--- name: open-world-loading-screen description: Turn your own pet, product, invented character or yourself (with consent) into a loading screen in the generic look of an open-world action game: heavy outlines, flat-shaded painterly comic art, hot sun-baked city colours, a cocky three-quarter pose in front of a busy street. The art is made with GPT Image 2 from your photo (cheap low draft, then a medium final), with no words in it. The hero's exact name, a one-line tagline, a LOADING label and a small loading bar are drawn locally, so every word is spelled right, and a 1:1 profile crop is cut from the same art. A check script fails any caption outside the 3:4 profile-grid crop or under the app buttons, any caption that differs from the plan, low contrast and missing consent. Funny and friendly: no weapons, no crime, no game interface. Describes a genre, never copies a game. Optional matching set of up to 4. For creators, pet accounts, social teams and brands. metadata: type: image version: 1.0.0 category: Social posts & covers tagline: Use this Skill to turn your pet, product or character into a cocky open-world loading screen with an exact caption and a loading bar. made-for: Creators, Pet accounts, Social teams, Brands generates-with: PNG loading screen 9:16 (1080 x 1920) + a 1:1 profile crop (1080 x 1080), optional set of up to 4 requires: Node.js 18 or newer for the scripts; the sharp package for the caption step (run `npm install sharp` in the Skill folder, Node.js 20.9 or newer); your own FAL_KEY only if you run scripts/fal-run.mjs outside-costs: About $0.14 for one screen on fal (GPT Image 2, a low draft about $0.02, a medium final about $0.12), up to about $0.41 with two re-rolls; a set of 4 is about $0.57. The caption, the loading bar, the profile crop and the checks are free. Billed to your own fal account or your Meleon credits; quote the total and get a yes before the first paid call. updated: 2026-11-13 title: Open-world loading screen availability: live summary: Turn your pet, product or character into a cocky open-world action-game loading screen, with an exact name, a one-line tagline and a loading bar. command: /open-world-loading-screen my terrier Patch, photo attached, tagline "Mayor of the dog park. Self-appointed." output: PNG ยท 9:16 loading screen + 1:1 profile crop required-input: Photo or description + name + one-line tagline creator: meleon --- # Open-world loading screen Make the loading screen your pet, your product or your character deserves: a cocky three-quarter pose in front of a busy, sun-baked city street, in heavy-outline comic art, with the hero's name and a one-line tagline on a flat panel and a small loading bar along the bottom. The image model draws the picture and never a word. The name, the tagline, the LOADING label and the bar are drawn locally, so every letter is exact, and they sit where the profile grid and the app buttons leave them alone. A 1:1 profile crop comes from the same art. This Skill describes a genre (an open-world action-game loading screen). It never names or copies a game, a studio, a logo, a font or an interface, and it is funny and friendly: a hot dog stand, a convertible, a skateboard. No weapons, no crime, no violence. ## What you need - **The hero**, one of: - **a pet** (a clear photo of the animal, the whole face visible); - **a product** (a clean photo of the thing); - **an invented character** (a written description, or an existing picture of the character, such as a character sheet); - **yourself, or a person who has agreed** (see People). - **A description** of the hero in one sentence, and the **features that must stay recognisable** (markings, colours, shape, collar, hair, outfit). Write them with the user, from the photo. - **A name** of 1 to 3 words, at most 20 characters. Drawn in capitals by default. - **A tagline**, one line of attitude, at most 60 characters. Funny, never about crime or violence (`references/screen-grammar.md`, "Captions"). No invented claims about a real person. - **A look**: a preset (`sunset-street` is the default, `midday-boardwalk`, `dusk-neon`) or the user's own one-sentence art direction. - **Optional:** which side the caption sits on (`left` or `right`), brand colours, a font file (`.ttf` or `.otf`), the label (`LOADING`) and how far the bar is filled. - **Optional:** 2 to 4 names and taglines for a matching set (the same hero in a different scene each time). - **People.** Pets, products and invented characters need no consent. A photo of a person is allowed only if it is the user themselves, or someone who has agreed in plain words to appear (written is best). They must be an adult. Ask, and stop if the answer is not a clear yes. Never a celebrity or public figure, never a child, never a photo of someone who has not agreed. The person stays exactly themselves: same identity, age range, skin tone, hair, features and body, drawn in the comic style. Make nothing that suggests they did something they did not do, and no caption that is untrue or unkind about them. A tagline about a real person is a joke they would laugh at, not one that could hurt. The scripts refuse a person without consent recorded. If the photo or the name is missing, ask for it. Do not invent a pet's name, a brand or a claim. ## What you make 1. `screens/screen-01.png` (and `screen-02.png` ... for a set): 1080 x 1920 PNG, the art with the caption panel, the LOADING label and the bar drawn on. Post it as a Reel, a Story, a Short or a TikTok photo. 2. `screens/profile-01.png` ...: 1080 x 1080 PNG, the plain art (no words) centred on the hero's face, ready for a profile picture. Apps show it as a circle; the face is kept inside it. 3. `screens/sheet.png`: a set of 2 to 4 side by side. 4. `screens/manifest.json` and `screen.json`: the caption boxes and the plan, so the words can be redrawn for free. 5. `art/`: the text-free drafts and finals. ## How to run it **Budget: about $0.14 for one screen (a low draft about $0.02, a medium final about $0.12), up to about $0.41 with two re-rolls. A set of 4 is about $0.57 (about $0.86 with two re-rolls). `--model nano-banana-pro` for a final is a flat $0.15. The caption, the bar, the profile crop and the checks are free.** Before the first paid call, run `node scripts/build-requests.mjs --plan screen.json --stage estimate`, quote the total it prints, and get one explicit yes on it. At every paid step, tell the user the model, the number of calls and the cost, and ask again if the spend will pass the quoted total. To send a request: if the Meleon tools are connected, call `edit_image` for a request with `image_urls` (or `generate_image` for a character with no picture): its `prompt`, its `image_urls` (write the requests with `--out requests --inline`, so local images are data:image/ URIs; Meleon deletes uploads after 7 days), `model` "gpt-image-2", its `image_size` as `size`, its `quality` and `output_format`, `count` 1, and `skill` "open-world-loading-screen", so the run shows under this Skill in the user's Meleon dashboard. It costs Meleon credits (the price is in the tool's description), or nothing if the user saved their own fal key on Meleon. Otherwise POST the printed `input` to the fal model id in the request with the user's own FAL_KEY (`node scripts/fal-run.mjs --request --out art --approved` does this; the Skill never stores keys), or paste the prompt and images into any GPT-Image-class editor. 1. **Brief (free).** Confirm the hero, the photo, the description and the features to keep, the name and tagline (letter for letter), the look, the caption side, and for a person the consent and adult questions above. 2. **Plan (free).** `node scripts/new-screen.mjs --out screen.json --kind pet|product|character|person --description "..." --keep "..." [--photo hero.jpg] --names "Name1|Name2" --taglines "Line one.|Line two." (--preset sunset-street | --look "...") [--side left|right] [--panel #hex --ink #hex --accent #hex] [--label LOADING --percent 64] [--consent own-photo|written-consent --adult-confirmed]`. Then write each screen's `pose` and `scene` in `screen.json` (`references/prompt-patterns.md`). Show the user the plan in plain words. 3. **Spend gate (free).** `node scripts/build-requests.mjs --plan screen.json --stage estimate`. Quote the total and wait for one explicit yes. 4. **Draft (about $0.02).** `--stage draft --out requests --inline`, send it and open the result. Check: the hero is recognisable, a cocky pose, a busy street, no stray text, and the hero (arms and tail too) stays in the half opposite the caption, with open street where the panel goes. A trot or a stretched arm runs under the panel: pick a narrow pose (standing, sitting, lounging) and say which side the props are on. Re-roll by changing one thing in the pose or scene. Put the approved draft's path in the screen's `draft`. 5. **Screen 1 final (about $0.12).** `--stage final --screen 1 --out requests --inline`, send it, open it at full size and compare the hero with the photo: markings, colours, face. If GPT Image 2 drifts from the hero, try `--model nano-banana-pro` (about $0.15) as a second opinion. At most 3 attempts per screen, one change each. When it is right, put its path in the screen's `art` and in `references.anchor`. 6. **Look at the art (free).** Open it at full size and confirm there are no letters, numbers, signs, logos, game interface or watermarks, no extra people and nothing violent. Set the screen's `art_text_free` to true. Set its `crop` `x` and `y` (fractions of the frame, 0 to 1) on the hero's face; `zoom` 1.0 to 1.6. 7. **Caption (free).** Run `npm install sharp` in the Skill folder once, then `node scripts/compose-screens.mjs --plan screen.json --out-dir screens --guides`. It draws the name, the tagline, the label and the bar inside the safe box (the panel at most 460 px wide, the bar in one column with it), cuts the profile crop and writes the manifest. Fonts: Helvetica Neue Bold by default (DejaVu on Linux); pass `--font-file brand.ttf --font "Family Name"` for brand type. To change a word, a colour, the side or the bar, edit `screen.json` and run it again: no new art needed. 8. **Check (free).** `node scripts/check-screens.mjs --plan screen.json --manifest screens/manifest.json`. It fails a caption outside the 3:4 grid crop or in an app interface zone, a caption that does not match the plan, a name under 84 px, contrast under 4.5:1, a face outside the profile circle, a caption about crime or violence, a person without consent, art not marked text-free, the wrong size, or files out of date. Then open the screen, the guides and the profile crop at full size and go through `references/output-spec.md` by eye, reading every word letter by letter. Fix and recompose until it passes. 9. **A set (optional).** Plan 2 to 4 screens in step 2. Do screen 1 first, then `--stage draft` and `--stage final --screen rest`: each carries the photo, screen 1 as the look reference and its own draft. Compose and check them together. 10. **Deliver.** The screen or screens, the profile crops, the sheet for a set, `screen.json` and the art. Say which were re-rolled and anything still imperfect. 11. **Clean up photos.** Delete the user's photo from the working folders unless they want it kept. fal keeps request inputs and outputs on its side for a period: tell the user they can delete them from their fal dashboard, and to check fal's current retention policy. ## Limits - GPT Image 2 can refuse or soften a realistic photo of a person. Pets, products and invented characters are the reliable cases. If a person is refused, say so; do not try to get round the check. - Fine markings (a patch over one eye, a one-sided ear) can flip or drift. Name them in the features to keep and say which side, then compare with the photo. - The model follows "head at 13 percent" and "right half" loosely and the feet not at all, so the hero can land larger or further left than asked. The panel is at most 460 px wide (a 116 px name) so a hero in the right 45 percent of the frame stays clear; check the draft before the final. - A parked car can carry a plate with letter-like marks. The panel or the strip may cover it; if not, turn the car or re-roll. - The art is text-free, so a signboard shows blank. Lettering printed on a product is the only exception, and it must match the photo. ## Files - `references/screen-grammar.md`: the look of the genre in words, the layout, the caption panel and the loading bar, safe zones, the profile crop, captions, sets, and why no game is named. - `references/prompt-patterns.md`: prompt anatomy, writing the pose and scene, each kind of hero, people and consent, drafts and finals, failure modes and fixes. - `references/output-spec.md`: deliverables, sizes, costs, the pass/fail checklist. - `assets/prompt-template.md`: the screen prompt with placeholders. - `assets/generation-input.json`: model ids, draft and final sizes and qualities, approximate costs, the composition, the safety rule, the hero wording per kind, the look presets. - `assets/safe-zones.json`: the frame, the safe box, every zone with its source, the caption layout and the minimum sizes. - `assets/screen.example.json`: the sample plan (Patch, an invented terrier, `sunset-street`). - `assets/example-input.jpg`: the sample's photo of the invented terrier. - `scripts/new-screen.mjs`: writes screen.json and checks the captions. - `scripts/build-requests.mjs`: the estimate and the draft and final requests. Calls no API. - `scripts/fal-run.mjs`: sends one request with the user's FAL_KEY after `--approved`, waits, downloads (never overwrites without `--force`). - `scripts/compose-screens.mjs`: draws the caption, the label and the bar, cuts the profile crop, writes the manifest, the guides and the sheet (needs sharp). - `scripts/check-screens.mjs`: safe zones, exact words, size, contrast, consent and freshness checks.