25 ChatGPT-5.5 Prompts for Production Image Workflows: Creative Briefs, Owned Photos, Precision Edits, QA, and Provenance

How to Use ChatGPT-5.5 as the Production Layer for Image Work

This prompt set treats ChatGPT-5.5 as the planning, governance, evaluation, and documentation layer for production image workflows, not as the image-rendering model. In the workflow used throughout this article, ChatGPT-5.5 helps a team convert a vague request into a usable creative brief, verify photo rights and consent assumptions, decide whether a job is generation or editing, write preservation instructions, create visual QA rubrics, draft alt text, and produce archive manifests. The actual image generation or image editing step is routed to OpenAI’s GPT-Image-2.5 models: Flare for faster everyday generation and Sunburst for precision-sensitive editing and premium control, as OpenAI describes them in its Images 2.5 and API documentation.

OpenAI’s image-generation guide documents two production routes that matter for these prompts. The Image API is the direct route for single-prompt generation or editing where the application sets the image model directly to gpt-image-2.5-flare or gpt-image-2.5-sunburst. The Responses API route is more suitable when a workflow needs conversation state, multi-step reasoning, or a top-level model coordinating an image-generation tool; in that route, the mainline model and the image tool’s model are separate choices. The prompts below are written so a producer, developer, or brand reviewer can specify the intended route before any rendering work begins.

The selected guide covers production-grade prompt engineering for frontier AI APIs, including layered prompt architecture, caching, and agentic task optimization. The complete The 2026 ChatGPT Prompt Engineering Best Practices Guide article provides the destination-specific detail for this section’s ChatGPT-5.5 Prompt Engineering decision because it is the strongest general prompt-engineering reference for readers who want to turn ChatGPT image prompts into repeatable, production-grade workflows.

The official model catalog for this article includes GPT-5.5, but no official gpt-5.5-mini identifier is documented in the supplied source set. Do not build examples, environment variables, routing rules, or procurement language around a miniature GPT-5.5 variant unless OpenAI documents that identifier in the official catalog you are using at implementation time. For this prompt pack, “ChatGPT-5.5” means the model used to structure decisions and review outputs; “Sunburst” and “Flare” mean the image models used to render or edit images.

The Production Rule: Plan Before You Render

Production image work fails most often when teams ask for a visual result before they define the operational constraints. A marketing manager may request “make this founder photo more premium,” but that phrase does not specify whether the organization owns distribution rights, whether the founder consented to modification, whether facial identity must remain unchanged, whether the office background is a legally sensitive location, whether visible product labels can be altered, or whether the final asset will appear in a paid ad, help-center article, investor deck, or recruitment campaign. The purpose of these prompts is to force those constraints into the request before a model is asked to generate pixels.

OpenAI describes Images 2.5 as improving areas such as detail, natural lighting, textures, reference-subject preservation, targeted editing, multi-turn edit consistency, instruction following, real-world information, style control, and transparent-background generation. Those are useful capabilities, but they do not remove the need for human constraints. A model can follow a precise edit request better when the request identifies the subject, the protected regions, the allowed transformation, the forbidden changes, the target channel, the acceptance checks, and the person who has authority to approve the asset.

For owned-photo workflows, the default assumption should be conservative: the user must confirm they own the photo or are authorized to distribute and modify it, and the prompt must not fabricate consent, licenses, releases, or ownership. If the image contains a real person, the prompt should preserve identity and non-target regions unless the person has authorized a specific transformation. If the image contains brand marks, licensed artwork, packaging, private spaces, sensitive documents, uniforms, or minors, the workflow should escalate to a human approval owner before editing rather than burying risk inside a style prompt.

Operational warning: provenance metadata and watermarking are not substitutes for rights review. OpenAI says Images 2.5 uses C2PA metadata and an invisible SynthID watermarking layer across ChatGPT, Codex, and the OpenAI API, and OpenAI explicitly states that no single provenance mechanism is sufficient. Treat C2PA and SynthID as provenance signals, not proof of copyright ownership, subject consent, factual authenticity, or permission to publish.

The Standard Prompt Contract for Every Image Request

Every prompt in this masterclass is built around a standard contract. The contract is intentionally more formal than a casual image prompt because production images travel through design review, legal review, web optimization, accessibility checks, analytics, and archival systems. When a team captures the contract at the start, it can compare generated candidates against the brief instead of relying on taste, memory, or whoever is loudest in the approval thread.

This Images 2.0 prompting guide presents 25 structured patterns for production-quality visual generation, including specification design, negative constraints, and reference-image conditioning that can be incorporated into the broader rights-aware workflow in this masterclass. The complete ChatGPT Images 2.0 Advanced Prompting: 25 Patterns That Get Production-Quality Outputs article provides the destination-specific detail for this section’s AI Image Generation Workflow decision because the target is a production-oriented image-prompt workflow and is more exact than a Codex UI-mockup playbook for this marker.

How the 25 Prompts Are Organized

The 25 prompts that follow are designed as reusable templates, not one-off examples. Each prompt section includes a purpose, a copy-paste prompt, required inputs, expected output, and an approval or safety note. The structure lets a creative-production team place these prompts inside a briefing form, project-management ticket, internal content operating procedure, or developer tool without rewriting the logic every time a new campaign or product image is requested.

The first group focuses on preparation: converting requests into creative briefs, identifying channel and aspect requirements, inventorying source assets, and confirming rights for user-owned photos. The second group controls image decisions: choosing generate versus edit, routing between Flare and Sunburst, defining subject-preservation instructions, and managing multi-turn edit constraints. The third group handles quality and governance: brand consistency, text-risk planning, contact-sheet review, alt text, provenance interpretation, filenames, optimization, approvals, and archive manifests.

Use the prompts in sequence when the asset has legal, brand, or identity sensitivity. For a low-risk internal concept image, a team may only need the brief, routing, and QA prompts. For a public campaign using a real person’s photo, the team should use the rights-confirmation, consent, preservation, edit-scope, approval, provenance, and archive-manifest prompts before publication. The operating rule is simple: the more real-world identity, ownership, or brand value an image carries, the more of the contract must be explicit and approved.

These prompts also assume that generated candidates can fail review. OpenAI’s documentation and safety materials describe safeguards, evaluations, C2PA metadata, and SynthID watermarking, but they do not promise flawless identity preservation, perfect text rendering, universal latency reductions, zero unsafe outcomes, or permanent provenance persistence through every downstream transformation. A production workflow should therefore retain source references, prompts, model-route decisions, reviewer notes, acceptance results, and final-file manifests so the organization can explain how an image was produced and why it was approved.

Prompts 1–9: Brief, Rights, Routing, Preservation, and Channel Specs Before Rendering

Use these first nine prompts with ChatGPT-5.5 as a planning, governance, and QA assistant before sending anything to an image-rendering workflow. OpenAI documents GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst as image-generation and editing models; ChatGPT-5.5 should be used here to structure briefs, define constraints, route work, and prepare acceptance criteria rather than being treated as the image model itself.

The selected article explains how to turn simple prompts into reusable AI skill workflows through prompt chaining, template design, cross-platform compatibility, and automation. The complete From Prompts to AI Skills: How to Build Reusable Prompt Workflows for ChatGPT, Claude, and Codex article provides the destination-specific detail for this section’s Image Prompt Design decision because image prompt design benefits from reusable templates and prompt workflow structure, which this article specifically covers.

Prompt 1: Campaign Brief Normalizer

Purpose: Convert a messy campaign request into a production-ready image brief that separates business intent, audience, creative direction, constraints, and approval criteria.

Copy-paste prompt:

You are my production image brief editor. Convert the notes below into a structured creative brief for an AI-assisted image workflow.

Return:
1. Campaign objective
2. Audience and context
3. Core visual idea
4. Required subjects, products, or scenes
5. Brand tone and style boundaries
6. Must-avoid items
7. Required deliverables
8. Open questions that must be answered before generation or editing
9. Human approval checklist

Important: do not invent missing legal rights, consent, product claims, or brand permissions. If the request involves a real person, owned photograph, trademarked asset, or client product, flag the approval dependency.

Raw campaign notes:
[PASTE NOTES]

Required inputs: Campaign notes, audience, brand or client context, required deliverables, and any legal or compliance constraints already known.

Expected output: A structured brief that a designer, marketer, developer, or image API operator can review before generating or editing assets.

Approval/safety note: Treat unanswered rights, consent, medical, financial, political, or product-claim issues as blockers, not as creative gaps for the model to fill.

Prompt 2: Content-to-Image Mapping

Purpose: Decide which parts of an article, landing page, email, or campaign actually need images, and define the job of each image before creating prompts.

Copy-paste prompt:

You are mapping content into a production image plan. Analyze the content below and recommend only the images that have a clear communication job.

For each recommended image, return:
- Placement or section
- Image purpose
- Key message the image must carry
- Suggested image type: hero, diagram, product visual, social crop, editorial illustration, thumbnail, comparison visual, or process graphic
- Whether this should be generated from text, edited from an owned image, or created manually
- Risks: rights, privacy, factual accuracy, brand mismatch, text-in-image, or misleading realism
- Acceptance criteria

Content:
[PASTE CONTENT]

Brand or campaign context:
[PASTE CONTEXT]

Required inputs: Source content, brand context, channel goals, and any required placements.

Expected output: A practical image map that avoids decorative overproduction and links every asset to a communication requirement.

Approval/safety note: Reject visuals that would imply facts not present in the source content, especially performance claims, endorsements, before-and-after results, or real-world events.

Prompt 3: Asset Inventory Builder

Purpose: Create an operational inventory of available images, logos, product shots, reference photos, sketches, style boards, and previous campaign assets before deciding what to generate or edit.

Copy-paste prompt:

You are building an image-production asset inventory. Organize the asset list below into a decision table.

For each asset, return:
- Asset name or identifier
- Asset type
- Ownership or license status as provided
- Whether it may be used as a reference, edited source, published output, or internal-only inspiration
- People, private locations, trademarks, products, or sensitive data visible
- Required consent or approval
- Recommended use in the workflow
- Do-not-use conditions

Do not assume rights that are not stated. Mark unknown rights as "needs confirmation."

Asset notes:
[PASTE ASSET LIST]

Required inputs: Asset filenames or descriptions, license notes, owner or client source, intended channels, and visible subject details.

Expected output: A reviewable table that separates usable production inputs from internal references, restricted assets, and blocked assets.

Approval/safety note: If an asset contains a recognizable person, private setting, confidential material, or third-party mark, require explicit confirmation before using it as an edit input or reference.

Prompt 4: Owned-Photo Rights and Consent Check

Purpose: Verify whether a real photo can be used in an AI-assisted editing workflow and document the permitted transformation before any edit instructions are written.

Copy-paste prompt:

You are my owned-photo rights and consent reviewer for an AI image edit.

Ask me for missing information, then produce a go/no-go decision.

Photo description:
[DESCRIBE PHOTO]

Intended edit:
[DESCRIBE EDIT]

Intended publication channels:
[LIST CHANNELS]

Known rights and consent:
[PASTE WHAT IS KNOWN]

Evaluate:
1. Do I own the photo or have authorization to distribute it?
2. Are recognizable people present?
3. Is consent documented for the intended edit and publication?
4. Are private locations, minors, uniforms, medical context, financial context, or sensitive attributes present?
5. Does the edit preserve identity and non-target regions?
6. Could the edit misrepresent the person, event, product, or setting?
7. What exact transformation is permitted?
8. What must not be changed?

Required inputs: Photo description, ownership status, consent terms, intended edit, publication channels, and subject sensitivity.

Expected output: A go, revise, or stop decision with permitted edits, prohibited edits, and unresolved approvals.

Approval/safety note: Do not proceed if ownership, authorization, or consent is unknown. The prompt must not fabricate permission or reinterpret a limited release as broad consent.

This intellectual-property analysis explores how disputes over AI talent and trade secrets can reshape ownership and governance expectations, reinforcing why production image teams must document source rights, consent, and permitted transformations rather than assume them. The complete How the Apple vs OpenAI Lawsuit Could Reshape AI Talent Wars and Intellectual Property in 2026 article provides the destination-specific detail for this section’s Copyright Safe AI Images decision because the target directly examines AI intellectual-property risk, and the bridge correctly uses it to support rights documentation without presenting it as an image-licensing ruling.

Prompt 5: Privacy and Metadata Choice

Purpose: Decide what metadata, provenance records, filenames, captions, and visible or invisible signals should be preserved, stripped, or documented for production use.

Copy-paste prompt:

You are preparing privacy and metadata instructions for an AI-assisted image workflow.

Given the asset and publishing context below, recommend:
- Metadata to preserve for internal archive
- Metadata to remove before public distribution
- Provenance notes to retain in the production record
- Caption or credit requirements, if provided
- Privacy risks from location, device, timestamp, identity, or embedded text
- Whether the public file should avoid sensitive descriptive filenames
- What the approval record should say about C2PA, invisible watermarking, or other provenance signals

Important: provenance signals can support traceability but do not prove copyright ownership, truth, or consent.

Asset context:
[PASTE CONTEXT]

Required inputs: Asset source, publication channel, privacy sensitivity, credit obligations, archive policy, and whether the asset is generated, edited, or photographed.

Expected output: A metadata handling plan for internal records and public distribution copies.

Approval/safety note: OpenAI says its provenance approach includes C2PA metadata and invisible watermarking, but no single provenance mechanism is sufficient; keep independent production records.

Prompt 6: Generate-Versus-Edit Decision

Purpose: Decide whether the next asset should be created from a text prompt, edited from an owned image, built as a composite by a human designer, or deferred.

Copy-paste prompt:

You are my generate-versus-edit decision assistant. Use the brief and asset inventory below to recommend the safest and most efficient production path.

Return a decision for each needed image:
- Generate from text
- Edit an owned and authorized image
- Use a human designer or photographer
- Do not create until approvals are resolved

For each decision, explain:
1. Why this path fits the objective
2. Required inputs
3. Rights or consent dependencies
4. Subject-preservation requirements
5. Accuracy risks
6. Whether text-in-image is essential or should be handled in layout software
7. Recommended acceptance criteria

Brief:
[PASTE BRIEF]

Asset inventory:
[PASTE INVENTORY]

Required inputs: Approved brief, asset inventory, rights status, channel requirements, and any real-person or product constraints.

Expected output: A decision table that prevents unnecessary edits to sensitive photos and avoids generating images where factual or legal precision requires another method.

Approval/safety note: If the workflow involves a real person, preserve identity and non-target regions unless the documented consent specifically allows the change.

Prompt 7: Flare-Versus-Sunburst Routing

Purpose: Route each image task to GPT-Image-2.5 Flare or GPT-Image-2.5 Sunburst based on production need, not unsupported assumptions about price or guaranteed quality.

Copy-paste prompt:

You are routing image tasks between GPT-Image-2.5 Flare and GPT-Image-2.5 Sunburst.

Use OpenAI's documented positioning:
- Flare: fast, high-quality everyday image generation and rapid creative workflows
- Sunburst: premium workflows where editing precision and tighter control matter more, with longer generation times

For each task below, recommend:
- Flare, Sunburst, or non-AI/manual production
- Reason for routing
- Whether the task is generation, edit, or inpainting
- Required quality setting to test: low, medium, high, xhigh, max, or auto
- Human acceptance criteria
- Measurement notes: actual token usage, output review, revision count, and channel fit

Tasks:
[PASTE TASK LIST]

Required inputs: Task list, deadline, volume, editing precision requirements, reference images, quality expectations, and review capacity.

Expected output: A routing plan that uses Flare for speed-oriented everyday work and Sunburst for precision-sensitive edits or premium production tests.

Approval/safety note: OpenAI documents matching token rates for Flare and Sunburst, so do not route on a fabricated per-token discount; measure actual usage and acceptance outcomes.

Prompt 8: Reference-Preservation Contract

Purpose: Write a strict preservation contract for edits that use an owned reference image, especially where identity, product shape, packaging, clothing, layout, or non-target regions must remain unchanged.

Copy-paste prompt:

You are drafting a reference-preservation contract for an AI image edit.

Reference image description:
[DESCRIBE IMAGE]

Confirmed rights and consent:
[STATE CONFIRMATION]

Permitted transformation:
[DESCRIBE ONLY THE ALLOWED CHANGE]

Create an edit contract with:
1. Elements that must remain unchanged
2. Elements that may change
3. Elements that must not be inferred or invented
4. Identity-preservation instructions, if a person is visible
5. Product-preservation instructions, if a product is visible
6. Background or non-target-region preservation rules
7. A concise edit prompt
8. A QA checklist to compare before and after

Do not authorize any edit outside the stated permission.

Required inputs: Reference description, explicit rights confirmation, consent details, allowed edit, prohibited changes, and review criteria.

Expected output: A preservation-first edit prompt plus a checklist for comparing the generated edit against the source image.

Approval/safety note: OpenAI describes improved reference-subject preservation and targeted editing for Images 2.5, but that is not a guarantee; human review must verify the subject and non-target regions.

Prompt 9: Channel and Aspect-Ratio Specification

Purpose: Convert campaign needs into exact channel deliverables, crops, safe zones, file notes, and review criteria before image generation or editing begins.

Copy-paste prompt:

You are preparing channel specifications for production image assets.

Using the campaign brief below, create a deliverables table with:
- Channel or placement
- Asset purpose
- Required aspect ratio or crop family
- Primary subject placement
- Safe-zone notes for text overlays, UI chrome, or platform cropping
- Whether transparent background is needed
- Whether the image should be generated, edited, or manually laid out
- Export and naming notes
- Approval checklist for each channel

Do not invent platform requirements. If dimensions, file-size limits, or naming rules are not provided, mark them as "client/platform spec needed."

Campaign brief:
[PASTE BRIEF]

Known channel specs:
[PASTE SPECS]

Required inputs: Campaign brief, known platform or CMS specs, target channels, crop needs, overlay requirements, and export conventions.

Expected output: A channel-ready production table that prevents one beautiful image from being forced into unsuitable crops.

Approval/safety note: Keep layout-critical typography, legal disclaimers, and regulated claims outside the generated pixels when precision is required; add them later in controlled design software.

Prompts 10–18: Editorial Generation, Authorized Photo Edits, Brand Controls, and Variation Decisions

Use these nine prompts when the work moves from planning into production decisions: generating a new editorial concept, editing authorized portraits and product photos, controlling backgrounds and composition, managing multi-turn changes, keeping brand behavior consistent, reducing text-rendering risk, and selecting variations with documented reasons. In each case, ChatGPT-5.5 is the planning, governance, and review assistant; the actual rendering or editing step belongs to the appropriate image workflow, such as the Image API for single generation or edits, or the Responses API image-generation tool for conversational and multi-step work.

Prompt 10: New Editorial Image Generation Brief

Purpose: Use this prompt when no source photo is required and the team needs an original editorial image brief for a news post, analysis page, explainer, report cover, or social card. The prompt separates the article’s factual topic from the visual metaphor so the image does not fabricate events, people, product screenshots, documents, charts, or logos that the team does not have rights to use.

Copy-paste prompt:

You are my production image brief editor. Create an original editorial image brief for a new image, not an edit of an existing photo.

Article or asset topic: [TOPIC]
Audience: [AUDIENCE]
Channel and placement: [CHANNEL]
Required dimensions or aspect ratio: [SIZE_OR_RATIO]
Brand tone: [BRAND_TONE]
Things that must not appear: [EXCLUSIONS]
Any real people, logos, documents, products, or locations that must be avoided unless licensed: [RIGHTS_SENSITIVE_ITEMS]

Produce:
1. A one-sentence editorial concept.
2. A detailed image prompt for generation.
3. A negative prompt or exclusion list.
4. A factual-risk note identifying any elements that could imply a real event, endorsement, legal claim, or third-party rights.
5. A recommended routing note: use Flare for fast everyday generation, or consider Sunburst only if precision-sensitive editing or tighter control is required later.
6. Three safe alternate concepts.

Required inputs: Provide the actual topic, publication context, visual tone, placement, dimensions, and a list of prohibited rights-sensitive elements. Do not ask the model to invent credentials, awards, news photos, screenshots, or product interfaces.

Expected output: Expect a production-ready generation brief with exclusions and alternate directions. A useful answer should make clear whether the image is symbolic, illustrative, diagrammatic, or documentary-style, because documentary-style output can mislead audiences if it looks like evidence of a real scene.

Approval/safety note: Approve the concept only if it avoids fabricated rights, avoids implying endorsement by real people or brands, and does not present a synthetic scene as factual documentation. Provenance signals such as C2PA metadata or invisible watermarking can help communicate generation history, but they do not prove ownership, consent, or truth.

Prompt 11: User-Owned Portrait Edit Contract

Purpose: Use this prompt before editing a recognizable person in a portrait, headshot, team image, creator photo, executive profile, or contributor bio. It forces confirmation of ownership or authorization, explicit consent from recognizable people, and a narrow list of allowed transformations.

Copy-paste prompt:

You are my portrait-edit production controller. I will provide or describe a source photo only if I own it or am authorized to distribute and edit it.

Rights confirmation: [I_OWN_OR_AM_AUTHORIZED]
Recognizable-person consent confirmation: [CONSENT_CONFIRMED]
Source photo description: [PHOTO_DESCRIPTION]
Permitted transformation: [ALLOWED_CHANGE]
Regions that must not change: [NON_TARGET_REGIONS]
Identity-preservation requirements: [IDENTITY_REQUIREMENTS]
Disallowed changes: [DISALLOWED_CHANGES]
Final use: [USE_CASE]

Create:
1. A safe edit instruction that preserves the person's identity and all non-target regions.
2. A mask or region description for the intended edit.
3. A list of features that must remain unchanged, including face shape, age impression, body type, clothing unless specified, background unless specified, and accessories unless specified.
4. A rejection note for any requested change that would misrepresent the person, fabricate consent, or imply an event that did not occur.
5. A reviewer checklist for before-and-after approval.

Required inputs: Supply an affirmative rights confirmation and consent confirmation. If either is missing, the workflow should stop or return a rights gap, not proceed with a cosmetic or identity-affecting edit.

Expected output: Expect a controlled edit instruction that names the target region and protects everything else. For example, it may permit “remove glare from eyeglasses” while prohibiting “make the person look younger” unless the subject explicitly requested and approved that representation.

Approval/safety note: Do not use this prompt to create misleading images of real people, to imply endorsements, to alter identity-sensitive traits without consent, or to fabricate authorization. Human approval is required before publication because identity preservation is a production requirement, not a guaranteed model behavior.

Prompt 12: Authorized Product-Photo Precision Edit

Purpose: Use this prompt for owned or authorized product photography where the product itself must remain accurate while specific defects, lighting issues, crops, dust, or staging problems are corrected. This is a strong candidate for precision-sensitive editing, where OpenAI positions Sunburst as the model option for tighter control, while teams should still test outcomes against their own acceptance criteria.

Copy-paste prompt:

You are my product-photo edit spec writer. I will use only a source image that my organization owns, created, licensed, or is otherwise authorized to edit and distribute.

Rights confirmation: [RIGHTS_CONFIRMED]
Product name or SKU: [PRODUCT]
Source image description: [PHOTO_DESCRIPTION]
Target edit: [EDIT_REQUEST]
Regions that must remain unchanged: [NON_TARGET_REGIONS]
Product accuracy requirements: [MATERIAL_COLOR_SHAPE_LABEL_REQUIREMENTS]
Forbidden changes: [FORBIDDEN_CHANGES]
Output placement: [PLACEMENT]

Write:
1. A precise image-edit instruction.
2. A target-region description.
3. A non-target preservation list.
4. A product-accuracy checklist.
5. A warning list for edits that could change the advertised product, packaging, regulatory label, or included accessories.

Required inputs: Provide product identifiers, rights confirmation, the permitted edit, and accuracy constraints for color, materials, dimensions, labels, ingredient panels, compliance marks, or accessories. If labels or claims are visible, specify whether they must remain pixel-identical or be excluded from the final crop.

Expected output: Expect a concise edit spec that distinguishes retouching from product alteration. “Remove dust on the tabletop” is materially different from “make the device thinner,” because the latter may misrepresent the product sold.

Approval/safety note: Preserve all non-target regions and reject any output that changes product claims, safety labels, brand marks, included components, or material appearance beyond the approved transformation. Do not invent licenses, model releases, product approvals, or retail permissions.

Prompt 13: Background Replacement Without Subject Drift

Purpose: Use this prompt when replacing a background behind a person, object, product, pet, artwork, or set piece while preserving the foreground subject. Background replacement is a common place for subtle errors: edges change, hair is redrawn, product silhouettes warp, shadows become inconsistent, or the new setting implies a location or endorsement that is not true.

Copy-paste prompt:

You are my background-replacement edit planner.

Source ownership or authorization confirmed: [YES_NO]
Recognizable-person consent confirmed, if applicable: [YES_NO_OR_NOT_APPLICABLE]
Source image subject: [SUBJECT]
Current background: [CURRENT_BACKGROUND]
Replacement background: [NEW_BACKGROUND]
Foreground regions to preserve exactly: [FOREGROUND_PRESERVATION]
Allowed edge refinements: [EDGE_RULES]
Lighting and shadow requirements: [LIGHTING_RULES]
Misleading-context risks to avoid: [RISKS]

Produce:
1. A background-only edit instruction.
2. A foreground preservation contract.
3. A list of subject features that must not be regenerated or restyled.
4. A lighting and shadow consistency note.
5. A review checklist for detecting subject drift, false location implications, and rights-sensitive background elements.

Required inputs: Confirm rights to the source image and consent for recognizable people. Describe the replacement background in generic or licensed terms, not by asking for an unlicensed brand location, private venue, copyrighted artwork, or identifiable event backdrop.

Expected output: Expect a background-only instruction with clear boundaries. The answer should instruct the editing step to preserve the foreground subject, clothing, expression, product geometry, and all non-target details unless the user has explicitly authorized a different change.

Approval/safety note: Reject outputs where the replacement background makes a real person appear to attend an event, visit a location, use a product, or endorse an organization without consent. Provenance metadata may travel with the asset, but it cannot substitute for subject consent or location rights.

Prompt 14: Composition Change With Protected Regions

Purpose: Use this prompt when converting an image to a new crop, layout, banner, thumbnail, cover frame, or ad composition. Composition changes can be safer than full regeneration if the goal is channel fit, but they can still alter identity, product proportions, legal text, or scene meaning if the non-target regions are not protected.

Copy-paste prompt:

You are my composition-change planner for production images.

Source rights confirmed: [YES_NO]
Recognizable-person consent confirmed, if applicable: [YES_NO_OR_NOT_APPLICABLE]
Original image description: [ORIGINAL_DESCRIPTION]
New placement and aspect ratio: [PLACEMENT_AND_RATIO]
Primary subject: [PRIMARY_SUBJECT]
Secondary elements to preserve: [SECONDARY_ELEMENTS]
Safe areas for copy or UI overlays: [SAFE_AREAS]
Elements that must not be extended, invented, cropped, or moved: [PROTECTED_ELEMENTS]
Allowed composition changes: [ALLOWED_CHANGES]

Return:
1. A composition edit brief.
2. A protected-region list.
3. A crop and extension risk assessment.
4. A safe-area recommendation.
5. A before-publication QA checklist.

Required inputs: Provide the new placement, aspect ratio, overlay needs, protected regions, and any content that must remain visible for legal, accessibility, or editorial reasons. If the image contains a real person, confirm consent for the revised context and distribution channel.

Expected output: Expect a layout-aware edit brief that tells the image workflow where expansion, cropping, or repositioning may occur. The output should identify high-risk areas such as faces, hands, product edges, package labels, background signage, badges, and fine text.

Approval/safety note: Do not approve a composition change that changes the apparent relationship among people, products, locations, or claims. Preserve non-target regions and record the final crop rationale for auditability.

Prompt 15: Multi-Turn Edit Plan and Stop Conditions

Purpose: Use this prompt before starting a sequence of edits where several rounds may build on the same source image or prior outputs. OpenAI’s image-generation guide distinguishes single-prompt Image API work from conversational or multi-step workflows using the Responses API image-generation tool, but multi-turn consistency still requires disciplined change control and human review.

Copy-paste prompt:

You are my multi-turn image edit producer. Build a step-by-step plan that minimizes drift and keeps a record of every approved change.

Source rights confirmed: [YES_NO]
Consent confirmed for recognizable people, if applicable: [YES_NO_OR_NOT_APPLICABLE]
Initial image description: [INITIAL_IMAGE]
Final goal: [FINAL_GOAL]
Approved transformations: [APPROVED_TRANSFORMATIONS]
Non-target regions to preserve across all turns: [NON_TARGET_REGIONS]
Maximum number of edit turns: [MAX_TURNS]
Stop conditions: [STOP_CONDITIONS]
Reviewers and approval gates: [REVIEWERS]

Create:
1. An ordered edit sequence.
2. The instruction for each turn.
3. What must be compared after each turn.
4. A drift log template.
5. Stop conditions for identity drift, product drift, text corruption, background rights risk, or unapproved changes.
6. A final acceptance checklist.

Required inputs: Provide the source rights status, consent status, the final goal, the allowed transformations, and the maximum number of iterations. A production team should define stop conditions before rendering, not after becoming attached to a near-miss variation.

Expected output: Expect a sequenced plan that isolates edits. For example, correct exposure first, replace background second, then adjust crop last, with a comparison checklist after every turn.

Approval/safety note: Stop if a turn changes a person’s identity, alters a product, damages text, removes provenance cues, or introduces unauthorized background elements. Do not rely on multi-turn editing to preserve every detail automatically; keep source references and version records.

Prompt 16: Brand-Consistency Matrix for Image Outputs

Purpose: Use this prompt to turn brand rules into an evaluation matrix that reviewers can apply to generated images and edited photos. The matrix helps teams compare visual style, tone, subject treatment, accessibility, channel fit, and rights controls across multiple assets instead of approving images based on personal taste.

This collection of GPT-5.5 prompts for marketing directors covers campaign strategy, brand positioning, and cross-channel optimization, giving creative teams a structured brand context to translate into consistent visual briefs and approval criteria. The complete 50 GPT-5.5 Prompts for Marketing Directors: Campaign Strategy, Budget Allocation, Performance Analytics, Brand Positioning, and Cross-Channel Optimization article provides the destination-specific detail for this section’s Brand Consistent AI Visuals decision because the target explicitly addresses brand positioning and cross-channel marketing prompts, making it a stronger operational companion than a general future-of-content article.

Copy-paste prompt:

You are my brand image governance assistant. Convert the following brand rules into a practical scoring matrix for generated and edited production images.

Brand attributes: [BRAND_ATTRIBUTES]
Color and lighting guidance: [COLOR_LIGHTING]
Composition rules: [COMPOSITION_RULES]
Subject treatment rules: [SUBJECT_RULES]
Prohibited aesthetics or clichés: [PROHIBITED_STYLES]
Accessibility requirements: [ACCESSIBILITY_REQUIREMENTS]
Rights and consent requirements: [RIGHTS_REQUIREMENTS]
Channels covered: [CHANNELS]

Produce a matrix with columns:
Criterion, Pass standard, Warning signs, Reject conditions, Reviewer notes.
Include rows for style, tone, subject accuracy, product accuracy, protected regions, text handling, accessibility, provenance handling, and channel fit.

Required inputs: Provide actual brand rules rather than vague preferences. Include prohibited visual tropes, required contrast standards, color constraints, treatment rules for people, and any legal review requirements for claims, logos, packaging, or sponsored imagery.

Expected output: Expect a table that reviewers can use on contact sheets, first drafts, and final exports. A strong matrix has explicit reject conditions, such as “recognizable person altered without consent,” “package claim changed,” or “synthetic document appears to be real evidence.”

Approval/safety note: The matrix should never create rights that do not exist. Brand consistency is a quality-control framework; it does not authorize use of third-party IP, unlicensed photography, or recognizable people without consent.

Prompt 17: Text-Risk Strategy for Images

Purpose: Use this prompt when an image might contain headlines, captions, product labels, UI snippets, signage, charts, document fragments, or small typography. Even with improved instruction following and quality settings, production teams should treat rendered text as a risk area that needs planning, review, and often deterministic post-production.

Copy-paste prompt:

You are my text-risk strategist for AI-assisted image production.

Image purpose: [PURPOSE]
Text that seems necessary: [TEXT_ELEMENTS]
Text that can be avoided or added later in design software: [POST_PRODUCTION_TEXT]
Legal, medical, financial, safety, or product claims involved: [CLAIMS]
Languages and character sets: [LANGUAGES]
Minimum legibility requirements: [LEGIBILITY]
Output channels: [CHANNELS]

Recommend:
1. Which text should be excluded from generation and added later.
2. Which text, if any, may be included in the image prompt.
3. A prompt wording strategy to reduce text errors.
4. A QA checklist for misspellings, hallucinated words, claim changes, trademark misuse, and accessibility.
5. A final handoff note for designers or web producers.

Required inputs: List every required text element, separate decorative from legally meaningful copy, and identify whether the final asset will go through design software, a CMS, or automated image optimization. If text contains claims, specify the approved wording.

Expected output: Expect a strategy that often recommends generating a clean visual without critical text, then adding typography through deterministic layout tools. The answer should flag text that cannot tolerate spelling drift, such as prices, dosages, compliance labels, SKU names, deadlines, and legal disclaimers.

Approval/safety note: Do not approve image text that changes a claim, invents a certification, corrupts a label, or makes a misleading promise. Higher quality settings do not guarantee flawless typography, so human review remains mandatory.

Prompt 18: Variation Selection and Production Decision Log

Purpose: Use this prompt after producing a contact sheet or small set of variations. It converts subjective preference into a documented selection process based on brief fit, rights status, identity or product preservation, brand consistency, text risk, provenance handling, and channel readiness.

Copy-paste prompt:

You are my image variation selection reviewer. Evaluate candidate images against the approved brief and produce a production decision log.

Approved brief summary: [BRIEF]
Candidate descriptions or filenames: [CANDIDATES]
Source-photo rights status for each candidate, if applicable: [RIGHTS_STATUS]
Consent status for recognizable people, if applicable: [CONSENT_STATUS]
Brand matrix or criteria: [BRAND_CRITERIA]
Known risks: [KNOWN_RISKS]
Final channel: [CHANNEL]
Required export or handoff notes: [EXPORT_NOTES]

For each candidate, score:
Brief fit, visual quality, subject preservation, product accuracy, text risk, rights/consent readiness, brand consistency, accessibility, and provenance handling.

Then provide:
1. Recommended winner.
2. Required edits before approval.
3. Reject list with reasons.
4. Human review questions.
5. Final production log entry.

Required inputs: Provide the approved brief, candidate identifiers, rights status, consent status, brand criteria, and known risks. If a candidate derives from a real photo, include confirmation that the team owns or is authorized to edit and distribute the source and that recognizable people consented to the transformation and use.

Expected output: Expect a ranked evaluation with reasons, not a beauty contest. The best variation is the one that satisfies the brief with the lowest unresolved production risk, not necessarily the most dramatic or polished image.

Approval/safety note: Do not select a variation with unresolved rights, missing consent, altered identity, changed product facts, corrupted text, or misleading provenance assumptions. Record the chosen file, rejection reasons, required edits, reviewer, and approval date so later teams can audit why the asset entered production.

Prompts 19–25: QA, Provenance, Web Delivery, Approval, and Archive Control

Use these final seven prompts after image generation or editing, not as a substitute for visual inspection. ChatGPT-5.5 can structure QA, write acceptance checklists, normalize metadata, and prepare records; it is not the image-rendering model, and it should not be treated as proof that an output is lawful, authentic, consented, or visually flawless.

Prompt 19: Contact-Sheet QA Triage

Purpose: Convert a batch of candidate images into a structured contact-sheet review so editors can compare composition, channel fit, brand alignment, and production risk before selecting finalists.

Copy-paste prompt:

You are my production image QA coordinator. Review the candidate image set described below and produce a contact-sheet triage table.

Context:
- Campaign or page:
- Channel placements:
- Required aspect ratios or crops:
- Brand rules:
- Audience:
- Candidate image IDs or filenames:
- Generation or edit prompt version:
- Image model used, if known:
- Quality setting, size, format, and compression settings, if known:

For each candidate, evaluate:
1. Best use case.
2. Composition strength.
3. Subject clarity.
4. Brand consistency.
5. Channel/crop risk.
6. Text-in-image risk, if any.
7. Rights or consent dependencies.
8. Provenance metadata expectations.
9. Required human checks before approval.

Return:
- A ranked table.
- A reject/hold/advance decision for each candidate.
- A short note explaining why the top candidate should proceed.
- A list of follow-up edits that must not change protected subjects, identities, logos, or non-target regions unless explicitly authorized.

Required inputs: Candidate IDs, target channels, prompt version, brand rules, crop requirements, and any known generation settings.

Expected output: A ranked QA table with clear advance, hold, and reject decisions that can be copied into a creative review ticket.

Approval/safety note: If any candidate uses a real person, owned product photo, recognizable property, or customer-supplied asset, require confirmation that the user owns or is authorized to distribute the image and that the proposed use matches the granted permission.

Prompt 20: Artifact and Anatomy Review

Purpose: Create a focused inspection checklist for visual defects that commonly escape high-level creative review, including hands, limbs, reflections, shadows, seams, product geometry, and localized edit artifacts.

Copy-paste prompt:

You are my visual defect reviewer. Build a human inspection checklist for the image described below.

Image description or candidate ID:
Intended use:
Real person, product, property, or synthetic-only image:
Protected regions that must not change:
Known edits already applied:
Target output size and crop:

Inspect for:
- Hands, fingers, limbs, teeth, eyes, ears, hairlines, and facial symmetry.
- Product proportions, labels, seams, buttons, ports, packaging edges, and reflections.
- Lighting direction, shadow consistency, contact shadows, and perspective.
- Background continuity, masking halos, transparent edges, blur, noise, and texture repetition.
- Unwanted text, pseudo-logos, distorted signage, or brand-confusing marks.
- Any change that could misrepresent a real person, product, place, or event.

Return a severity table with: issue, location, severity, likely cause, acceptable fix, and whether the image must be regenerated, edited, or rejected.

Required inputs: Image description, intended placement, protected regions, known edits, and whether the image includes real people or products.

Expected output: A severity-ranked checklist that separates cosmetic fixes from rejection-level defects.

Approval/safety note: Do not approve anatomy or identity corrections on a real person unless the user confirms the edit is authorized and the output will not misrepresent the person’s identity, health, age, event participation, or consent.

Prompt 21: Exact-Copy Review Against the Brief

Purpose: Compare the final candidate against the approved creative brief and edit contract so the team can catch missing constraints, accidental style drift, or unauthorized changes before publication.

Copy-paste prompt:

You are my final image compliance reviewer. Compare the final candidate against the approved brief and identify exact matches, deviations, and approval blockers.

Approved brief:
Original source asset notes, if any:
Permitted transformation:
Protected regions:
Required visual elements:
Prohibited visual elements:
Brand rules:
Channel requirements:
Final candidate description or reviewer notes:

Return:
1. A requirement-by-requirement compliance table.
2. A list of exact-copy requirements that appear satisfied.
3. A list of deviations that require human review.
4. A list of unauthorized changes, if any.
5. A publication decision: approve, approve with documented exception, revise, or reject.
6. The minimum edit instruction needed if revision is recommended.

Required inputs: Approved brief, protected regions, required and prohibited elements, final image notes, and publication context.

Expected output: A compliance matrix that makes creative approval auditable instead of relying on memory or informal comments.

Approval/safety note: For user-owned photos, the reviewer must confirm that the final image preserves identity and non-target regions unless the rights holder explicitly authorized those changes.

Prompt 22: Provenance Interpretation Note

Purpose: Prepare an internal provenance explanation that distinguishes metadata and watermark signals from legal conclusions. OpenAI states that ChatGPT Images 2.5 uses C2PA metadata and invisible SynthID watermarking, while also stating that no single provenance mechanism is sufficient.

Copy-paste prompt:

You are my provenance documentation assistant. Draft a plain-language provenance note for this image.

Image ID or filename:
Creation path:
- ChatGPT, Codex, Image API, or Responses image-generation tool:
- Image model used, if known:
- Generation or edit action:
- Source assets:
- Owner or authorization status for each source asset:
- Prompt version:
- Human approver:
- Known provenance signals present or expected:
- Distribution channel:

Write:
1. A short internal provenance summary.
2. A public-facing disclosure option, if disclosure is required by policy.
3. A legal-risk note explaining what provenance does NOT prove.
4. A checklist for preserving metadata during export, compression, CMS upload, and syndication.
5. A warning if ownership, consent, or source-asset authorization is missing.

Required inputs: Creation path, model information where available, source assets, authorization status, prompt version, and destination channel.

Expected output: A provenance note that can travel with the asset record and inform editorial, legal, and platform teams.

Approval/safety note: C2PA metadata or SynthID watermarking should never be described as proof of copyright ownership, factual authenticity, subject consent, or distribution rights.

Prompt 23: Alt Text and Caption Drafting

Purpose: Generate accessible alt text, editorial captions, and optional credit lines from confirmed image facts while avoiding unsupported claims about what happened, who consented, or what the image proves.

Copy-paste prompt:

You are my accessibility and caption editor. Draft publication-ready text for the image below.

Confirmed image facts:
Page or article topic:
Audience:
Image type: generated, edited from owned photo, product photo, illustration, diagram, or other:
People shown, if any:
Products or brands shown, if any:
Facts that must not be inferred:
Required disclosure or caption policy:
Tone:

Return:
- Concise alt text focused on visual information needed for the page.
- Longer descriptive alt text option if the image carries essential meaning.
- Editorial caption.
- Optional production note for internal use.
- Phrases to avoid because they imply unverified events, ownership, consent, or authenticity.

Required inputs: Confirmed visual facts, page topic, image type, policy requirements, and prohibited inferences.

Expected output: Alt text and captions that support accessibility and editorial accuracy without overstating provenance or factual certainty.

Approval/safety note: Captions for real people or owned photos must not imply endorsement, consent, participation, location, or identity details beyond what the rights holder and editorial record authorize.

Prompt 24: Deterministic Web Optimization and Filename Plan

Purpose: Turn an approved image into a repeatable web-delivery plan covering filenames, formats, compression, responsive variants, and CMS notes. OpenAI’s image-generation guide documents output customization for quality, size, format, and compression; this prompt keeps those choices explicit rather than improvised.

Copy-paste prompt:

You are my image delivery operations planner. Create a deterministic web optimization plan for the approved image.

Approved asset ID:
Page or campaign:
Primary keyword or descriptive subject:
CMS or delivery system:
Required placements:
Target crops or dimensions supplied by the team:
Allowed formats:
Compression policy:
Metadata preservation policy:
Accessibility text status:
Provenance note status:
Version number:
Publication date or sprint:

Return:
1. Filename pattern using only lowercase letters, numbers, and hyphens.
2. Proposed filenames for master, web, thumbnail, and social variants.
3. Export table with placement, dimensions supplied by the team, format, compression setting, and metadata handling.
4. Cache-busting or versioning convention.
5. QA checks after upload.
6. Rollback filename or archive rule.

Required inputs: Approved asset ID, descriptive subject, CMS constraints, allowed formats, compression rules, metadata policy, and version number.

Expected output: A deterministic export and naming table that developers and content teams can reproduce across environments.

Approval/safety note: Do not strip provenance metadata by default when organizational policy requires preservation; if optimization removes metadata, record that decision and the approver.

The selected playbook covers AI-driven WordPress content pipelines using GPT-5.2, Claude Opus 4.7, WP-CLI over SSH, and Hostinger hosting for high-volume publishing. The complete WordPress + AI Content Pipelines: The Hostinger + WP-CLI Production Playbook article provides the destination-specific detail for this section’s WordPress Image Optimization decision because it is the closest WordPress production-workflow match and supports the publishing side of optimizing AI-generated images inside WordPress content pipelines.

Prompt 25: Final Approval and Archive Manifest

Purpose: Produce the final manifest that links the brief, source assets, rights confirmation, prompt versions, model records, QA decisions, provenance notes, web exports, and human approvals.

Copy-paste prompt:

You are my production archive manager. Build the final approval and archive manifest for this image workflow.

Project:
Final asset IDs:
Source asset IDs:
Rights and consent confirmation:
Permitted transformation:
Prompt versions used:
Text planning model:
Image model and snapshot if recorded:
Image API or Responses image-generation tool route:
Generation or edit action:
Quality setting:
Size, format, and compression settings:
Human QA reviewers:
Approval decision:
Known exceptions:
Alt text and caption:
Provenance note:
Web export filenames:
Archive location:
Retention or deletion policy:

Return a manifest with:
- Executive summary.
- Rights and consent record.
- Technical generation/edit record.
- QA and acceptance record.
- Provenance and disclosure record.
- Web delivery record.
- Exceptions and unresolved risks.
- Final approval line with approver, date, and decision.

Required inputs: Final assets, source assets, rights confirmation, prompt versions, model records, QA outcomes, web exports, and approval names.

Expected output: A complete archive manifest suitable for internal audit, rollback, re-export, localization, or future campaign reuse.

Approval/safety note: If rights, consent, provenance handling, or protected-region preservation is unresolved, the manifest should mark the image as not approved for publication.

Implementation Guidance for Teams Running These Prompts

Recommendation: Run a pilot before embedding these prompts into a production CMS, design system, or asset pipeline. Select a small but representative set of image tasks: one synthetic editorial image, one authorized product edit, one owned-photo portrait edit, one background replacement, and one web-export workflow. The pilot should measure whether the prompts improve review consistency, reduce missing approvals, and produce archive records that another team can understand without interviewing the original creator.

Model and route recording: Record ChatGPT-5.5 as the planning, governance, or QA model when it is used for these text workflows. Separately record the image model and route used for rendering or editing. OpenAI positions GPT-Image-2.5 Flare for fast everyday image generation and GPT-Image-2.5 Sunburst for precision-sensitive editing; both are documented for image generation, image edits, and inpainting, and both accept text and image input and output images. When using the Responses API image-generation tool, record the top-level mainline model separately from the image tool model, because those are different selections in OpenAI’s documented routing.

Snapshot discipline: Where your logs expose a model snapshot, capture it in the manifest. The official model pages list default snapshots for the 2.5 image models, including gpt-image-2.5-sunburst-2026-09-08 and gpt-image-2.5-flare-2026-09-08. If your workflow records only the model ID, preserve the model ID, run date, prompt version, quality setting, action, size, format, and compression settings so later reviewers can understand the production context.

Prompt versioning: Treat each prompt template like production configuration. Assign a stable ID, owner, version number, approval date, and change note. Do not silently edit a rights prompt, preservation prompt, or provenance prompt after assets have been approved; create a new version and record which assets used the old one. This prevents a future audit from applying today’s policy language to yesterday’s output.

Human acceptance: Require a named human approver for every published asset. The acceptance criteria should include visual quality, brief compliance, rights confirmation, consent status where relevant, protected-region preservation, provenance handling, accessibility text, and web-export verification. A model-generated QA table can organize the decision, but it should not be the decision-maker for publication.

Failure recovery: Define what happens when a render fails, an edit drifts, metadata is stripped, or a reviewer finds a blocker after upload. The safe default is to preserve the source asset, prompt version, failed output, error note, reviewer comment, and rollback path. Retry only when the task is idempotent or when duplicate outputs will not create publication confusion. For real photos, do not “fix” failure by broadening permissions or inventing consent; return to the rights holder or project owner.

Useful Links

• https://openai.com/index/introducing-chatgpt-images-2-5/
• https://developers.openai.com/api/docs/guides/image-generation
• https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst
• https://developers.openai.com/api/docs/models/gpt-image-2.5-flare
• https://deploymentsafety.openai.com/chatgpt-images-2-5
• https://developers.openai.com/api/docs/models/gpt-5.5