convez.ai
SkillDesign & UIIntermediário
cvas-544

Image Studio CC

cvas-544/image-studio-CC

Pipeline completo de geração de imagens com análise de referência e refinamento iterativo

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill de criação de imagens com IA que analisa referências visuais, constrói prompts otimizados para Nano Banana, gera imagens e as refina em workspace local. Inclui memória de estilo, identidade de marca e engenharia de prompts multi-modo.

Instalação

$npx convez-ai cvas-544/image-studio-CC

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
28%
Último commit
há 8 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Criar imagens alinhadas à identidade visual de marca
  • Refinar e iterar sobre gerações de imagem em workspace local
  • Aprender estilos visuais e reutilizá-los em futuras criações

Veredito

Skill bem estruturada que eleva a geração de imagens com IA ao nível de produção, agregando análise de referência, memória de estilo e workspace iterativo. Ideal para equipes que precisam manter consistência visual em múltiplas gerações.

README

A tradução para o português deste README ainda está na fila.

image-studio-CC

A complete AI image creation workflow skill for Claude Code. Combines intelligent prompt engineering, brand identity, style learning, and image generation into a single end-to-end pipeline — powered by the nano-banana skill for generation and a local refinement workspace for iteration.


What It Does

image-studio-CC is the "brain" layer on top of raw image generation. It:

  • Analyzes reference images and extracts visual fingerprints
  • Asks you how you want to use the reference (recreate, inspire, brand-align, or customize)
  • Builds a precise, natural-language prompt optimized for Nano Banana / Imagen
  • Calls the nano-banana skill to generate the image
  • Opens a local refinement workspace with the prompt pre-loaded for annotation and iteration
  • Learns your visual style over time and saves it to a portable memory file

Installation

# Clone directly into your Claude Code skills directory
git clone https://github.com/cvas-544/image-studio-CC.git ~/.claude/skills/image-studio-CC

# Run setup — installs nano-banana dependency and checks requirements
cd ~/.claude/skills/image-studio-CC && bash setup.sh

Then load the skill in Claude Code:

image-studio-CC

Files

image-studio-CC/
├── SKILL.md                     ← Full skill logic (loaded by Claude Code)
├── nano-banana-server.js        ← Local bridge server for the refinement workspace
├── nano-banana-playground.html  ← Visual refinement workspace (opens in browser)
├── setup.sh                     ← Installer — checks dependencies, sets up nano-banana
└── README.md                    ← This file

Dependencies

  • nano-banana skill — handles actual image generation via Gemini CLI. Must be installed at ~/.claude/skills/nano-banana/
  • Gemini CLI — must be installed and configured with a valid GEMINI_API_KEY
  • Node.js — required to run the local bridge server

Set your API key

export GEMINI_API_KEY="your-key-here"

Get your key at aistudio.google.com.


How to Use

Step 1 — Load the skill

In Claude Code, type:

image-studio-CC

Or pass a reference image path directly:

image-studio-CC Test.jpg

Step 2 — Choose a command

CommandWhat it does
/promptAnalyze a reference image and generate an optimized prompt
/generateFull pipeline — prompt → generate image → open workspace
/brandQuick brand-aligned image creation using saved brand settings
/recreateInstantly recreate a reference image as closely as possible
/inspireUse a reference image's style, apply it to new content
/setupSet up your brand identity (colors, style, tone, font)
/changeStyleUpdate saved brand settings
/showBrandView your current brand settings
/carouselGenerate 3–6 brand-aligned prompts for a carousel series
/postGenerate an image prompt + LinkedIn post + blog post together
/exportPDFExport the prompt + image + brand settings as a PDF brief

The Full Workflow

You load image-studio-CC (with optional image path or attachment)
        ↓
Skill silently loads:
  - Annotation patterns (nano-banana-skill-memory.json)
  - Style memory      (~/.config/image-studio-cc/style-memory.md)
  - Brand settings    (~/.config/image-studio-cc/brand.json)
        ↓
You run /generate (or /prompt)
        ↓
Skill asks: which mode?
  1. Recreate closely
  2. Style inspiration
  3. Brand-aligned recreation
  4. Recreate with my changes  ← type your own modifications
        ↓
Prompt is built silently (no further questions)
        ↓
nano-banana skill generates the image
        ↓
Refinement workspace opens in browser
  - Prompt is pre-loaded
  - Annotate regions, add numbered notes
  - Bake refined prompts, track iteration history
  - Annotation patterns saved back to skill memory

Mode Selection

When you provide a reference image, you pick one of four modes:

Mode 1 — Recreate Closely

Reproduce the reference as faithfully as possible. Same composition, lighting, palette, materials, mood.

Mode 2 — Style Inspiration

Capture the visual aesthetic from the reference and apply it to new content you specify. The style transfers, the subject changes.

Mode 3 — Brand-Aligned Recreation

Use your saved brand colors, style, and tone as the visual language — apply it to new content inspired by the reference composition.

Mode 4 — Recreate with My Changes

Closely match the reference but layer your own modifications on top. You type what you want changed (e.g., "change the shoes to orange", "replace left screen with an AI revolution image", "make the background warmer").


Brand Setup

Run /setup once to save your brand identity:

Brand name, accent color, background color, text color,
secondary color, visual style, tone, font preference

Saved to: ~/.config/image-studio-cc/brand.json

Once saved, brand settings load silently every session — never asked again unless you run /setup or /changeStyle.


Style Memory

Every reference image you analyze is automatically added to your style memory file:

~/.config/image-studio-cc/style-memory.md

This file accumulates your visual style profile over time — recording visual style, color story, lighting character, composition tendencies, mood, and storytelling approach from each reference.

This file is portable. Paste it into ChatGPT, Gemini, or any other AI to instantly transfer your style context to that model.

Example entry:

## Reference: Man from behind at dual monitors in minimalist studio — 2026-03-04

**Visual Style:** Cinematic editorial photography, photorealistic, fashion-tech aesthetic
**Color Story:** Near-monochromatic — white, charcoal black, chrome silver
**Lighting:** Soft diffused overhead ambient, even with no harsh shadows
**Composition:** Rear-centered, symmetrical, mid-level camera angle
**Mood:** Contemplative, focused, editorial, slightly clinical yet premium
**Storytelling:** Subject immersed in work — tension between human and screen
**Prompt patterns learned:** "seen from behind", "symmetrically framing", "near-monochromatic palette"

Refinement Workspace

The local workspace (nano-banana-playground.html) lets you refine generated images visually:

  • Click or draw regions on the image to pin annotations
  • Each annotation ties to a numbered card in the side panel
  • Annotations merge into the prompt automatically
  • Repeated annotation patterns are tracked in skill memory and suggested in future sessions
  • Bake versions to track your prompt evolution over time

The bridge server (nano-banana-server.js) runs locally on port 3333 and connects Claude Code to the workspace. It starts automatically when you run /generate.

Manual server control

# Start
node ~/.claude/skills/image-studio-CC/nano-banana-server.js

# Stop
kill $(lsof -ti:3333)

# Check logs
tail -f /tmp/nb-server.log

Smart /generate Behavior

/generate is the primary command. It's smart about context:

SituationWhat happens
/prompt already ran this sessionReuses that prompt — no re-prompting
Image provided, no prior promptAsks mode (1/2/3/4) → builds silently → generates → opens workspace
No image, no prior promptAsks one-line description → builds silently → generates → opens workspace

Relationship with nano-banana

image-studio-CC and nano-banana are separate, complementary skills:

SkillRole
image-studio-CCOrchestration layer — prompt engineering, brand, style memory, workspace, mode selection
nano-bananaGeneration engine — executes Gemini CLI commands to produce the image

image-studio-CC calls nano-banana internally during /generate. You never need to invoke nano-banana directly for image studio workflows.


Skill Memory Files

FilePurpose
~/.config/image-studio-cc/brand.jsonYour saved brand identity
~/.config/image-studio-cc/style-memory.mdCumulative visual style journal (portable)
~/.claude/projects/.../nano-banana-skill-memory.jsonAnnotation patterns from workspace refinements

👨‍💼 Author

Built by Vasu Chukka

📬 vasu.chukka@outlook.com
💻 https://www.linkedin.com/in/vasu-chukka-1a3569116/

Ver o original no GitHub

Bom para

Relacionados