🎨 Art style lab

Style direction for the book, iterated live. The test subject is fixed — chiva at a colonial plaza, two munchos aboard, the pig running — so every generation below differs only in style. Model: bytedance/seedream/v5/pro.

New test

Results (1)

Prompt

BASELINE TEST: hand-painted children's book watercolor, loose ink outlines, muted earthy palette, naive flat characters with dot eyes, visible paper texture

Reference — style samples from the source book

Reference: rich landscape spread, whimsical hidden details

Reference: quiet station scene, warm greens, hand-painted signage

Reference: giant-object architecture (mushroom station), character lineup

Full prompt blocks (used by the book pipeline)

Style block

ART STYLE (crucial): a hand-painted children's book watercolor illustration in the spirit of Sven Nordqvist's picture books. Loose, sketchy pen-and-ink outlines; translucent watercolor washes that bleed slightly beyond the lines; clearly visible paper texture and pigment granulation; muted, earthy palette of moss green, ochre, warm wood brown, brick red and faded sky blue; soft natural daylight. Detail is SUGGESTED, not rendered: leave areas loose and unfinished, let the paper show through, keep generous calm space in the sky. CHARACTERS are drawn in a deliberately naive, stylized way — a few confident ink strokes, rounded bean-shaped bodies, dot eyes, flat color fills, minimal facial features; no detailed anatomy, no realistic fur or skin texture. Cozy, whimsical, handmade and slightly imperfect. Absolutely NO smooth digital airbrush look, NO glossy 3D-cartoon rendering, NO saturated vector-flat colors, NO hyper-detailed realism — if it looks like a polished Pixar-style digital painting, it is wrong.

Chiva block

SETTING & VEHICLE (Colombian adaptation): Replace the little steam train with a traditional Colombian "chiva" — a rustic, open-sided wooden bus on a truck chassis, every plank hand-painted in bold red, yellow and blue with decorative folk-art scrollwork and geometric borders, a ladder to the roof rack loaded with baskets of fruit, chickens, and rolled hammocks. The driver wears a sombrero vueltiao (black-and-white woven hat) and a light shirt. The two main characters, the munchos Menos and Más, are small capybara pups (chigüiros) with round bean-shaped brown bodies, dot eyes and tiny rounded ears, wearing mini ruana ponchos (one brown, one beige) and tiny sombreros vueltiao. The other passengers are small COLOMBIAN ANIMALS: armadillos, iguanas, howler and titi monkeys, toucans, frogs, a spectacled-bear cub — all drawn as simple naive storybook figures with dot eyes.

LANDSCAPE: Colombian coffee-region countryside with colonial architecture — white-washed walls, terracotta roofs, brightly painted wooden balconies and doors (Barichara / Villa de Leyva / Salento style), steep green Andean hills, wax palm silhouettes, coffee bushes with red berries, plantain and banana leaves, bougainvillea and heliconia flowers, misty mountains in the far background, dirt roads instead of rails.

KEEP FROM THE ORIGINAL: one hidden arithmetic equation painted into the scenery of every spread (on a rock, a wall, a banner, a cow's side, hanging laundry, a sail); leave a calm area of sky in the top-left for narration text; the same warm watercolor storybook style; the same gentle humor and hidden details.

Workflow notes

Flux Pro via fal.ai — workflow:
1. STYLE BLOCK + CHIVA BLOCK + the frame's scene description = the prompt.
2. Frame 1 is generated text-to-image (fal-ai/flux-pro/v1.1). Every next frame is generated with the PREVIOUS frame as reference (fal-ai/flux-pro/kontext) so characters, the chiva and the style stay consistent.
3. Cast members entering or leaving are listed per frame; their look strings are injected into the prompt verbatim.
4. Aspect ratio: wide panorama (the originals are double-page spreads ≈ 2:1).
5. Text-in-image (the painted sums) is unreliable — check each result; regenerate or hand-letter in post if garbled.