Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.
JavaScript 3d-generationai-agentsanimationavatardigital-humangenerative-aiglbgltfimage-to-3dllmmcpmcp-servermodel-context-protocolsolanatext-to-3dthreejsweb-componentswebglwebxrx402
// Forge prompt generator: truly random prompts that reliably forge well.
//
// Instead of sampling a fixed list, "Surprise me" and the example chips draw
// from curated slot grammars: each family combines a subject pool with only
// the materials, finishes, and bases that suit it, so every one of the tens
// of thousands of possible prompts stays inside what the TRELLIS lane
// reconstructs beautifully (single isolated subject, chunky silhouette, a
// named material, no thin wires, no glass-only builds, no multi-object
// scenes). The family design mirrors the prompt coach in
// forge-prompt-studio.js: every generated prompt carries a material word and
// never trips the multi-subject heuristics.
//
// Pure module, no DOM: consumed by the Forge page UI and unit-tested
// directly (tests/forge-prompt-gen.test.js). `rng` is injectable so tests
// can seed it; the UI passes nothing and gets Math.random.
const pick = (arr, rng) => arr[Math.floor(rng() * arr.length)];
// Article-aware join for slots that can start with a vowel ("an amethyst
// geode"). Good enough for this vocabulary; no silent-h words in the pools.
const art = (phrase) => `${/^[aeiou]/.test(phrase) ? 'an' : 'a'} ${phrase}`;
// ── Family: statues & sculpture ─────────────────────────────────────────
// Sculpture is the single strongest TRELLIS category: solid silhouette,
// and patina/roughness reads as intentional surface detail.
const STATUE_SUBJECTS = [
'an astronaut',
'a ballerina mid-twirl',
'a seated lion',
'a coiled dragon',
'a meditating monk',
'a chess knight',
'a perched owl',
'a rearing horse',
'a leaping koi fish',
'a curled-up cat',
'a standing bear',
'a twisting abstract ribbon',
];
const STATUE_MATERIALS = [
['weathered bronze', 'green patina'],
['carved white marble', 'soft grey veining'],
['polished black obsidian', 'glassy sheen'],
['terracotta', 'warm matte finish'],
['carved jade', 'polished deep green'],
['sandstone', 'wind-worn texture'],
['cast concrete', 'brutalist finish'],
['gilded bronze', 'warm gold sheen'],
['carved walnut', 'oiled wood grain'],
['celadon ceramic', 'crackle glaze'],
];
const STATUE_BASES = [
'on a square stone plinth',
'on a round marble base',
'on a black granite pedestal',
'museum piece, soft studio light',
];
// ── Family: creature figurines ──────────────────────────────────────────
// Style pairs a material adjective with the finish cue that sells it.
const FIGURINE_STYLES = [
['low-poly', 'faceted matte surface'],
['glazed porcelain', 'soft studio light'],
['plush felt', 'stitched seams'],
['brass clockwork', 'polished gears'],
['carved wood', 'hand-painted details'],
['hand-sculpted clay', 'chunky rounded shapes'],
['knitted wool', 'chunky yarn texture'],
];
const FIGURINE_ANIMALS = [
'red fox',
'axolotl',
'hedgehog',
'barn owl',
'octopus',
'humpback whale',
'sea turtle',
'bear cub',
'tree frog',
'penguin',
'highland cow',
'dachshund',
'chameleon',
'baby elephant',
];
// ── Family: potted plants ───────────────────────────────────────────────
const PLANTS = [
'monstera',
'fiddle-leaf fig',
'jade succulent',
'aloe vera',
'trailing pothos',
'bonsai juniper',
'barrel cactus',
'lavender bush',
'boston fern',
'flowering orchid',
];
const POTS = [
'a terracotta pot',
'a glazed cobalt ceramic pot',
'a ribbed concrete planter',
'a woven rattan basket',
'a hammered copper pot',
'a matte black ceramic planter',
'a speckled stoneware pot',
];
// ── Family: crystals & geodes ───────────────────────────────────────────
const MINERALS = [
'amethyst',
'rose quartz',
'emerald fluorite',
'golden citrine',
'deep blue azurite',
'smoky quartz',
'turquoise',
'malachite',
];
// ── Family: food & bakery ───────────────────────────────────────────────
const FOODS = [
['a glazed chocolate donut with rainbow sprinkles', 'glossy icing'],
['a layered strawberry cake on a ceramic stand', 'smooth buttercream'],
['a rustic sourdough loaf', 'floury matte crust'],
['a ripe pomegranate split open', 'glossy ruby seeds'],
['a stack of pastel macarons', 'glossy smooth shells'],
['a honey jar with a wooden dipper', 'warm golden light'],
['a steaming bowl of ramen', 'glossy ceramic bowl'],
['a watermelon with a cut wedge', 'glossy rind, juicy red flesh'],
];
// ── Family: crafted props ───────────────────────────────────────────────
// Each builder owns its object's compatible variants, so a teapot never
// comes out in wrought iron nor a lantern in buttercream.
const PROP_BUILDERS = [
(rng) =>
`a glazed ceramic teapot, ${pick(['cobalt blue', 'oxblood red', 'matte sage green', 'speckled cream'], rng)}, studio lighting`,
(rng) =>
`a ${pick(['brass', 'copper', 'wrought iron'], rng)} storm lantern, weathered patina, warm glow`,
(rng) => `a treasure chest, ${pick(['dark oak', 'weathered cedar', 'ebony wood'], rng)}, tarnished iron bands`,
(rng) =>
`a ${pick(['worn cognac leather', 'mustard velvet', 'deep green velvet', 'charcoal wool'], rng)} armchair, studio lighting`,
(rng) =>
`a knight helm, ${pick(['dark polished steel', 'brass-trimmed steel', 'blackened iron'], rng)}, plumed crest`,
(rng) =>
`a tall ceramic vase, ${pick(['crackle-glazed turquoise', 'matte terracotta', 'cobalt-striped white'], rng)}`,
(rng) =>
`a retro radio, ${pick(['cream bakelite', 'mint green enamel', 'walnut with brass dials'], rng)}`,
(rng) =>
`a vintage film camera, brushed aluminium, ${pick(['black', 'tan', 'burgundy'], rng)} leather grip`,
(rng) =>
`a wizard staff, gnarled ${pick(['oak', 'blackthorn', 'driftwood'], rng)}, embedded ${pick(['amber', 'sapphire', 'emerald'], rng)} gem`,
(rng) =>
`a stack of hardcover books, ${pick(['aged cloth binding', 'embossed leather covers', 'linen spines, gilt titles'], rng)}`,
];
// ── Family: cozy miniatures ─────────────────────────────────────────────
const MINI_BUILDERS = [
(rng) =>
`a cozy mushroom cottage, thatched roof, ${pick(['moss-covered stone base', 'round oak door', 'stone chimney'], rng)}`,
(rng) => `a small lighthouse, ${pick(['red', 'teal', 'navy'], rng)}-striped enamel paint, rocky stone base`,
(rng) =>
`a low-poly camper van, ${pick(['pastel teal', 'butter yellow', 'salmon pink'], rng)} paint, chrome bumper`,
(rng) =>
`a hot air balloon, ${pick(['striped red-cream', 'patchwork rainbow', 'teal-gold'], rng)} canvas, wicker basket`,
(rng) => `a dutch windmill, ${pick(['red brick', 'whitewashed stone', 'dark timber wood'], rng)} tower, canvas sails`,
(rng) => `a tiny houseboat, painted wooden hull, ${pick(['sage green', 'ochre', 'sky blue'], rng)} rounded cabin`,
];
// One builder per family. Exported so tests can exercise each family
// directly; generateForgePrompt samples them uniformly.
export const FAMILIES = [
function statue(rng) {
const [adj, detail] = pick(STATUE_MATERIALS, rng);
return `a ${adj} statue of ${pick(STATUE_SUBJECTS, rng)}, ${detail}, ${pick(STATUE_BASES, rng)}`;
},
function figurine(rng) {
const [adj, finish] = pick(FIGURINE_STYLES, rng);
return `a ${adj} ${pick(FIGURINE_ANIMALS, rng)} figurine, ${finish}`;
},
function plant(rng) {
return `a potted ${pick(PLANTS, rng)} in ${pick(POTS, rng)}, soft studio light`;
},
function crystal(rng) {
const mineral = pick(MINERALS, rng);
return rng() < 0.5
? `a cluster of ${mineral} crystals on raw grey rock, glossy facets`
: `${art(mineral)} geode, sparkling crystal core, soft studio light`;
},
function food(rng) {
const [subject, finish] = pick(FOODS, rng);
return `${subject}, ${finish}`;
},
function prop(rng) {
return pick(PROP_BUILDERS, rng)(rng);
},
function mini(rng) {
return pick(MINI_BUILDERS, rng)(rng);
},
];
// Family draw weights, proportional to each grammar's combinatorial size:
// uniform selection would over-sample the small pools (food, crystals) and
// make repeats feel common. Order matches FAMILIES.
const FAMILY_WEIGHTS = [5, 3, 2, 1, 1, 2, 1];
const WEIGHTED_FAMILY_INDEX = FAMILY_WEIGHTS.flatMap((w, i) => Array(w).fill(i));
/** One random prompt, guaranteed to be a single isolated subject with a
* material cue (the shape the Forge lane meshes cleanest). */
export function generateForgePrompt(rng = Math.random) {
return FAMILIES[pick(WEIGHTED_FAMILY_INDEX, rng)](rng);
}
/** `count` distinct prompts, none of which appear in `avoid`, used so
* "Surprise me" and "More ideas" never echo what is already on screen.
* `maxLength` filters to prompts that fit compact UI (example chips). */
export function generateDistinctForgePrompts(count, avoid = new Set(), rng = Math.random, maxLength = Infinity) {
const out = [];
const seen = new Set(avoid);
// The combinatorial space is ~10^4, so collisions are rare; the attempt
// cap only guards against a caller asking for more than the space holds.
for (let attempts = 0; out.length < count && attempts < count * 40; attempts++) {
const prompt = generateForgePrompt(rng);
if (prompt.length > maxLength || seen.has(prompt)) continue;
seen.add(prompt);
out.push(prompt);
}
return out;
}
/**
* A replacement set for a row of example chips that cannot make the row taller.
*
* The chip rows on /forge are flex-wrap containers, so what moves the page is
* the row's TOTAL width, not any single label: a rotation whose labels are
* wider overall wraps onto a new line and pushes the rest of the page down. A
* per-label cap alone does not prevent that (a 64-character cap over labels
* averaging half that reliably added a row), and it measured as the largest
* single layout shift on /forge: 0.19 of a 0.256 CLS against production on
* 2026-08-15.
*
* `currentLabels` are the labels being replaced. The returned prompts are
* distinct, absent from `avoid`, each no longer than the longest current label,
* and no longer in total than the current labels combined, so the row's wrap
* geometry can only stay the same or shrink. When the budget is too tight to
* fill the row with distinct prompts the shortest candidates are returned
* instead, so the row still rotates rather than freezing.
*/
export function generateForgeChipSet(
currentLabels,
avoid = new Set(),
rng = Math.random,
maxLength = Infinity
) {
const lengths = currentLabels.map((label) => String(label).trim().length);
const count = lengths.length;
if (!count) return [];
const budget = lengths.reduce((a, b) => a + b, 0);
const longest = Math.min(Math.max(...lengths), maxLength);
// Over-generate so there is a real pool to fit inside the budget. The
// combinatorial space is ~10^4, so this stays cheap. A row whose labels are
// all shorter than any prompt the generator can emit filters the pool to
// nothing; fall back to the caller's own cap there so the row still rotates
// rather than freezing on its static copy.
let pool = generateDistinctForgePrompts(count * 6, avoid, rng, longest);
if (!pool.length) pool = generateDistinctForgePrompts(count * 6, avoid, rng, maxLength);
if (!pool.length) return [];
// Walk the pool in its generated (random) order so the row keeps its
// variety, but only accept a prompt while the shortest remaining candidates
// could still fill the leftover slots inside the budget. Without that
// lookahead a couple of wide early picks strand the tail and the row comes
// back short.
const shortest = [...pool].sort((a, b) => a.length - b.length);
const minLength = shortest[0].length;
const picked = [];
let used = 0;
for (const prompt of pool) {
if (picked.length >= count) break;
const slotsAfter = count - picked.length - 1;
if (used + prompt.length + slotsAfter * minLength > budget) continue;
picked.push(prompt);
used += prompt.length;
}
// Budget too tight even for the shortest full row (labels narrower than any
// prompt): top up with the shortest candidates. Fewer characters than the
// labels they replace can only remove a wrap, never add one, and the caller
// pins the row's height across the swap so that cannot shift the page.
for (const prompt of shortest) {
if (picked.length >= count) break;
if (!picked.includes(prompt)) picked.push(prompt);
}
return picked;
}