# Moments R Forever — AI Enhance, Shared Credits and Human Photo Services

## Implementation prompt

Act as a senior product designer, accessibility specialist and full-stack engineer. Extend the EXISTING Moments R Forever Site B. Inspect its current source, routes, AI prompt studio, 100 animation presets, photo viewers/editors, Story Builder, Interview Builder, photo booth, account services, storage and billing before changing anything. Preserve saved memories, originals, edits, permissions, working tools and Site B's current audience. Do not modify Site A or its DNS.

This is an implementation task, not a mockup-only task. Complete the supported work, test it and publish the verified update to Site B. Do not claim paid generation works without a connected, tested provider and billing backend. If infrastructure or credentials are missing, finish independent UI/prompt work, disable unavailable paid actions, and report the exact blocker.

### 1. One recognizable AI Enhance action wherever a photo appears

Create a reusable `AIEnhanceAction` with a clear sparkle/photo icon and accessible name **“AI Enhance photo.”** Place it consistently on individual photo cards, selected-photo toolbars, full-size viewers and photo editors across shoeboxes, Story Builder, interview photo attachments, scanning results and the photo booth wherever those surfaces exist. Inventory all photo-rendering surfaces; report any missed or unsupported location.

For thumbnails, use the existing overflow menu when an overlay would cover faces or overcrowd controls. Make the action discoverable without hover. On desktop offer the label “AI Enhance”; on phones keep a visible icon with an accessible label and a plainly labeled menu item. Use at least 44×44px touch targets, visible focus states, keyboard activation, adequate contrast and reduced-motion support.

The action must retain the selected photo, source, orientation, permissions, originating screen and intended story/interview destination. Do not make users find or upload that photo again. Enable it only for a photo the current user may process. A guest's right to view a photo must not automatically grant permission to send it to an AI provider.

Do not initiate an upload, generation, payment or camera permission merely by opening this action.

### 2. A simple two-card choice

Open an accessible “What would you like to do?” screen or panel with two large cards:

**Bring this moment to life — Photo to Video**
“Add a little movement: a blink, a smile, a shared glance or a gentle dance.”

**Digital Makeover — Optional Photo Retouching**
“Choose a few finishing touches. Keep what makes you, you.”

Show the selected photo and a clear Back/Cancel action. Explain that Photo to Video aims to preserve appearance, while Digital Makeover intentionally changes only the details the user selects. Do not apply makeover instructions to animation mode by default.

Keep all makeover options unchecked initially. Never describe a person as defective, unattractive or “fat.” Use warm, neutral wording. Do not suggest a makeover automatically because a face or body was detected.

### 3. Photo to Video

Reuse the existing 100 presets, grouped into 10 categories, with TWO dependent dropdowns:
1. Movement group.
2. Specific movement, updated when the first choice changes.

Retain the people-count controls for one person, two people and groups, as well as scene/pet compatibility. Include custom movement text, an optional “Give extra attention to…” note, and an advanced “Edit master prompt” section. Offer 2-, 3- and 4-second story excerpts. Validate actual model durations; show the complete billable generation length and cost if a provider requires a longer minimum, then trim a derived copy to the selected excerpt.

Use this appearance-preserving baseline, adapted to the selected provider:

> Use the supplied photograph as the visual reference and starting frame. Preserve each person's recognizable identity, apparent age, facial proportions, skin tone and texture, hairstyle, hairline, facial hair, glasses, clothing and accessories. Preserve the original photographic tone, grain, lighting, background and composition. Use one small, natural movement in a continuous shot with a steady camera. Keep visible anatomy and existing contacts stable. Keep cropped body parts outside the frame. Preserve silent, relaxed lips unless a separately approved speech feature is selected. Treat this as an imagined animation, not an authentic recording.

Append the selected motion and a clearly separated priority note. A priority note emphasizes an instruction; it is not a guarantee or a supported numeric model weight unless the selected API actually offers that control. Flag conflicting notes for review rather than silently overriding appearance preservation.

Example couple preset:

> Starting from their camera-facing pose, the two people turn their eyes slightly toward each other and share a small closed-mouth smile. Keep head movement minimal and their existing hand positions steady. Settle naturally after the glance.

Label full-body dancing, hand gestures, embraces and changes of pose as higher risk. Prefer source photos already close to the requested pose, with the moving body parts visible. Never promise perfect fingers, unchanged faces or distortion-free output. The presets must be described as prepared suggestions, not proven optimal results until tested.

### 3A. Choose the most effective AI for each task

Use task-specific model routing behind the simple customer interface. Do not assume the best animation model is also the best still-photo retoucher.

Initial production candidates, to verify against current official documentation and actual account access before implementation:
- Photo to Video: Runway Gen-4.5 (`gen4.5`) as the starting default for short image-to-video requests. Compare it with current Kling image-to-video models on the same rights-cleared test set, especially for couples, hand contact and dancing.
- Digital Makeover: OpenAI GPT Image 2.5 Sunburst (`gpt-image-2.5-sunburst`) as the starting candidate for precise image editing. Verify current edit/mask support and fidelity behavior; use only parameters actually supported by the chosen model.
- Choose final production defaults using acceptable-result rate, recognizable likeness, preservation of unselected details, anatomy, latency and total cost per accepted result. Do not call a model a proven winner without this evaluation.

Use at least 20 representative, disposable, rights-cleared photographs across single portraits, older photographs, couples, groups, glasses, facial hair and different lighting/skin tones. Test each task separately. Reject models/settings that consistently alter identities or unselected people. Keep a provider adapter and versioned model configuration so a better model can be adopted without changing the wallet or photo interface.

Prefer quality and precise instruction-following over the cheapest attempt. Do not silently substitute a lower-quality model, route a private photo to an undisclosed provider, or launch paid automatic retries. A model change that alters the price or data recipient requires an updated disclosure/quote and appropriate confirmation. Record the exact model and settings in the job receipt.

Keep these customer-facing app credits independent of any provider's monthly credits. Customer rollover must continue to work if the app changes AI providers.

References checked November 1, 2026:
- https://docs.dev.runwayml.com/api-details/api_changelog/
- https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst
- https://developers.openai.com/api/docs/guides/tools-image-generation

These are implementation starting points, not a claim that either provider guarantees likeness or anatomy. Recheck model access, pricing and terms at implementation time.

### 4. Digital Makeover checkboxes

For photos with multiple people, require the user to select the person or draw/refine a target region before retouching. Provide a way to correct an automatic mask. Do not silently alter every person in the image.

Offer these neutral checkbox labels and professional background instructions:

| Checkbox | Prompt instruction applied only when selected |
| --- | --- |
| Reduce temporary blemishes | Reduce visible temporary pimples and localized redness in the selected region. Preserve pores, freckles, birthmarks, permanent scars, surrounding skin tone and natural texture. |
| Soften fine lines | Gently reduce the contrast of fine lines in the selected region. Retain natural folds, expression and recognizable age; avoid waxy or airbrushed skin. |
| Soften deeper wrinkles | Conservatively soften selected deeper lines without flattening facial structure or erasing all signs of age. Keep the person recognizable. |
| Remove visible braces | Remove visible brackets and wires and create a plausible, natural-looking dental appearance. Keep the existing smile, mouth opening, lips and facial proportions. Do not whiten or reshape the smile beyond the requested removal. Warn that hidden tooth detail must be inferred. |
| Remove eyeglasses | Remove the selected glasses and reconstruct plausible eye/skin detail beneath them. Preserve the person's recognizable eye shape and surrounding features as closely as the source permits. Warn that obscured detail is inferred and may be inaccurate. |
| Reduce under-eye darkness | Gently reduce dark circles and discoloration while preserving skin texture, eye shape, lighting and surrounding skin tone. |
| Soften under-eye puffiness | Subtly reduce the appearance of under-eye bags while retaining the natural lower-eyelid shape and recognizable face. |
| Reduce under-chin fullness | Make a small, localized adjustment beneath the chin. Preserve the recognizable chin, neck alignment, jaw proportions and natural shadows; do not change the rest of the face. |
| Gentle facial contouring | Make conservative adjustments only in the user-selected facial region. Preserve recognizable identity, expression and overall age. Do not change unrelated eye, nose or lip geometry. This is cosmetic photo editing, not a medical procedure. |
| Gentle waist/abdomen adjustment | Make a restrained adjustment to the selected adult subject's waist/abdomen outline or garment drape. Keep plausible anatomy, posture, limb proportions, clothing seams, patterns and nearby background lines. Do not create an extreme body transformation. |
| Smooth clothing around the waist | Reduce selected garment bunching or folds while maintaining believable fabric, pattern alignment and body proportions. Do not reshape the person's body when only clothing smoothing is selected. |
| Even localized skin tone | Gently balance selected uneven patches under the original lighting. Preserve the person's natural complexion and ethnic appearance; do not globally lighten or darken skin. |

Do not silently add teeth whitening, makeup, age changes, skin lightening, body reshaping, facial-hair removal or changes to unselected people. If a requested edit cannot be isolated reliably, show the limitation before spending credits.

Default to **Subtle** intensity. Offer a clearly explained Natural/Moderate alternative only where the provider supports useful control. Do not present invented model strength settings as real controls. Restrict body-contouring and cosmetic face-contouring workflows to adults; do not infer permission or age solely from a face detector. Allow ordinary blemish correction without offering sexualized or extreme transformations.

Avoid marketing labels such as “digital surgery,” “facelift result” or “tummy tuck result.” If those phrases appear in help text, explain that this is only a photo edit and does not predict medical outcomes.

### 5. Makeover master prompt and composition

Use a versioned master prompt with deterministic checkbox-to-instruction mapping. Example baseline:

> Edit the supplied photograph conservatively. Apply only the explicitly selected retouching tasks inside the approved target region. Preserve the person's recognizable identity, original expression, pose, hairstyle, facial hair, clothing, photographic style and all unselected details. Retain realistic skin texture and the original lighting. Keep other people unchanged. Preserve plausible anatomy, clothing geometry and background lines. Blend localized edits naturally without beautifying unrelated areas.

Append, in order:
1. Target person/region and selected intensity.
2. Instructions for the checked options only.
3. The user's optional priority note, limited to the selected tasks.
4. Provider-specific settings supported by its current API.

If a note requests an unchecked change, ask the user to select/confirm that option. Do not let a hidden master instruction silently enable a makeover. Advanced master-prompt editing must show the resulting complete request and flag conflicts with selected options or target boundaries.

Save the exact checkbox selections, mask/reference, prompt version, user note and model settings with each generation. Keep this metadata private and separate from the untouched original.

### 6. Before/after review and feedback

After a successful paid job, show the original and generated derivative side by side or with an accessible before/after slider. Include keyboard-accessible Original/Enhanced buttons. Clearly label AI-edited or AI-animated content.

Actions: Keep this version, Discard this version, Download, Save as a new shoebox item, Use in my story/interview, and Try another version. A new generation must display a new price and require confirmation; never retry automatically because someone dislikes an image.

Never overwrite or delete an original. Preserve aspect ratio and orientation unless the user explicitly chooses otherwise. Do not overwrite existing story edits. Restoring the original must be straightforward.

Collect optional feedback: overall rating; likeness changed; excessive smoothing; eyes/teeth/glasses issue; hands/limbs issue; body/clothing distortion; background warped; requested edit missing; someone else changed; emotional discomfort; other. Allow a note. Separate feedback from a billing dispute. Ask explicit permission before sending a photo or generated result to support. Local-only feedback must be labeled as local, not as a submitted support request.

### 6A. Human photo professional — fixed service or custom quote

Offer a clearly labeled **“Have a photo professional help”** option:
- Beside the two main enhancement cards, as a secondary alternative.
- On AI result/review screens, especially after **“I’m not happy with this result.”**
- From the original photo, so trying AI first is never required.

Use neutral copy: “Prefer a personal touch? Ask a photo professional to retouch this picture.” Do not imply human work is free or that an unhappy AI result automatically creates a complimentary human order. Keep AI billing disputes separate from a new human service purchase.

Support two admin-configured purchase paths:

**A. Book a fixed-price photo service**
Show a service menu only for procedures with an approved price, defined scope and an available fulfillment professional. Examples may include blemish cleanup, under-eye retouching, wrinkle softening, glasses/braces removal, clothing cleanup or photo restoration. Do not invent prices. Show the actual price, applicable tax, number of photos/people covered, deliverable resolution, turnaround, included revisions and exclusions before checkout. Complex or uncertain work should route to a quote rather than accepting a misleading fixed-price order.

**B. Request a custom quote**
Carry forward the selected photo securely; also allow a new local upload. Let the customer choose what is shared: untouched original, selected AI attempt if relevant, and optional reference photos. Do not send every generation or other shoebox items by default.

Collect:
- A plain-language description of the requested work, with optional existing makeover checkboxes.
- Target person/region and anything that must remain unchanged.
- Intended use and desired delivery size.
- Optional deadline, reference image and before/after preference.
- Contact details needed for the request, prefilled only when appropriate.
- Confirmation of permission to share these specific files with the assigned photo professional.

Show supported file types and size limits. Upload progress, interruption recovery and duplicate-submit protection must work. Keep files private and attached to this request only.

Default the quote request to no charge unless the business explicitly configures a quote-review fee. If a fee is configured, display its exact amount, tax, what it covers, whether it is credited toward accepted work, and the approved cancellation/refund treatment BEFORE submission/payment. Never turn a quote request into an undisclosed charge.

A submitted quote request is not an accepted project or authorization to charge the enhancement price. The professional must send a concrete proposal containing the work description, exclusions/limitations, price and currency, tax, delivery estimate, revision allowance, quote expiration and approved cancellation/refund terms. The customer can accept and pay, ask a question, or decline. A material change in scope or price requires a new accepted proposal. No work beyond the paid scope is charged automatically.

**Order fulfillment and customer review**
Create a real, authenticated staff workflow for reviewing requests, preparing quotes, assigning a professional, exchanging scoped notes and delivering proof/final files. Do not present a working human-service button unless a monitored queue, fulfillment staff and necessary storage/payment services actually exist.

Track request and order states separately: draft, upload pending, submitted, under review, quote sent, accepted, awaiting payment, assigned, in progress, proof ready, revision requested, delivered, declined and cancelled. Keep payment/refund status explicit rather than hiding it inside a general “complete” state. Show the customer their request/order number, current status, receipt, next step and promised response/delivery times only when those times are configured and staffed.

Capture only the approved fixed-service or quoted amount through the configured secure checkout. Use verified, idempotent payment webhooks. Do not start paid work based only on a browser success redirect. If human services are priced in money, show them in money; do not silently deduct AI credits. Supporting human services with app credits would require a separately approved conversion/pricing policy shown before checkout.

Allow the customer to compare the original and proof, request revisions within the agreed scope, approve delivery and download the finished file. Deliver a new derivative and offer “Save to my shoebox” or “Use in my story/interview.” Never replace the original or existing story edits. Additional work outside the agreed scope needs a new quote.

If the complaint is about AI video, explain whether the available human service covers only the source photograph or also video correction. Do not promise professional video restoration or perfect animation unless that service is actually staffed and priced.

**Privacy and feedback**
Grant each staff member/professional access only to assigned requests and their specifically shared files, not the entire customer account or shoebox. Use authenticated access or short-lived scoped links, audit access and downloads, and revoke access when appropriate. State the actual file retention/deletion policy, professional/provider access and any external transfer before submission. Never display family photos in a public portfolio without separate consent.

Provide in-app request status and optional notifications through configured services, respecting customer communication preferences. Do not claim an email or support message was sent unless delivery was actually attempted and its state is recorded.

Offer optional professional-service feedback after delivery, distinct from AI-output ratings. Include an accessible way to report a problem with the order, ask about a charge or request help under the approved service terms. Do not advertise satisfaction guarantees, unlimited revisions or turnaround promises that have not been approved and staffed.

### 7. Shared credits and month-to-month rollover

Create one authenticated, server-side **Moments R Forever AI Credit Wallet** usable across Photo to Video, Digital Makeover, paid booth AI and future AI features. Do not tie the customer's wallet balance to an AI provider's monthly website subscription balance. Do not store money or enforce spending caps in localStorage.

Default product policy to implement once approved purchase terms are in place:
- Separately purchased paid credit packs do not expire or reset each month.
- Paid subscription credit allocations roll over rather than disappearing at month-end. Default to retaining already earned paid credits after cancellation; cancellation stops future allocations. Any different subscription policy requires explicit approval and clear disclosure before sale.
- Separately purchased credits remain available after subscription cancellation.
- Promotional/free credits may have a disclosed expiration; distinguish them from paid credits and use expiring credits first where appropriate.
- Moving between phones, browsers or computers does not change the account balance.
- Show available credits, reserved credits and transaction history. Use simple wording such as “Unused purchased credits stay in your account.”

Credit amounts must be integer units with server-side pricing. Quote the full billable provider request, including minimum duration, resolution and selected options. Clearly distinguish a customer's app credits from the provider's internal credits or token units. Do not invent prices or assume every model is billed in tokens.

Add per-generation and optional monthly spending caps. Count pending reservations toward those caps. Treat caps as limits, never as permission to spend automatically. Keep automatic top-ups off unless the customer separately enables them with a clear limit.

### 8. Payment and generation integrity

Credit a wallet only after a verified successful payment-provider webhook. Validate account, product, amount, currency and payment status. Duplicate webhooks must not credit twice. A successful browser redirect alone is not payment proof.

Authenticate photo ownership/processing permission and destination access before quoting or starting a job. Use private media storage, scoped temporary provider access and server-held API credentials. Explain the provider transfer and obtain consent before the first upload. Do not repurpose family photos for unrelated uses.

Use an append-only ledger with purchase, grant, reservation, settlement, release, refund and adjustment records. Use atomic balance checks so concurrent requests cannot overspend. Apply unique transaction/job/payment identifiers and idempotency keys. Record the prompt/photo hash, quote expiry, model, provider task ID, actual cost and receipt.

Job states must distinguish quoted, reserved, submitted, processing, ready, failed, cancelled and uncertain/time-out outcomes. Reconcile uncertain provider jobs before resubmitting or releasing funds. Never double-dispatch or double-charge after a retry or duplicate callback.

Release unused reservations after a confirmed unbilled technical failure. Reconcile provider-billed failures under the approved billing policy. Treat a delivered but disliked generation, an undelivered job, a duplicate charge and a payment dispute as separate cases. Show the proposed resolution and account record accurately.

Before enabling purchases, finalize clear purchase/credit/refund terms. Explain that AI processing does not guarantee a desired appearance and that a completed generation may consume credits even if disliked. Do not implement a blanket disclaimer that claims to remove every responsibility or denies all reimbursements regardless of technical failure or duplicate billing.

### 9. Examples that explain the features honestly

Photo to Video: show a rights-cleared couple photo followed by an actual generated clip where they begin facing the camera, make a slight mutual glance and smile. Label the AI animation and identify the illustrative nature of the example. Preserve a static comparison for reduced motion; do not autoplay audio.

Digital Makeover: show a rights-cleared consenting adult example with subtle blemish/under-eye retouching. List exactly which checkboxes were selected. Avoid humiliating before labels or dramatic weight-loss promises.

If no provider is connected, use a clearly labeled storyboard or explain that the actual sample is unavailable. Never present a still-image pan, CSS animation or fabricated preview as a successful AI transformation.

### 9A. Graphical prototype and example assets

Build a highly polished, accessible warm-ivory/maroon/sage interface with real photographic demo assets rather than placeholder boxes. Start with two clear cards: Photo to Video and Digital Makeover, with a secondary professional-help route. Include a rollover-wallet explanation, workflow navigation and an explicit example/demo label.

Provide matched close-up example pairs for blemish/under-eye finishing, glasses removal and braces removal. Where available, add a fully clothed adult waist/clothing example. Use a before/after slider plus Original and Retouched buttons, and a larger close-up dialog. Each asset pair must state what was changed and what the model inferred. Show examples as examples, not live results of checkbox selection. Do not imply an unselected procedure has been applied to a customer's photograph.

For animation, provide an actual generated-video example only when one is available. Otherwise show clearly labeled first/last storyboard stills with separate frame buttons; never call a CSS crossfade a generated video. Include the original camera-facing couple and a matching gentle glance frame.

Give feedback a designed screen with overall rating, issue tags, written notes and a clear human-professional alternative. A standalone prototype may save demo feedback only in the current browser; it must not claim the note was sent to support. Its quote form may preview an on-screen request but must not claim a staffed order was submitted.

Use fictional, generated adult subjects for the demo and preserve their consistency within each pair. Mark those assets as fictional illustrations, not evidence of a production model's quality or a promise to customers. Keep before labels neutral. Do not use real customer memories in demonstrations without separate consent.

A demonstration may use a small explicitly labeled subset of animation presets, but the real product must reuse all 100 existing presets. Any displayed credit balance must be labeled illustrative when no account ledger exists. Do not invent prices, transactions, reviews, staff identities or turnaround promises. Include working downloads of the build brief and example boards, keyboard focus management, 44px targets, reduced-motion handling and mobile layouts.

### 10. Validation and delivery

Use disposable test photos and test payment events only. Verify:
- The Enhance action is available and accessible across the inventoried photo surfaces and carries the correct photo/destination.
- Canceling preserves work; opening an action causes no upload, payment or generation.
- Both cards route correctly; video preservation and intentional makeover settings stay separate.
- The 100 presets and dependent selectors remain correct; custom prompts, priority notes and master edits compose predictably.
- Every checkbox maps to only its requested instruction; empty selection cannot trigger a paid makeover; other people remain outside the edit target.
- Before/after, download, save-as-new and story/interview insertion preserve original file hashes and existing edits.
- Paid-credit rollover, subscription cancellation, promotional expiry and transaction history match approved policy. Changing providers does not reset, expire or lose customer balances.
- Human fixed-price services and quote requests carry the right photo and instructions; quote-review fees are disclosed; accepting a quote is separate from submitting a request; scope/price changes require approval.
- Professional access is limited to assigned files; interrupted uploads, duplicate submissions/payments, quote expiry, cancelled orders, proof/revision delivery and cross-account/staff access are tested.
- Human-service receipts, revision allowances, retention rules and status notifications match the actual configured service. An unhappy AI result never silently starts a paid human order.
- Forged/duplicate payment webhooks, duplicate generation retries, concurrent spending caps, insufficient balance, quote changes/expiry, failed jobs, late callbacks and cross-account access are handled safely.
- Local drafts and feedback never claim account persistence or submission when they are browser-only.
- Desktop/mobile layouts, keyboard focus, 44px targets, screen-reader names and reduced-motion behavior work.

Run appropriate checks and report actual results. Publish the verified update to Site B while preserving its audience. Finish with what existed, what changed, checks passed/failed/not performed, precise infrastructure blockers, deployed URL and saved version/commit. Do not stop at a plan or claim an unavailable service works.
