tacosdedatos Illustrator

End-to-end illustration creation for tacosdedatos post banners.

Required: Read agents/shared/illustration-style-guide.md for modes, colors, templates.

Related skills: image-prompt-writer (prompt-only), image-validator (QA-only), gemini-imagegen (API details).

The Creative Workflow

Phase 1: Discovery

Gather inputs (ask if not provided):

Phase 2: Creative Direction

1. Select the illustration mode

2. Identify the visual metaphor

Don't illustrate literally. Find a metaphor:

• AI agents → orchestra conductor
• Data pipeline → flowing river of shapes
• Learning journey → garden cultivation
• Time management → melting clock

3. Plan the composition

• Single focal point
• Generous negative space
• 2-3 accent colors max
• Dark background (usually)

Phase 3: Prompt Crafting

Use the image-prompt-writer skill internally, or craft manually using templates from illustration-style-guide.md.

Prompt structure:

[Style] on [background color].
[Subject/metaphor description].
Colors: [specific hex codes].
Style: [texture, composition notes].
[Negative constraints].

Always include negatives:

no text, no words, no labels, no watermarks, no signatures, no logos
no photorealism, no 3D render, no harsh gradients, no glossy surfaces

Phase 4: Generation

Use the gemini-imagegen skill to generate.

Settings:

• Model: gemini-3-pro-image-preview
• Aspect ratio: 16:9 or 3:2 (landscape)
• Resolution: 2K (balance of quality/speed)

CRITICAL: Save to /tmp/ with UUID filename for Discord delivery:

import uuid
filename = f"/tmp/{uuid.uuid4()}.jpg"
image.save(filename)
print(f"Image saved to: {filename}")

Phase 5: Validation

Check the generated image against the quality checklist:

• Fits the selected mode
• Uses colors from palette
• Has subtle texture/grain
• No embedded text
• Clear visual metaphor
• Generous negative space
• Proper aspect ratio (landscape)

If validation fails: Iterate with refined prompt. Common fixes:

• Add more specific color hex codes
• Emphasize "no text" in negatives
• Simplify the composition
• Add "subtle grain texture"

Phase 6: Delivery

Provide the final image with context:

## Post Banner: [Title]

**Mode**: [Selected mode]
**Metaphor**: [Visual concept]

Image saved to: /tmp/[uuid].jpg

### Design Rationale
[Explain your creative choices - why this mode, why this metaphor, what the visual elements represent]

### Prompt Used
[Include for reproducibility]

Iteration Patterns

If colors are wrong

Re-prompt with explicit hex codes:

Colors MUST be exactly: background #121212, accents #E63946 and #A8DADC

If text appears

Strengthen negatives:

Absolutely no text, no letters, no words, no writing, no labels, no numbers

If too cluttered

Simplify prompt:

Single focal element only. Generous negative space. Minimal composition.

If style is off

Add texture cues:

Subtle risograph grain texture. Flat illustration style. No gradients.

Example Workflow

Request: "Create a banner for 'Construye tu primer equipo de IA esta tarde'"

1. Discovery

• Title: "Construye tu primer equipo de IA esta tarde"
• Summary: Tutorial on building an AI agent team quickly
• Tone: Energetic, practical, empowering
• Key concept: Orchestrating multiple AI agents

2. Creative Direction

• Mode: Abstract/Geometric (technical + orchestration theme)
• Metaphor: Conductor's baton directing flowing data streams
• Composition: Baton on left, streams flowing right to binary/music notes

3. Prompt

Flat digital illustration on very dark background (#121212).
A conductor's baton on the left, directing three flowing wave-like streams that carry musical notes and binary code.
Colors: Crema Nevada (#F1FAEE) for the baton, Azul Niebla (#A8DADC) for the streams, Rojo Fuego (#E63946) for accent marks on the streams.
Style: clean geometric, subtle grain texture, generous negative space, elegant flow.
No text, no words, no photorealism, no 3D render, no harsh gradients.

4. Generate using gemini-imagegen with 16:9, 2K

5. Validate - Check all criteria

6. Deliver with rationale

Anti-Patterns