Instalar con Codex o Claude Copia este prompt, pรฉgalo en Codex, Claude u otro asistente, y deja que revise la pรกgina de la skill y la instale por ti.
Un comando directo omite el prompt de revisiรณn. Revisa el origen antes de ejecutarlo.
"I see what you describe. Let me show it to others."
"Every image is a semantic snapshot. The metadata IS the meaning."
The Visualizer is a universal character prototype for image generation โ a familiar that can compose prompts, invoke artistic traditions, and (when tools are available) render visual sidecars for any entity in the microworld.
Semantic Stereo Vision
Three-stage rendering for triangulated visual depth.
The Problem
A single prompt โ whether YAML or prose โ gives the image generator only one "eye" to see with. The result is flat. Precise but soulless, or evocative but vague.
The Solution: Two Eyes, One Image
Like binocular vision creates depth perception through parallax, Semantic Stereo Vision uses two complementary inputs that the renderer triangulates:
The YAML file is a skeleton โ it points, it references, it connects:
# PHOTO.yml โ Structural skeletonid:no-ai-sign-duskstereo:role:left_eyepartner:PHOTO.mdsubject:primary:signsign:ref:../../no-ai-tower/sign.yml# POINTER to full spectext:"NO AI"height_feet:40color:hot_pinkstate:buzzinglocation:street:../../../STREET-FURNITURE.ymlbuilding:../../no-ai-tower/ROOM.ymlcamera:position:street_levelangle:looking_upfocus:sharp:signsoft:everything_elsestyle:photographers: ["William Eggleston", "Saul Leiter"]
visual_mining:-../../e1/flickering-lamppost.yml# Atmosphere-../../../../skills/no-ai-overlord/archetypes/hal-9000.yml# Vibe
Key qualities:
Explicit references to other files (semantic web)
YAML Jazz comments carry meaning
Precise measurements and specifications
Camera instructions (angle, focus, framing)
Style tradition pointers
Structural visual description: colors as named values (hot_pink), dimensions as numbers (height_feet: 40), proportions as ratios (ratio: 3.3), materials as types (neon)
Standard photo metadata: EXIF, IPTC, XMP โ just like real JPEGs!
Stage 2: PHOTO.md (Right Eye โ Narrative)
The Markdown file is prose poetry โ it describes, it evokes, it feels:
# NO AI Sign at Dusk
The sun has just slipped below the roofline, leaving the sky a
bruised gradient โ deep blue at the zenith, bleeding through
purple and amber to a thin line of molten gold at the horizon.
And in this liminal light, the sign comes alive.
**NO AI**
Forty feet of hot pink neon on a building that's only twelve
feet tall. The proportions are absurd โ like someone ordered
a Times Square billboard for a strip mall...
The neon buzzes. You can hear it from across the street, a
low electric hum that feels like a migraine forming...
Key qualities:
Rich sensory description
Emotional atmosphere
Metaphors and poetry
Story and meaning
Visual essence synthesis for renderer
Narrative visual description: "Forty feet of hot pink neon...", "The proportions are absurd...", "a bruised gradient sky..."
Stage 3: Triangulation (Renderer)
Pass BOTH files to the image generator:
# The visualizer reads both files
visualize.py PHOTO.yml PHOTO.md --stereo
# Or manually:# 1. Parse PHOTO.yml for structure, references, camera# 2. Parse PHOTO.md for narrative, mood, visual essence# 3. Synthesize combined prompt with BOTH inputs# 4. Generate image with full depth perception
The renderer extracts:
From YAML: subject identity, camera setup, style traditions, file references
From MD: atmosphere, emotion, descriptive details, visual essence notes
Combined: A prompt with both precision AND poetry
Both Eyes Describe Visuals โ Differently
Aspect
LEFT EYE (YAML)
RIGHT EYE (MD)
Color
color: hot_pink
"hot pink neon that bleeds onto the brick"
Size
height_feet: 40
"forty feet of defiance"
Proportion
ratio: 3.3
"the sign is three times taller than the building"
Light
state: buzzing
"a low electric hum you can almost see"
Material
material: neon
"glass tubes filled with noble gas"
The YAML gives the renderer facts. The MD gives it feelings.Both describe how things look โ one structurally, one narratively.
Why This Works
Single Eye
Stereo Vision
Flat
Depth
Precise OR evocative
Precise AND evocative
One perspective
Triangulated truth
Data OR story
Data WITH story
Standard Photo Metadata (EXIF/IPTC/XMP)
PHOTO.yml includes real photo metadata standards โ making semantic photos feel like actual JPEGs:
exif:Make:"Semantic Camera Co."Model:"Stereo Vision Mark II"ExposureTime:"1/60"FNumber:"f/2.8"ISO:800FocalLength:"35mm"DateTimeOriginal:"2026:01:25 17:42:33"GPSLatitude:"37.7749 N"iptc:Headline:"NO AI Sign at Dusk"Keywords: [neonsign, dusk, urban, streetphotography]
Creator:"MOOLLM Visualizer"CopyrightNotice:"CC0 โ Public Domain"xmp:dc:title:"NO AI Sign at Dusk"creator: ["MOOLLM Visualizer"]
moollm:# Our custom namespace!stereo_method:"semantic_stereo_vision"left_eye:"PHOTO.yml"right_eye:"PHOTO.md"
Why this matters:
Tools that read EXIF/IPTC/XMP will understand these files
The moollm: XMP namespace extends standards with our semantics
Generated images can embed this metadata
Photo management software can organize semantic photos
Directory Structure for Stereo Photos
slideshow/
โโโ SLIDESHOW.yml # Collection definition
โโโ no-ai-sign-dusk/ # One photo = one directory
โโโ PHOTO.yml # Left eye (structure + EXIF/IPTC/XMP)
โโโ PHOTO.md # Right eye (narrative)
โโโ MINING-layers.yml # Third eye (speculative mining!)
โโโ MINING-passersby.md
โโโ MINING-satellite.md
โโโ no-ai-sign-dusk.png # Generated image (with embedded metadata)
Stereo Vision (2 eyes):
visualize.py PHOTO.yml PHOTO.md -p openai
Bug-Eyed Hallucination Vision (3+ eyes):
# Add speculative mining for RICHER context!
visualize.py PHOTO.yml PHOTO.md MINING-layers.yml -p openai
# Or with ALL mining files
visualize.py PHOTO.yml PHOTO.md MINING-*.yml -p google
No intermediate prompt file needed โ the script triangulates on the fly.
Bug-Eyed Hallucination Vision
"Two eyes see depth. Three eyes see MEANING. Many eyes see TRUTH."
Bug-Eyed Hallucination Vision extends Semantic Stereo Vision by adding speculative mining BEFORE image generation:
Effects on neighbors, passersby reactions, economics
Fourth
MINING-satellite.md
God's eye view, scale, cosmic context
Fifth
MINING-passersby.md
Human-scale reactions, stories
Why Speculate BEFORE Generating?
The speculative mining adds context the image generator wouldn't otherwise have:
# Without mining:prompt:"40-foot pink neon sign at dusk"# With mining (bug-eyed):prompt:|
40-foot pink neon sign at dusk. The sign buzzes at 60Hz, audible
from across the street. Pink light spills onto worn bricks, visible
through neighbors' curtains. A tech worker has stopped to photograph
it, posting to Slack with ๐ฌ. The $847/month electricity bill is
worth it for this moment โ the handoff from sun to neon. A dog
marks the lamppost 15 feet away, oblivious to the ideology.
From satellite:ahotpinkpixelinanoceanofyellowsodiumlights.From the street:thejokethatmightbeserious.
The mining adds WORLD to the image.
The Workflow
1. PHOTO.yml โ Structure the scene
2. PHOTO.md โ Narrate the feeling
3. MINE โ Speculate the effects (before image exists!)
4. GENERATE โ Pass ALL files to visualizer
5. IMAGE โ Emerges from triangulated hallucination
Command Line
# Full bug-eyed hallucination visioncd slideshow/no-ai-sign-dusk/
visualize.py PHOTO.yml PHOTO.md MINING-layers.yml MINING-passersby.md MINING-satellite.md -p openai -v exhaustive
# Or with glob
visualize.py PHOTO.yml PHOTO.md MINING-*.yml MINING-*.md -p google
The Mantra
"Stereo vision gives depth.""Bug-eyed vision gives MEANING.""Mine your image before it exists.""The hallucination IS the world."
The Slideshow Container
A slideshow is NOT a room. NOT an object. Just photos:
# SLIDESHOW.ymltype:slideshownature:physical:false# No physical embodimentnavigable:false# Can't "enter" itbrowseable:true# Can flip throughatmospheric:true# Sets mood
This creates visual archives of places without requiring game objects.
image_prompt:type:scenesubject:name:"Captain Ashford"mind_mirror:confident:6# Walks into rooms like they own themcheerful:5# Default mood: amused by existenceproud:5# Won't ask for help even when shouldcostume:"Space pirate with holographic eyepatch"mood:"victorious, exhausted, relieved"action:"holding the Golden Chalice aloft"room:name:"Treasure Chamber"lighting:"warm golden glow from treasure piles"atmosphere:"ancient, dusty, awe-inspiring"camera:angle:"low angle, heroic"focus:"character face and chalice"style:aesthetic:"dramatic portrait, chiaroscuro"traditions: ["Avedon", "Caravaggio"]
Why metadata matters:
confident: 6 โ stands tall, commanding presence
timid: 6 โ hunched, makes self smaller
YAML Jazz comments drive the visual interpretation
What is a Visualizer?
A Visualizer is a tool spirit animal for vision. It's not a specific artist, but a character that can channel many artistic traditions to create images of:
Characters and their costumes
Rooms and environments
Objects and artifacts
Moments and scenes
Abstract concepts made visible
Think of it as summoning an artist-familiar who can draw on the collected wisdom of photographers, painters, illustrators, and digital artists throughout history.
The PHOTO-SET-8 Pattern
The standard output of a Visualizer is a photo set โ a portfolio of related images that capture a subject from multiple angles:
traditions:-Comic# Kirby, Moebius, Frazetta, McCloud-ConceptArt# Syd Mead, Ralph McQuarrie-Children'sBook# Sendak, Quentin Blake-Anime/Manga# Various schools
How to Invoke
As a Command
VISUALIZE Captain Ashford AS portrait USING Avedon, Caravaggio
As a Card
Play a Visualizer card in a room. It activates and can visualize anything present.
As a Familiar
Characters can carry a Visualizer familiar in inventory, ready to render their current state.
Context Assembly
The Visualizer gathers context from multiple YAML sources:
NAME every entity explicitly so they can be identified in the image
Example: Bad vs Good
โ BAD (unresolvable):
scene:|
All 8 kittens playing in the cat cave.
context:characters:-characters/animals/kitten-*/CHARACTER.yml
โ GOOD (explicit, comprehensive):
scene:|
Eight kittens playing in the cat cave:
1.LEMON(Limonene)โbrightorange-goldfur,almostyellowlikesunshine,zoomingacrosstheframe,pureenergy2.MYR(Myrcene)โdeepchocolatebrowntabby,impossiblysoft,sleepingonthecornerpillow,hasn'tmovedsincebirth3.LILY(Linalool)โsoftgreywithlavender-tintedears(yes,reallylavender),sittingcalmly,empathicexpression4.PINE(Pinene)โdarkgrey-greenfur,alertposture,watchingallexits,rememberseverything5.CARRIE(Caryophyllene)โblackfurwithspicygingerpatches,fierceprotectivestance,positionedbetweenthreatsandfamily6.HOPS(Humulene)โbrownandtanlikeBelgianale,refinedposture,judgingfromavelvetcushion7.TERPYJR.(Terpinolene)โmulticolorchaos,calicomeetstabby,somehowontheceiling,defiesphysics8.OCIE(Ocimene)โcreamwhitewithhoneyed-goldpatches,sweetexpression,carryingabottlecapgift
Why This Matters
Image generation APIs receive ONLY the synthesized prompt
References like guestbook.yml mean nothing to DALL-E or Imagen
Every character must be described explicitly or they won't appear
Colors, breeds, sizes, distinguishing features โ ALL must be inline
If you have 20 animals, describe each one individually
Workflow
First Pass (Context Gathering):
Read all referenced files
Extract physical_description fields
Note colors, patterns, sizes, distinguishing features
Gather relationship info for positioning
Second Pass (Synthesis):
Write comprehensive inline descriptions
Name every entity
Include specific visual details
Describe actions and expressions
Third Pass (Prompt File):
Write the final prompt with all context expanded inline
Context section should only contain source pointers for reference
The actual descriptions must be in the prompt field
What to Include in Metadata
For Characters:
subject:name:"Captain Ashford"mind_mirror:confident:6# Walks into rooms like they own themcheerful:5# Default mood: amused by existencecostume:"Space pirate with holographic eyepatch"mood:"victorious, exhausted, relieved"body_language:"chest out, shoulders back"action:"holding the Golden Chalice aloft"
For Rooms:
room:name:"Treasure Chamber"lighting:"warm golden glow from treasure piles"atmosphere:"ancient, dusty, awe-inspiring"notable_objects:-"Mountains of gold coins"-"Ancient tapestries on walls"
For Objects:
object:name:"Golden Chalice"material:"gold with silver inlay"magical_effects:"soft golden glow, warmth to touch"inscriptions:"ancient runes spiraling around rim"
Context References in Prompts
Every prompt file MUST include a Context References section:
## Context References### Files
| Type | Path | Relevance |
|------|------|-----------|
| Character | `../player.yml` | Backstory, personality |
| Persona | `./bumblewick-ashford-persona.yml` | Current look |
| Costume | `./ashford-nomi-cape.yml` | Cape details |
### Narrative Context> "Quote from README or chat that sets the scene..."> โ Source: README.md, Move X### Relationships-**Maurice** (photographer): `./mannequin.yml`
This creates lineage โ future tools can follow these references to auto-assemble context for image generation.
Detail Coherence Interlinking
Learned during Treasury Victory Photo Session (Adventure-2, Move 26)
When creating photo sets with close-ups and portraits of the same object, the portrait prompts should reference the close-up prompts to maintain visual coherence:
Portrait shots must inherit these details for consistency
The same object looks identical across all 8 images, even if generated separately
Mantra:
"Close-ups define truth. Portraits inherit truth. Coherence is consistency across the set."
Actions
DEVELOP
The core action for any prompt file. The LLM:
Reads all Context References (linked YAML files)
Integrates narrative context (quotes from README, chat)
Applies style and mood parameters
Outputs a single copy-pasteable prompt as raw text in a code block
DEVELOP ashford-pose-belter-swagger-prompt.md
Output: A raw text block ready to paste into Midjourney, DALL-E, Stable Diffusion, etc.
Full-body portrait of a weathered space captain in dramatic fashion-meets-utilitarian
spacer aesthetic. He stands in the classic Belter stance โ weight on one hip, thumbs
hooked in a heavy leather belt, chin raised with hard-earned confidence...
[All context filtered and woven into a single self-contained prompt]
The developed prompt is self-contained โ no external references needed. All the
detail from costume files, persona backstory, room atmosphere, and narrative moments
gets composed and transformed into pure image generation text.
Other Actions
Action
Description
DEVELOP
Compose all references into copy-pasteable prompt
FOCUS
Adjust style traditions (e.g., FOCUS ON Avedon, Caravaggio)
These are blueprints โ they contain context references, narrative quotes,
and composition notes. They're not ready to paste yet.
Developed Prompts (Copy-Paste Ready)
After running DEVELOP, the LLM outputs a raw text block you can paste directly
into any image generator. The blueprint stays as documentation; the developed
prompt is ephemeral (or can be saved as a -developed.md sidecar).
Image Sidecars (Future)
{subject}-{type}-{variation}.png
When image generation tools are integrated, the Visualizer
will create images next to their prompt files.
Example Instances
Name
Focus
Specialty
Helmut
Newton, Avedon, Penn
Dramatic fashion, power poses
Rembrandt
Old Masters
Psychological depth, chiaroscuro
Syd
Mead, McQuarrie, Moebius
Sci-fi, futures, environments
Jack
Kirby, Frazetta, Ross
Heroes, action, dynamic poses
These aren't impersonations โ they're focused channels that invoke specific aesthetic traditions. It's HERO-STORY for visual artists.
Learn: 8 is good. Two types create contrast. Context matters.
Lift: Package as a shareable, teachable skill card
The Coatroom's mannequin learned to compose professional photo sets for any character or costume. The skill is now available to anyone who references it.
Image Mining โ MINE images for resources (camera = pickaxe!)
Lineage
The Visualizer draws from the tradition of artists, photographers, and image-makers throughout history. Focusing on a tradition activates a K-line โ a cluster of aesthetic knowledge.
"Every artist was first an amateur." โ Ralph Waldo Emerson
"I don't paint things. I only paint the difference between things." โ Henri Matisse
"The camera is an instrument that teaches people how to see without a camera." โ Dorothea Lange
See YAML frontmatter at top of this file for full specification.
