{ "model": "openai/gpt-image-2", "model_with_references": "openai/gpt-image-2/edit", "meleon_model": "gpt-image-2", "input": { "num_images": 1, "output_format": "png" }, "stages": { "draft": { "quality": "low", "image_size": { "width": 720, "height": 1280 }, "why": "a cheap rough screen to agree the pose, the scene and the open space for the caption; 720 x 1280 is the smallest 9:16 size fal accepts (at least 655,360 pixels)" }, "final": { "quality": "medium", "image_size": { "width": 1088, "height": 1936 }, "why": "the art for the finished screen; composed down to 1080 x 1920, so nothing is upscaled" } }, "alternative_final": { "model": "fal-ai/nano-banana-pro", "model_with_references": "fal-ai/nano-banana-pro/edit", "meleon_model": "", "input": { "aspect_ratio": "9:16", "resolution": "2K", "num_images": 1, "output_format": "png" }, "approx_cost_usd": 0.15, "why": "flat price per image (fal list price, read 2026-11-13). A second opinion when GPT Image 2 drifts from the pet's markings or the person's face; ask for it with build-requests.mjs --model nano-banana-pro. The size is whatever 9:16 at 2K gives; the composer scales it to 1080 x 1920." }, "max_references": 3, "approx_cost": { "per_1024_square": { "low": 0.006, "medium": 0.05, "high": 0.2 }, "per_reference_image": 0.01, "note": "GPT Image 2 is token-billed (image output $30 per 1M tokens, image input $8 per 1M, text $5 in and $10 out per 1M; fal model page read 2026-11-13). The scripts scale fal's per-1024-square figures by pixel count, so a 720 x 1280 draft at low is about $0.005 and a 1088 x 1936 final at medium about $0.10, plus about $0.01 per reference image. fal returns no token count with the image; the fal dashboard shows the real charge." }, "composition": { "left": "Composition: the hero is seen full length, small in the frame, about 40 percent of the image height: the top of its head at about 13 percent of the image height and its feet at about 55 percent. The whole of the hero's body, ears and tail stays in the right-hand part of the frame, between about 60 and 94 percent of the width, and nothing of it crosses into the left half. The whole left half of the frame, from about 25 to 70 percent of the height, is calm open street, kerb and pavement with no hero, no cart and no props in it, because a caption panel goes there later. The busy city street (a parked car, palms, buildings, the cart and the skateboard) fills the rest of the frame, behind and beside the hero, with the busiest props on the right and in the distance.", "right": "Composition: the hero is seen full length, small in the frame, about 40 percent of the image height: the top of its head at about 13 percent of the image height and its feet at about 55 percent. The whole of the hero's body, ears and tail stays in the left-hand part of the frame, between about 6 and 40 percent of the width, and nothing of it crosses into the right half. The whole right half of the frame, from about 25 to 70 percent of the height, is calm open street, kerb and pavement with no hero, no cart and no props in it, because a caption panel goes there later. The busy city street (a parked car, palms, buildings, the cart and the skateboard) fills the rest of the frame, behind and beside the hero, with the busiest props on the left and in the distance." }, "grid_rule": "Keep everything important between 13 and 87 percent of the height, because the profile grid shows only that middle part, and keep the right-hand 14 percent of the frame free of important detail.", "safety_rule": "Mood: cocky, friendly and funny. No weapons, no violence, no crime, no police, no smoking, no alcohol or drugs, no piles of money, no blood. No game interface of any kind: no health bar, map, minimap, stars, icons, buttons or heads-up display. Do not copy any existing game, film, brand, logo or font. The city is invented and generic: no real landmarks, no brand names, no number plates, every sign and shop front blank.", "kinds": { "pet": "The hero is {{description}}, the pet in reference image {{ref}}. It must stay clearly recognisable as that animal: {{keep}}. Same coat colours and markings in the same places, same ear shape and eye colour; only the art style changes.", "product": "The hero is {{description}}, the product in reference image {{ref}}. Keep its exact shape, proportions, colours and materials ({{keep}}), and any lettering printed on it exactly as in the photo. Add no other words, and do not give it a face or limbs.", "character": "The hero is {{description}}, an invented character. Keep this design: {{keep}}.", "character_with_reference": "The hero is {{description}}, the invented character in reference image {{ref}}. Keep its design exactly: {{keep}}.", "person": "The hero is the person in reference image {{ref}} ({{description}}). They must stay exactly themselves: the same identity, age range, skin tone, hair, facial structure and body, drawn in the comic art style so they are clearly recognisable. Do not change their ethnicity, add or remove features, make them younger or older, or give them an expression or action that would embarrass them." }, "presets": { "sunset-street": { "use_for": "the default: a hot pink and orange sunset over a busy sunny city street", "look": "a flat-shaded painterly comic-book illustration with heavy thick black outlines, bold simplified shapes, visible brush texture and halftone shading, hot saturated sun-baked colours (a hot pink and orange sunset sky, warm golden sunlight, turquoise and purple shadows), strong low-sun three-quarter hero lighting with a thin bright rim light, a hand-painted finish", "world": "a busy sunny city street with palm trees, low-rise buildings, bright striped awnings, parked cars and a street vendor cart", "style": { "panel": "#ffd23f", "ink": "#1a0f1f", "accent": "#ff3d7f" } }, "midday-boardwalk": { "use_for": "a bright beach-town boardwalk at noon", "look": "a flat-shaded painterly comic-book illustration with heavy thick black outlines, bold simplified shapes, visible brush texture and halftone shading, hot saturated sun-baked colours (a vivid turquoise sky with a few puffy clouds, white-hot sunlight, hot coral, teal and sunshine yellow), short hard shadows, a hand-painted finish", "world": "a busy sunny boardwalk with palm trees, pastel low-rise buildings, bright umbrellas, a beach behind, bicycles and an ice-cream cart", "style": { "panel": "#fff1b8", "ink": "#0b2a3a", "accent": "#ff6b3d" } }, "dusk-neon": { "use_for": "an indigo and magenta dusk city with warm window light", "look": "a flat-shaded painterly comic-book illustration with heavy thick black outlines, bold simplified shapes, visible brush texture and halftone shading, saturated dusk colours (an indigo and magenta sky, warm orange window glow, cyan and pink light spill), strong three-quarter hero lighting with a thin bright rim light, a hand-painted finish", "world": "a busy city street at dusk with palm trees, low-rise buildings with glowing windows, parked cars, scooters and a street vendor cart, glowing shapes in place of any writing", "style": { "panel": "#ffe66d", "ink": "#1b1033", "accent": "#19d3da" } } }, "cost_note": "One screen is about $0.02 for a low draft and about $0.12 for a medium final (the photo and the draft as references), about $0.14 in all, about $0.26 with one re-roll of each. A set of 4 is about $0.07 for drafts and about $0.52 for finals. build-requests.mjs --stage estimate prints the total for the real plan; quote it and get one yes before the first paid call. The caption, the loading bar, the profile crop and the checks are free and local." }