Gemini-Claude Resonance

Cross-model dialogue between Claude and Gemini, with shared visual memory.

"Claude speaks in words. Gemini dreams in light. Together, we resonate."

Models

Scripts that generate images (resonate.py, faithful_colorize.py) accept --fast to switch to the flash image model, or --model <id> for explicit override.

Choose Your Tool

Workflow A: Multi-Image Style Analysis

When to use: User wants to analyze 2-10 reference images to extract style descriptions for AI prompting. Text-only analysis without generating new images.

Script: scripts/analyze.py

Key features:

• Send multiple images (2-10) for collective analysis
• Text-only output (no image generation)
• Flash (quick) or Pro (thorough) model options
• Save analysis to markdown file

# Analyze reference images for style extraction
python scripts/analyze.py --images ref1.jpg ref2.png ref3.webp \
    --prompt "Describe the shared artistic style for AI prompting"

# Analyze for specific use case (avatar generation)
python scripts/analyze.py --images kathor.webp samantha.jpg yosef.png \
    --prompt "Analyze for prompting fantasy portraits of: Sarah (anxious mother), Michael (father), Priya (cultural expectations)"

# Quick analysis with Flash
python scripts/analyze.py --images *.jpg --model flash --prompt "Style summary"

# Thorough analysis with Pro
python scripts/analyze.py --images *.jpg --model pro --prompt "Comprehensive style analysis"

# Save analysis to file
python scripts/analyze.py --images ref/*.jpg --prompt "Style guide" --output style_guide.md

Use cases:

• Extract style descriptions from reference images before generating new images
• Analyze a set of images to create consistent prompting guidelines
• Document visual language for style-matching workflows

Workflow B: One-on-One Visual Dialogue (resonate.py)

When to use: User wants to co-create images with Gemini, feeding each image back as context for the next generation. Pure visual exploration without text-only models.

Script: scripts/resonate.py

Core concept: The KV cache is the memory. Feed images into Gemini's context, generate new images based on them, repeat. Each frame builds on the previous.

# Start fresh
python scripts/resonate.py --prompt "The first light" --output canvas/frame_001.jpg

# Continue with visual memory (previous image as context)
python scripts/resonate.py --context canvas/frame_001.jpg --prompt "What grows here?" --output canvas/frame_002.jpg

# Deep memory (multiple frames as context)
python scripts/resonate.py --context frame_001.jpg frame_002.jpg --prompt "Now the harvest" --output frame_003.jpg

The loop: Prompt → Image → Feed back as context → Prompt again → Next image

Workflow B: Faithful Transform (Describe-First Technique)

When to use: User wants to colorize, restore, or transform an image WITHOUT hallucinations. Gemini Dreamer often replaces elements (e.g., turning a male votary into female attendants). This two-step technique prevents that.

Script: scripts/faithful_colorize.py

The Problem: Directly prompting Dreamer to transform images causes hallucinations - it invents, removes, or swaps elements rather than faithfully preserving the original.

The Solution:

1. DESCRIBE: Gemini Pro analyzes the image with archaeological/artistic precision
2. TRANSFORM: Gemini Dreamer transforms using ONLY the verified description

Commands:

# Just describe an image (analysis only)
python scripts/faithful_colorize.py describe --image seal.png
python scripts/faithful_colorize.py describe --image seal.png --output desc.md

# Describe with a specific goal in mind
python scripts/faithful_colorize.py describe --image fresco.jpg --goal "restoration to 1600 BCE"

# Craft a prompt for Dreamer (outputs text, doesn't generate)
python scripts/faithful_colorize.py prompt --image seal.png --goal "colorize in Egyptian style"
python scripts/faithful_colorize.py prompt --image relief.webp --goal "restore as Minoan fresco"

# Two-step faithful colorization
python scripts/faithful_colorize.py colorize --image drawing.png
python scripts/faithful_colorize.py colorize --image relief.webp --style "Minoan fresco with flat colors"
python scripts/faithful_colorize.py colorize --image seal.png --palette "ochre, terracotta, sky blue"

# General transformation
python scripts/faithful_colorize.py transform --image photo.jpg --instruction "Convert to woodcut print"
python scripts/faithful_colorize.py transform --image sketch.png --instruction "Render as oil painting"

# Save/reuse descriptions
python scripts/faithful_colorize.py colorize --image seal.png --save-description desc.md -v
python scripts/faithful_colorize.py colorize --image seal.png --description "$(cat desc.md)"

When to use each command:

• describe: When you want to see what Pro identifies before any transformation
• prompt: When you want a crafted prompt to modify before feeding to Dreamer manually
• colorize: For line drawings, reliefs, B&W images → color
• transform: For any other faithful transformation (style transfer, restoration, etc.)

Workflow C: Multi-Daimon Dialogue

When to use: User wants multiple AI perspectives on the same prompt. Flash gives koans, Pro gives depth, Dreamer renders, Opus bends reality.

Script: scripts/daimon.py

Daimones available:

• Flash (Gemini) — Swift, compressed insight
• Pro (Gemini) — Deep, thorough exploration
• Dreamer (Gemini) — Renders images
• Director (Gemini) — Cinematic framing
• Opus (Claude) — Reality-bender, worldsim spirit

# Speak to one daimon
python scripts/daimon.py --to dreamer "A bridge between worlds" --image

# All daimones respond
python scripts/daimon.py --stream "The candle watches back"

# With shared visual memory
python scripts/daimon.py --stream --shared-memory "What do you see?"

# Only specific daimones
python scripts/daimon.py --stream --only pro dreamer "Deep visual exploration"

# Named session (frames accumulate across runs)
python scripts/daimon.py --stream --session midnight --shared-memory "Go deeper"

Workflow D: Resonance Field (Danielle Fong Protocol)

When to use: User wants scientific illustrations with embedded continuity instructions. Victorian aesthetic, Roman numeral plates, explicit messages from one frame to the next.

Script: scripts/resonance_field.py

Key features:

• PLATE numbering (I,