# Art plan - THE LIVERAMP CRYPTS > **Production model: FLUX Schnell (`fal-ai/flux/schnell`) for boss / monster portraits.** 2 head-to-head tests confirmed FLUX wins on D&D rulebook character art: the "many eyes" and "four arms" details that nano-banana 2 normalizes away, FLUX renders. Caveat: FLUX literalizes weird prompts in uncanny ways (fused head+tablets in one test). When that happens, fall back to nano-banana 2 (`fal-ai/nano-banana-2`). For environment/background art with text or specific compositional rules, prefer nano-banana 2 by default. See [`../../ART-VALIDATION-WORKFLOW.md`](../../ART-VALIDATION-WORKFLOW.md). V1 ships with CSS-drawn placeholder portraits and a parchment-textured background, so the variant runs with zero API keys. This file is the recipe for swapping in real fantasy art when keys are available. ## Recommended API: fal.ai (FLUX-based) Picked fal.ai for: - **Quality**: hosts FLUX and SDXL pipelines that handle "fantasy painting, parchment, D&D style" prompts well without LoRA fiddling. - **Cost**: ~$0.01-0.05 per image on the standard endpoints. 25 images per full art pack is roughly $0.25-$1.25. - **Simple HTTP**: a single POST with prompt + image dims returns either an image URL or base64. No model hosting, no GPU rental. - **One vendor**: keeps the regen script and credentials surface area small. Alternatives considered: Replicate (similar pricing, more model sprawl), Recraft (excellent for illustration, no in-house fantasy LoRA reputation), Stability AI direct (cheaper per call but rougher API ergonomics). fal.ai wins on developer ergonomics for this scope. ## Style guide (applies to every prompt) Append this to every prompt as the trailing style block so the pack feels cohesive: > oil painting, dramatic chiaroscuro, parchment-aged palette, muted golds and deep browns, hand-painted D&D rulebook illustration, no text, no watermark, no border, centered subject Suggested negative prompt: > text, watermark, signature, ui, hud, blurry, modern clothing, photo, anime, low quality, deformed hands ## Asset list All images go in `assets/img/`. The placeholder pipeline reads these paths if present and falls back to CSS art if missing. Recommend 768x768 for portraits, 1536x768 for banners. ### Title banner (1 image) - **Path**: `assets/img/title-banner.png` - **Dims**: 1536x512 - **Prompt**: `A vast subterranean crypt entrance hewn from black stone, torchlight, seven descending arches numbered I through VII fading into darkness, a heavy iron-bound door at the threshold, the words "THE LIVERAMP CRYPTS" carved in serif Roman capitals on a stone lintel` ### Dungeon level backgrounds (7 images, one per week) - **Path**: `assets/img/level-{1..7}-bg.png` - **Dims**: 1536x768 - **Prompts**: - W1 `assets/img/level-1-bg.png`: `Candlelit underground temple lined with floating scrolls, a stone altar at the center inscribed with the words "PREDICT, DO NOT RECALL," wisps of golden text drifting in the air` - W2 `assets/img/level-2-bg.png`: `An endless library carved into rough stone, shelves climbing into shadow, books bound in cracked leather, scattered loose pages curling on the floor, a single librarian's lectern lit by green flame` - W3 `assets/img/level-3-bg.png`: `A dwarven foundry deep in the crypts, anvils stamping rune-engraved tablets reading "CLAUDE.md," sparks raining over a half-built cathedral of code` - W4 `assets/img/level-4-bg.png`: `A circular warding chamber, three glowing runes (one silver for private data, one purple for untrusted content, one crimson for external comms) burning at the points of a triangle on the floor, smoke curling toward an iron grate` - W5 `assets/img/level-5-bg.png`: `A guild hall hung with banners of different orders, several robed figures arguing over a great round table strewn with maps and tool icons, one figure pointing decisively at a single map` - W6 `assets/img/level-6-bg.png`: `A marble courtroom carved underground, twelve judges' chairs raised on a dais, a single scroll spread on the central podium, gavels of obsidian glinting in cold blue light` - W7 `assets/img/level-7-bg.png`: `A sunlit overworld vista seen through the cracked ceiling of the final crypt, banners with OKR sigils flapping in the wind, a path winding away into farmland and merchant towns` ### Monster portraits (9 reusable, 3x3 difficulty x type) All `assets/img/monster-{slug}.png`, 768x768. | slug | prompt | |----------------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------| | `goblin-scribe` | A small green goblin in tattered scholar robes, clutching a half-written scroll and a dripping quill, candlelit, mischievous expression | | `slime-archivist` | A translucent green slime wearing tiny spectacles, sitting atop a tower of leather-bound index cards, ink dripping from its surface | | `imp-apprentice` | A small red-skinned imp juggling three glowing scenario-orbs, candle-lit, eager and chaotic | | `kobold-loremaster` | A wiry kobold in a moth-eaten librarian's robe, thumbing through a crumbling glossary, dim green torchlight | | `owlbear-cataloger` | An owlbear with reading glasses, arranging four numbered relics on a stone slab, study-room background | | `shade-tinker` | A shadowy hooded figure mid-craft, mending a broken mechanical trap in midair, sparks of dim blue light | | `wraith-of-definitions` | A pale, half-transparent wraith drifting through stone walls, mouthing silent words, faint pale-blue aura | | `doppelganger` | A featureless humanoid wearing three overlapping faces, each face one of a set of plausible options, eerie candlelight | | `stone-golem` | A hulking stone golem with runic carvings, holding a giant slab inscribed with a scenario, dust falling from its joints | ### Boss portraits (7, one per week) All `assets/img/boss-{slug}.png`, 768x768. | slug | prompt | |-----------------------|-------------------------------------------------------------------------------------------------------------------------------------------------------| | `boss-omniseer` | A many-eyed celestial seer wreathed in scrolls, each eye showing a different next token, golden light, throne of vellum | | `boss-librarian` | A towering curator with a thousand scrolls braided into its robe, holding an empty thimble in one hand and a brimming chalice in the other | | `boss-architect` | A four-armed architect in mason's leather, lifting a layered "cake" of stone tablets each labelled with a CLAUDE.md tier, blueprint scattered around | | `boss-lich` | A skeletal lich with three runes burning on its chest (silver, purple, crimson), wreathed in black salt and cold light | | `boss-agent-king` | A crowned figure in heavy robes, surrounded by tool icons orbiting like satellites, one hand on a compass, the other on a leash trailing into mist | | `boss-judge` | A black-robed judge with a featureless white mask, gavel of obsidian raised, rubric scroll unrolling endlessly off the bench | | `boss-executive` | A pale executive in a tailored merchant's coat, holding a scroll of pilots in one hand and a balanced scale in the other, sunlit doorway behind | ## Integration The renderer assigns a CSS class `monster-` to the portrait div. To replace placeholder CSS with a real PNG, add this rule per slug to `assets/css/crypt.css`: ```css .monster-goblin-scribe { background: url("../img/monster-goblin-scribe.png") center / cover no-repeat, #2a1d0a; } .monster-goblin-scribe::before, .monster-goblin-scribe::after { display: none; } ``` (Or batch it: one wrapping selector that hides `::before` and `::after` plus per-slug `background-image`.) When all assets are present, the CSS-drawn placeholder hides itself. ## API keys needed | Env var | Service | Where to obtain | |------------------|---------|--------------------------------------------------| | `FAL_KEY` | fal.ai | https://fal.ai/dashboard - create a service key | Optional (only if you'd rather use a different provider): `REPLICATE_API_TOKEN`, `STABILITY_API_KEY`. Generation script would need a small switch; not implemented in v1. ## Sample fal.ai call (Node, illustrative) A regen script is not bundled in v1. Here's the call shape so it's ready to wire when needed: ```js // scripts/generate_art.mjs (NOT included; reference only) import fs from "node:fs/promises"; import path from "node:path"; const FAL_KEY = process.env.FAL_KEY; if (!FAL_KEY) { console.error("FAL_KEY required"); process.exit(1); } const STYLE = "oil painting, dramatic chiaroscuro, parchment-aged palette, muted golds and deep browns, hand-painted D&D rulebook illustration, no text, no watermark, no border, centered subject"; const NEGATIVE = "text, watermark, signature, ui, hud, blurry, modern clothing, photo, anime, low quality, deformed hands"; async function generate(prompt, outPath, w = 768, h = 768) { const res = await fetch("https://fal.run/fal-ai/flux/schnell", { method: "POST", headers: { "Authorization": `Key ${FAL_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ prompt: `${prompt}. Style: ${STYLE}.`, negative_prompt: NEGATIVE, image_size: { width: w, height: h }, num_inference_steps: 4, }), }); if (!res.ok) throw new Error(`fal.ai ${res.status}: ${await res.text()}`); const data = await res.json(); const imageUrl = data.images?.[0]?.url; if (!imageUrl) throw new Error("no image url in response"); const img = await fetch(imageUrl); const buf = Buffer.from(await img.arrayBuffer()); await fs.mkdir(path.dirname(outPath), { recursive: true }); await fs.writeFile(outPath, buf); console.log(`wrote ${outPath}`); } // Then loop the asset list above and call generate() for each. ``` ## Cost estimate 25 images x $0.05 worst case = $1.25 for a full pack. Re-rolls are cheap; budget $5 to iterate on prompts.