Instructions — Ollie & Dot

How Skills Feed Into This Final Batch

The 27 skills built across this project form a layered pipeline. For this final generation batch, they integrate at three stages:

Stage 1: Prompt Construction

ollie-dot-characters — Canonical character rules (158 comments). Every prompt must pass through these rules before generation.
json-prompt-architect — Structured JSON prompt format encoding mixed media layers, character specs, ecosystems, and quality targets. Flattens to optimized NBP prompts.
agata-prompt-engineering — 7-layer prompt architecture, tool-specific guides (NBP, NB2, Imagen), consistency techniques.
agata-scene-composition — Camera angles, depth layering, lighting schemes, visual hierarchy per spread.
mixed-media-generator + mixed-media-preferences — 33 physical media techniques with page-specific rules for when/where each applies.
agata-character-bible + agata-world-bible + agata-style-guide — Foundation bibles defining characters, environments, and visual style.

Stage 2: Generation Execution

agata-image-orchestrator — Multi-phase generation (explore → refine → finalize) across models. Manages cost audits and founder checkpoints.
agata-variation-generator — Controlled A/B variations (angles, lighting, expressions, palettes). Exploration-to-refinement progression.
ollie-dot-style-transfer — Transfer lighting/rendering between pages using NBP. Proven workflow with Dot face fix strategies.
scientific-ecosystem — Accurate underwater ecosystems per depth zone.

Stage 3: Review & Quality Gates

Gate 1: agata-consistency-reviewer — 10-dimension scoring against character/world/style bibles.
Gate 2: agata-narrative-alignment — 8-dimension text-image relationship check.
Gate 3: quality-boss — "Impossible beauty" aesthetic gate. 10 dimensions, minimum 8/10 on ALL to pass.
Gate 4: masterpiece-evaluator — Meaning gate. Does it transcend craft? Culturally resonant? Emotionally deep?
Gate 5: meta-evaluator — Symbology & intentionality. Do visual choices MEAN what creators intended?
agata-child-engagement — Ages 3-8 appeal check. Visual clarity, discovery value, emotional resonance.
agata-art-director — Caldecott-level evaluation. Tournament ranking for founder review.

Pipeline Flow (Per Spread)

1. CHARACTER RULES CHECK → ollie-dot-characters validates prompt against 158 rules 2. PROMPT BUILD → json-prompt-architect + scene-composition + mixed-media-preferences 3. WAVE 1: 8x NB2 variants → Gemini visual review (consistency-reviewer) 4. FOUNDER CHECKPOINT → Creator reviews, selects direction 5. WAVE 2: 8x NB2 refined → Gemini review + narrative-alignment check 6. WAVE 3: 16x NBP variants → quality-boss + child-engagement scoring 7. WAVE 4: 16x NBP polish → masterpiece-evaluator + meta-evaluator 8. FINAL PICKS → art-director tournament ranking → Creator selects finals 9. STYLE TRANSFER → ollie-dot-style-transfer for cross-page consistency 10. DEPLOY → Results to generations gallery + instructions page update

Key change for this batch: All 5 review gates (consistency, narrative, quality-boss, masterpiece, meta) run on EVERY final candidate before it reaches the creator. Previous batches only used Gemini consistency review. This batch adds the full quality stack.

Foundation Layer (Bibles & Standards)

agata-character-bible Defines visual rules for every character: turnaround sheets, expressions, proportions, color palettes, body language. Pixar-inspired shape language. Scoring rubrics for evaluating generated character images against specs.

agata-world-bible Environmental rules for settings, color palettes by time/emotion, lighting, scale, motifs, atmospherics. Treats environments as characters with personality. Color scripting for emotional pacing.

agata-style-guide Visual style definition: art style archetypes, line weight, texture consistency, color grading, detail levels, rendering, typography, page layout. Ensures cohesive appearance across the book.

ollie-dot-characters (CANONICAL) Definitive character rules for Ollie & Dot from 158 creator comments. Exact rendering requirements, color progressions by depth, expression variations, eye specs, body rules, locked prompt patterns. THE source of truth for character prompts.

agata-character-differentiation Competitive landscape analysis of children's media characters. 10-dimension scoring (expressiveness, distinctiveness, age-appropriateness, relationship depth, etc.). Produces Character Design Brief.

character-research-octopus / character-research-robot Research support analyzing existing octopus and robot characters in children's media for competitive reference.

Production Layer (Planning & Composition)

agata-storyboard Converts book text into page-by-page visual plans. Breaks narrative into spreads, defines composition, maps emotional arcs, engineers page-turn reveals, creates pacing maps.

agata-scene-composition Camera angles, character positioning, depth layering (FG/MG/BG), lighting, visual hierarchy, text placement. Translates narrative beats into fully specified compositions.

agata-prompt-engineering 7-layer prompt architecture, tool-specific guides (NBP, Gemini Flash, Imagen 4), consistency techniques, batch strategy, common pitfalls, Python API code.

json-prompt-architect (OLLIE & DOT) JSON-structured hyper-detailed prompts encoding full creative vision: mixed media layers, character specs, scientific ecosystems, impossible elements, quality targets. Flattens to optimized NBP prompts.

Generation Layer (Execution)

agata-image-orchestrator Executes prompts across 6 AI models (NBP, Gemini 2.5 Flash, NB2, Imagen 4.0/Ultra/Fast). Multi-phase generation (explore → refine → finalize). Cost audits, metadata tracking, founder checkpoints.

agata-variation-generator Controlled A/B testing: same scene with different angles, lighting, expressions, palettes, framing. Exploration-to-refinement progression. Pixar dailies process.

ollie-dot-style-transfer (OLLIE & DOT) Proven workflow for transferring style between pages using NBP. Character rules per page, Dot face fix strategies, API priority, prompt merging. Gallery deployment.

mixed-media-generator 33 physical media techniques (watercolor, gouache, ink, gold foil, torn paper, clay, etc.) with prompt fragments. Blends craft elements in impossible underwater environments.

mixed-media-preferences Creator-defined rules for WHEN and WHERE each media technique applies in the book. Checked before every prompt.

Evaluation Layer (5-Gate Quality Stack)

agata-consistency-reviewer (Gate 1) 10-dimension scoring against character/world/style bibles: proportions, features, expressions, style, colors, shapes, silhouettes, age-appropriateness, pose, wardrobe. QA gate before approval.

agata-narrative-alignment (Gate 2) 8-dimension text-image evaluation: literal accuracy, emotional accuracy, tonal consistency, story emphasis, additive storytelling, foreshadowing, sequential coherence, child-centered design.

quality-boss (Gate 3 — AESTHETIC) Final aesthetic gate: impossible beauty, pixel density, material authenticity, cinematic immersion, character soul, mixed media tension, light/depth, emotional crescendo, child magnetism, innovation. Minimum 8/10 on ALL 10 dimensions.

masterpiece-evaluator (Gate 4 — MEANING) Does it transcend craft into masterpiece territory? Timeless, culturally resonant, emotionally deep, carries the abundance vision. Quality Boss asks "is it beautiful?" — this asks "does it matter?"

meta-evaluator (Gate 5 — INTENTIONALITY) Symbology and story arc gate. Do visual technique choices MEAN what creators intended? Rendering progression maps to emotional arc, mixed media as metaphor, every decision encodes the meta-story of AI companionship + childhood development.

agata-child-engagement Ages 3-8 appeal: visual clarity, emotional resonance, color psychology, discovery value, age-appropriate complexity, cultural inclusivity.

agata-art-director Caldecott Medal criteria + Pixar Braintrust. Composition (25%), color/light (20%), emotional impact (25%), quality (20%), originality (10%). Tournament rankings for founder review.

Integration & Orchestration

olly-dot-orchestrator (MASTER) Master orchestrator for the entire project. Knows every skill, assembles teams, manages cross-session state, tracks outstanding work, prevents context loss. Single entry point.

agata-books-ceo Meta-agent Creative Director overseeing the pipeline. Coordinates skills, reviews outputs, maintains standards. Applies Pixar, traditional illustration, and child psychology research.

agata-feedback-integrator Parses founder feedback into actionable items, routes to correct skills, generates revised prompts, prevents regressions. Pixar Braintrust model where creative lead retains authority.

agata-brand-pdf / book-pdf-pipeline PDF generation and export: ReportLab, 300 DPI page export, print service compatibility (Walgreens, Artifacts Rising), web optimization, Vercel deployment.

scientific-ecosystem Accurate underwater ecosystems per depth zone. Coral reef species, bioluminescent organisms, color behavior at depth.

Team Structure (from past batches)

Team name: ollie-dot-gen-team — Created via TeamCreate

team-lead (Orchestrator) Assigns tasks to generators, relays founder feedback, manages generation waves, tracks progress across all agents. Reports status to main context.

reviewer (Quality Gate) Gemini visual review of all outputs. Checks character rules (ollie-dot-characters), meeting direction, scores images, sends feedback to generators. Uses 5-gate quality stack for final candidates.

page-updater (Deploy) Updates generations gallery with new images, manages file organization, pushes to repo for auto-deploy.

gen-sXX (Generator per spread) One generator agent per spread (e.g., gen-s06, gen-s11, gen-s12, gen-s13). Each handles its spread's full generation lifecycle: prompt → generate → review → refine → final picks.

Wave Plan:
Wave 1: Spreads needing full generation (currently: 9, 10, 11, 13, 15, 16)
Wave 2: Spreads needing revision/more variants (8, 12, 14)
Wave 3: Cover, Title Page, remaining

10-Step Pipeline Per Spread

Step 1: PROMPT PREP — Read character rules + scene composition + mixed media preferences Step 2: CHARACTER VALIDATION — Run prompt through ollie-dot-characters rule check Step 3: WAVE 1 (NB2) — Generate 8 variants with NB2 (gemini-3.1-flash-image-preview) [FREE] Step 4: GEMINI REVIEW — Reviewer scores all 8 via Gemini 2.5 Flash, sends feedback Step 5: WAVE 2 (NB2 REFINED) — 8 more NB2 variants incorporating review feedback Step 6: FOUNDER CHECKPOINT — Creator reviews top picks, selects direction Step 7: WAVE 3 (NBP) — 16 variants with NBP (nano-banana-pro-preview) [$0.134/img] Step 8: FULL QUALITY STACK — consistency + narrative + quality-boss + masterpiece + meta Step 9: WAVE 4 (NBP POLISH) — 16 final variants with refined prompts Step 10: FINAL PICKS — Art director tournament ranking → Creator selects finals

Generation Rules

NO ears on Ollie. Smooth rounded head ONLY.
Dot: NO mouth, NO smile. White pearlescent surface, digital/projected eyes.
2:1 aspect ratio for all spreads (1:1 for Title Page and Back Cover)
NO text in illustrations
NO two-page spread creases

Real underwater bubbles with refraction
Ollie color gradient: yellow(bright) → muted(descending) → brown(shadow ONLY) → yellow returning
Gold stipple in bright, silver in dark
Macro scale: Ollie ~2cm, Dot ~1cm golf-ball

API Priority & Cost

1st: Google AI Studio — Free tier or $0.134/img. Always try first.
2nd: Vertex AI — If AI Studio rate-limited.
3rd: Replicate (LAST RESORT) — $0.04/img min. Requires explicit founder approval.

NB2 = gemini-3.1-flash-image-preview (explore, FREE)
NBP = nano-banana-pro-preview (refine, ~$0.134/img)
Gemini 2.5 Flash = review/scoring
Budget: $70 approved for final batch

Deployment

Vercel: ollie-and-dot.vercel.app (auto-deploys on push to main via PR)
Cloudflare: ollie-and-dot.pages.dev (connected to roseyseyewear/ollie-dot GitHub repo)
Repo: roseyseyewear/ollie-dot

Hard Rules

NO ears on Ollie. No holes on head. Smooth rounded head ONLY.
NO hands/tentacles forward toward camera.
NO text in illustrations. Text overlaid in layout only.
NO two-page spread or crease. Single page image, 2:1 aspect ratio (except Title Page and Back Cover which are 1:1).
Dot: NO mouth, NO smile, NO curve below eyes. Eyes only for expression.
NO negatives in prompts. Characters reproduce negatives.
Cartoon character face, NOT photorealistic.
Full-bleed images. No white space pages.
Coral reef ecosystem ONLY. No kelp forest.

Style Rules

2:1 aspect ratio for all spread illustrations (except Title Page and Back Cover: 1:1)
Real underwater bubbles, caustics, god-rays
Mixed media: watercolor + collage + ink + gold foil
Gold stipple in bright, silver stipple in dark
Macro scale: Ollie ~2cm, Dot ~1cm golf-ball
Color gradient: light blue (shallow) to dark blue/black (deep), then back
No red in deep water (scientifically accurate)
Monochromatic blue transition zone before color returns

Character Quick Ref

Ollie body: golden-amber #E8A735, gradient to cream underside
Ollie iris: gold foil in bright, silver in dark
Dot: white pearl, pearlescent nacre, mother-of-pearl sheen
Dot eyes: yellow upside-down half-U (default), blue inverted (blue)
Snail Easter egg hidden on every page

Reviewer Checklist

Ollie has NO ears, holes, or protrusions on head
Ollie body is golden-amber (not brown, not red)
Ollie eyes: gold in bright, silver in dark
Gold foil stipple at correct density
Dot is white pearl (not Tahitian)
Dot has NO mouth, NO smile

Correct ecosystem (coral reef only)
Color matches spread position
No text in the illustration
2:1 aspect ratio (1:1 for Title/Back Cover)
No page crease / two-page artifacts
Macro scale maintained