create-variants
ChatGPTStart making alternate sets of a screenshot project to test against it with Apple's Product Page Optimization. Apple runs the test; these are the treatments you add in App Store Connect. Returns immediately: concepts take under a minute and one-change variants about 2 minutes, so call get-variants next, which waits for it. mode: - explore (default): up to 3 bold concepts. Each is one idea about why people install, built from 2 or 3 changes (a new headline angle, a different opening card, a new palette, how the phones sit). Test the one you believe in most against the original: a new or small app's test can only detect a large difference. - refine: one change per variant (headline angle, opening card, look, or phones), so a win says exactly what worked. For an app with steady product page traffic, or a concept that already won. Steer it (optional): - idea: one line of the owner's own idea, e.g. "lead with the review card". In explore the first concept is built from it; in refine it is one more variant on top of the picked axes. Either way it uses only what the set already says. - axes (refine): exactly which things to change, one variant each: hook, lead, visual, device. Every variant keeps the project's real app screens, keeps its copy readable and its phones visible; any variant that fails those checks is refunded. Costs 5 credits per variant, charged at the start. Apple tests up to 3 treatments and Google Play 2, so an iOS set holds at most 3 variants and an Android set 2. Apple Watch sets cannot have variants: Apple does not test Watch product pages.
edit-screenshots
ChatGPTMake changes to an existing screenshot project. Use natural language to describe what you want to change. Costs 5 credits per edit. What you can change: - Text: headlines, subtitles, badge text, font size, font family (Inter, Poppins, Montserrat, DM Sans, Space Grotesk, etc.) - Text styling: per-word color, bold, italic, underline, highlight pills (colored background behind words), gradient text, text stroke outlines - Colors: brand palette, gradient backgrounds, accent colors, text colors. A set can rotate MULTIPLE accents card by card ("pink, then coral, then brick") and each card keeps its own highlight colour. - Background textures (ask for one explicitly, they are never added on their own): diagonal-stripe, crosshatch, checkerboard, zigzag, hairline-grid, dot-grid, waves, grain, radial-rays, concentric-circles. Kept subtle by design so they cannot affect headline contrast. Tiled textures run continuously across the whole set; radial-rays and concentric-circles can be centred on the middle of the set so the pattern fans out across every card. - Layouts (10 types): text-top-device-bottom, text-top-device-tilted, device-hero, social-proof, review-clip, screen-hero, lifestyle-hero, stats-hero, metric-badge, annotated-feature. The AI picks a narrative arc (HOOK → EDUCATE → PROVE → CONVERT) across the set. - Layout params: deviceScale (small/medium/large), deviceSide (left/right), textPosition (above/below), textAlign (left/center) - Device mockups: perspective variants (flat, left-15, right-15, left-30, right-30, isometric, top-down, landscape-left, landscape-right), 2D rotation, resize, reposition - Frame color: recolor the device frame: "natural" (default), "black", "white", "gold". Examples: "make the iPhone gold", "black titanium finish", "white iPhone". Requires Growth plan or higher. - Switch devices: "make this for Apple Watch" / "duplicate for Android tablet" clones the project at the target device's canvas size - Add/remove cards: add a social proof card, remove card 3, add a marketing title card - Compound fields (auto-positioned): statRow, laurelStat, pressBanner, starRating, quote, credential, guarantee - Panoramic backgrounds: one wide AI or Pexels image sliced across multiple cards ("pano the background across cards 1-3"). The edit only tags the cards; call generate-panoramic-background afterwards to create and slice the wide image. - Panoramic element spans: stretch a foreground image, device-mockup, or shape across adjacent cards ("pano the device across cards 1-2") - Floating elements: add/edit badges, star ratings - Shapes: glow orbs, waves, blobs, rounded rectangles, circles, custom SVG paths - Decorative shapes: 77 library shapes: leaf, flower, cloud, sparkle, heart, rocket, trophy, crown, coffee-cup, airplane, dollar-sign, paw-print, wing-left/wing-right (for laurels), and many more - Backgrounds: solid, subtle-gradient, rich-gradient, photo (Pexels), or ai-generated; change gradient colors/angle; set a backgroundPrompt for AI-generated - Style: shadows, opacity, border radius, rotation, blur What isn't supported (the AI will flag these in unsupportedAsks): - Multiple devices side-by-side in a single card (e.g. iPhone + Watch in one scene). Each card renders one device. Use the duplicate-for-device workflow instead, or split across cards. - Uploading a specific user-supplied screenshot into a mockup: open the project URL and drop it into the frame in the builder, then edit again. Example edit messages: - "Make the headlines larger and use Bebas Neue font" - "Change the color scheme to blue (#2563EB) across all cards" - "Add a social proof card with 5 stars and a testimonial quote" - "Tilt the phone on card 2 to the left" - "Pano the background photo across cards 1-3" - "Stretch the device on card 1 across into card 2" - "Make this project for Apple Watch" - "Recolor the iPhone frame to gold across all cards" - "Replace card 3 with a CTA card saying Download Free" - "Switch card 2 to device-hero layout with statRow showing our three key metrics" - "Add decorative leaf and sparkle shapes scattered in t…
generate-background
ChatGPTGenerate an AI background image for a specific card using Gemini. The background is generated based on a text prompt and applied directly to the card. Costs 6 credits. Good prompts describe mood, lighting, and color, not objects or text: - "Deep purple nebula with soft pink and blue light rays" - "Warm sunset gradient with golden bokeh particles" - "Dark moody atmosphere with teal and emerald glow" - "Clean minimal white-to-light-gray gradient with subtle noise texture" The generated image is cropped to exact device dimensions and set as the card's background.
generate-panoramic-background
ChatGPTGenerate one wide background image and slice it across multiple cards so they read as a continuous scene when the App Store gallery scrolls. The store gallery gap is accounted for, so slices line up after Apple's gutter. Two image sources: - prompt: AI-generated via Gemini. Costs 6 credits. - pexels_query: stock landscape photo from Pexels. Free. If you omit both, the tool fulfills a panoramic the chat already set up: when generate-screenshots or edit-screenshots returns a design with a panoramic background (e.g. after "pano the background across cards 1-3"), the cards are tagged but the wide image isn't created yet. Calling this tool with just the project_id picks up that pending panoramic and generates it. Good prompts describe a continuous scene, not objects or text: - "Misty mountain range at dawn, soft pink-to-blue gradient sky" - "Calm ocean horizon at golden hour with gentle waves" - "Abstract flowing emerald and teal silk waves"
generate-screenshots
ChatGPTCreate a complete set of App Store screenshot designs for an app. Returns a project URL where the developer can preview, refine, and export final PNGs. The design runs in the background and takes about 1 to 2 minutes, so this tool returns right away with status "generating" and a project_id. Tell the user the set is being designed, then call get-project with that project_id. get-project waits up to about 40 seconds for the set before it answers, so if it still says "generating", call it again straight away until it says "ready". IMPORTANT: fill codebase_context before calling this tool. You will not have the app's source here, so gather the same facts from the user or from their App Store listing or website: what the app does and who it is for, its brand colours, its main screens, the value proposition, and the app category. The more context you provide, the better the screenshots will be. Call prepare-screenshot-brief first for the category playbook. The AI picks a narrative arc that fits the app's sell angle (trust-first for finance/health, visual-first for creative/lifestyle, problem-first for pain-relief apps, numbers-led for SaaS/analytics, community-first for social, feature-forward for multi-mode apps) and assigns a job to each card: HOOK → EDUCATE → PROVE → CONVERT. 10 layout types available per card: - text-top-device-bottom: headline top, flat device bottom (feature/educate) - text-top-device-tilted: headline left, tilted device right (or mirrored) - device-hero: headline + large centered device, product-forward - social-proof: stars + quote + laurel stat, no device (testimonial-led) - review-clip: rating-statement headline + stars + quote + bottom-clipped device - screen-hero: top-clipped device + centered headline + optional trust stat - lifestyle-hero: full-bleed photo background + text overlay (no device) - stats-hero: giant centered stat(s), no device, numbers-led - metric-badge: centered device + chunky "achievement card" floating over its screen - annotated-feature: tilted device + side callout chip linked by a connector line Each card can carry auto-positioned compound fields: statRow, laurelStat, pressBanner, starRating, quote, credential ("FDIC insured"-style trust pills), guarantee ("Cancel anytime"-style risk-reversal pills). Trust signals are opt-in: supply the real number, quote or press name, or the card ships clean rather than inventing one. Panoramic backgrounds slice one wide AI-generated or Pexels image across multiple cards for cohesion. Panoramic element spans stretch a foreground image or device across adjacent cards. Note: if the design comes back with a panoramic background, the chat only tags the cards; call generate-panoramic-background afterwards to actually create and slice the wide image. Design features: - Rich text with per-word color, weight, italic, underline, highlight pills, gradient fills, stroke outlines, and emoji - Device perspective variants: flat, left-15/30, right-15/30, isometric, top-down, landscape - 77 decorative shapes across 13 categories (nature, weather, celebration, social, tech, health, food, travel, abstract, finance, education, pets, emoji), plus 17 core geometric shapes - Laurel stats: wing-left + wing-right shapes flanking a hero number App Store 60/40 rule enforced: minimum 60% of cards must show a device mockup, maximum 40% can be marketing-only. Costs 5 credits per generation, charged when the design finishes.
get-project
ChatGPTRetrieve a screenshot project: whether its design is finished, and its full state. Use it to check on a set started with generate-screenshots, and to inspect cards before making edits. Free: no credit cost. status tells you where the design stands: - generating: still being designed. This call already waited up to about 40 seconds for it, so call get-project again straight away. - ready: the cards exist. Show the user the project URL and offer edits or export. - no_design: the design finished without cards. message holds the reply, usually a question to answer with edit-screenshots. - failed: the design did not finish. Credits were refunded; run generate-screenshots again. The project field holds the canvas state with: - cards[]: each card has an id, elements array, and optional background settings - Each element has: type (text, device-mockup, shape, badge, image, star-rating), position (x, y), size (width, height), zIndex, and type-specific properties - Text elements: fontFamily, fontSize, fontWeight, color, segments (for multi-color text with per-word color, bold, italic, underline, highlightColor) - Device mockups: perspectiveVariant (flat, left-15, right-15, left-30, right-30, isometric, top-down, landscape-left, landscape-right), screenshotImage (null if no upload), frameStyle (realistic | none), showIsland (false hides the Dynamic Island pill; Apple accepts screenshots either way) - projectMeta: globalVisualTheme, brandColors, mood, appCategory
get-variants
ChatGPTCheck on a run started with create-variants, and list a project's test variants. While the run is still going this call waits up to about 40 seconds before answering; if it still says generating, call it again straight away. Free: no credit cost. status: - generating: still being made. Call get-variants again now. - done: the variants are ready. Each has a builder URL to preview, refine in chat and export, a label, and the idea it tests. Any that were refunded are listed with the reason. - failed: nothing could be made this time and every credit was refunded. Run create-variants again. - none: no run has been started for this project.
list-devices
ChatGPTShow all supported device specs for App Store and Play Store screenshots. Use the device ID when generating screenshots.
prepare-screenshot-brief
ChatGPTGet a category playbook to prepare for screenshot generation. Call this BEFORE generate-screenshots, and also whenever the user asks what their App Store screenshots should show or lead with (for example "what should my first screenshot focus on?"): answer from the playbook it returns. Free (no credits). Pass app_category to get category-specific guidance pulled live from AppScreenshotStudio: the frame-1 hook playbook (default layout, what to avoid, caption patterns), the recommended story_flow and mood, and the exact facts worth gathering about THIS app for its category (e.g. security and compliance signals for finance, real gameplay art for games, the outcome for fitness). Returns: - A category playbook (frame-1 hooks + layout + arc + mood) for the app_category - A category-specific research focus (what to ask the user about this app) - Headline tips + the codebase_context schema to fill in The more you gather here, the better generate-screenshots does on the first try.
render-screenshots
ChatGPTExport a screenshot project to high-resolution PNG files at exact App Store dimensions. Returns download URLs for each card; URLs stay valid for 7 days, so save the PNGs to disk promptly (re-rendering is free if a URL has expired). Costs no credits, but export is unlocked by the first payment: on a trial this returns 402 PAYMENT_REQUIRED, so check the plan before promising a user their PNGs. Pass device_ids to get the same canvas at several device sizes in one call, re-laid-out for each, instead of building a second project: an App Store submission wants the 6.9-inch iPhone and, when the app supports iPad, the 13-inch iPad, and Google Play is a separate listing. Rendering blocks until the PNGs are ready: expect roughly 20-40s per iPhone set and 60-120s per iPad set (the larger canvas renders slower), and those add up per device, so allow up to ~2 minutes each before treating it as failed. Note: device mockups show empty frames until app screenshots are added in the builder at the project URL.
