention Time, do the non-anchor views perform clearly different subject functions?
• Does the chosen order pass the gradient and swatch tests rather than collapsing into repeated tones or an avoidable tonal zigzag?
• Does the final crop preserve a source-native tonal bridge, rather than jumping from the established scene into an isolated blur or empty color field?

D. Cut

• Is the artwork full-bleed with no outer frame or individual panel strokes?
• Do all four panels meet at exact 0px central cuts?
• Is there no visible gap, line, tonal strip, background, glow, fade, or edge treatment?
• Does the work feel like one photographic field interrupted by crop changes rather than four framed boxes?

E. Subtitle

• Is there only one conventional lower-middle subtitle block?
• Does it belong to the whole moment rather than Frame 1 or a page layout?
• Is it the same quiet, thin white/warm-white handwritten subtitle style with only minimal legibility support?
• Does the sentence name a real temporal relation without becoming a slogan?

The result fails if Motion Time reads as four nouns, if Attention Time automatically places the Source Anchor first despite a stronger viable order, repeats a dominant subject in adjacent frames, lets a high-salience half-object hijack a frame, ends on unanchored blur, breaks a handoff, produces an avoidable tonal zigzag or abrupt final swatch, or loses time when the subtitle is removed. Rebuild the content set or order; do not repair it with styling.

Revision Behavior

• “更有时间感”：first audit the mode. Strengthen a real carrier in Motion Time, or reorder Attention Time around a stronger gradient and settlement; do not add events.
• “还是四个 crop”：in Motion Time, rebuild Frames 3–4 as transmission and residue; in Attention Time, rebuild the content set and its three final-order handoffs.
• “概念上通、视觉上断”：require a concrete source fragment at both handoffs; shared color or an explained concept is insufficient.
• “第二、三格主体一样”：apply the noun test. Keep the old subject only as a secondary edge handoff and make a different connector dominant in Frame 3.
• “第三、四格太像”：apply the noun and function tests; the later frame must change the dominant role and land on a different complete subject.
• “最后一格没有主体”：restore one complete settling subject. If the source supports an intentional spatial or naturally soft ending, use one coherent crop containing an inherited sharp cue, a physical threshold, and a directional soft field; empty blur alone still fails.
• “第四格和第一格太像”：choose a distinct settlement crop with different coverage and primary structure; the ending may not replay the Source Anchor.
• “后三格色调太像”：run the swatch test and reroute through source-visible dark/cool, neutral, and warm/highlight zones. At least two tonal roles must differ when the source permits; never grade panels independently.
• “第四格色调突然跳开”：keep more source-native foreground or midground material inside Settle so hue and value cross a real visual threshold before reaching the destination; do not repair the jump with a separate grade.
• “让清晰线索和模糊远方互动”：use a source-existing sharp-to-soft depth relay. Align the inherited cue, real object edge or spatial threshold, and naturally soft destination within one crop; never copy the cue onto the background or generate a clearer road ahead.
• “原图不该在第一格”：test the Source Anchor in every plausible slot; keep the order with the strongest handoffs, smoothest gradient, and clearest settlement.
• “只露出半个主体”：run the salience override test. If the fragment dominates, either restore the complete subject and make it primary, or crop it out enough to become genuinely subordinate.
• “文案像设计语句”：switch fully to lower-middle subtitle mode and rewrite around contact, delay, arrival, or persistence.
• “文字放错格”：treat the whole composite as one scene; center the subtitle across the bottom pair or move it into Frame 4.
• “像四格漫画”：remove outer matte, individual strokes, and all gutters; join the four crops edge-to-edge.
• “中间还有缝”：set both central cuts to a true 0px gap and remove every divider, strip, fade, glow, and edge treatment.
• “更像原图”：restore exact crop provenance, geometry, palette, clothing, objects, waves, and reflections.
• “只改文案”：keep crops and layout unchanged.

Delivery Format

![Pocket Memory four-frame scene](absolute-image-path-or-rendered-image)

*Exact subtitle from the artwork, if any.*

**场景说明：** [One sentence naming the factual anchor, visual carrier, and arrival point.]

When a local saved path exists, include it. Do not call the result a historical reconstruction, and do not add promotional copy.