gemini-gen

Gemini Gen — Visual Understanding & Multimodal Analysis

Gemini 3.1 Pro via Renoise gateway. Zero npm dependencies, native fetch only. Handles files of any size automatically — small files are sent inline, large files (>20MB) are uploaded first.

Quick Start

# Analyze a product photo
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file photo.jpg --mode product

# Extract a video script with timestamps
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file clip.mp4 --mode video-script

# Extract visual style keywords from a reference
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file reference.jpg --mode style

# Free-form analysis
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file photo.jpg "Describe this image in detail"

Analysis Modes

Preset modes auto-select optimal resolution and output format. Use --mode for common tasks instead of writing custom prompts.

Product Analysis (--mode product)

Analyzes product photos and returns structured JSON with type, color, material, selling points, brand tone, and scene suggestions. Auto-selects high resolution for maximum detail.

node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file product.jpg --mode product

Output:

{
  "type": "resistance loop bands",
  "color": "Pink 10lb, Blue 15lb, Mint green 20lb",
  "material": "TPE elastic, matte finish",
  "selling_points": ["3 resistance levels", "foldable and portable", "pastel color scheme"],
  "brand_tone": "Youthful athletic, trendy fitness",
  "scene_suggestions": ["living room workout", "hotel room fitness", "outdoor park"]
}

Video Script Extraction (--mode video-script)

Watches a video and outputs timestamped dialogue, scene descriptions, and camera movements. Auto-selects low resolution to reduce token consumption.

node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file clip.mp4 --mode video-script

Style Extraction (--mode style)

Extracts visual style keywords from a reference image or video: color palette, lighting, camera language, composition, and mood.

node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file reference.jpg --mode style

CLI Usage

# Text only
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs "Explain quantum computing"

# Analyze an image (high resolution for product detail)
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file photo.jpg --resolution high "Describe this product"

# Analyze a video (low resolution to save tokens)
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file clip.mp4 --resolution low "Summarize this clip"

# Multiple images
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file a.jpg --file b.jpg "Compare these two"

# JSON output mode
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --json "Return a JSON object with name and age"

Options

Flag	Default	Description
--file <path>	—	Attach local file (repeatable). Files >20MB auto-uploaded
--file-uri <uri>	—	Attach uploaded file by URI (requires --file-mime)
--file-mime <mime>	—	MIME type for --file-uri
--resolution <level>	medium	low / medium / high / ultra_high
--model <name>	gemini-3.1-pro	Model name
--temperature <n>	1.0	Temperature
--max-tokens <n>	8192	Max output tokens
--json	off	Request JSON response format
--mode <name>	—	Preset analysis mode: product, video-script, style

Resolution Levels

mediaResolution controls token allocation per image/frame:

Level	Image Tokens	Video Frame Tokens	Best For
low	280	70	Bulk processing, video analysis
medium	560	140	General use (default)
high	840	210	Product photos, fine text
ultra_high	1120	280	Extreme detail

Large File Handling

Files larger than 20MB are automatically uploaded before analysis — no manual steps needed. The upload goes through the Renoise file gateway and returns a temporary URL (valid for 1 hour).

# This just works — the script detects file size and auto-uploads if needed
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file large-video.mp4 "Analyze this video"

For manual uploads (e.g. reusing the same large file across multiple calls):

# Upload once
FILE_URL=$(node ${CLAUDE_PLUGIN_ROOT}/skills/renoise-gen/scripts/upload.mjs large-video.mp4)

# Use the URL in multiple calls
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file-uri "$FILE_URL" --file-mime video/mp4 "Summarize"
node ${CLAUDE_SKILL_DIR}/scripts/gemini.mjs --file-uri "$FILE_URL" --file-mime video/mp4 "Extract dialogue"

When to Use vs When Not

Use gemini-gen for	Use renoise-gen for
Analyzing product photos	Generating images
Understanding video content	Generating videos
Extracting scripts from video	Text-to-video / image-to-video
Comparing visual assets	Product design sheets
OCR / text extraction	Scene backgrounds
Describing scenes for prompts	—

API Reference (Advanced)

Endpoint

POST https://renoise.ai/api/public/llm/proxy/v1beta/models/{model}:generateContent?key={RENOISE_API_KEY}

Request Format

{
  "contents": [
    {
      "role": "user",
      "parts": [
        { "inlineData": { "mimeType": "image/jpeg", "data": "<base64>" }, "mediaResolution": { "level": "media_resolution_high" } },
        { "text": "Describe this image" }
      ]
    }
  ],
  "generationConfig": {
    "temperature": 1.0,
    "maxOutputTokens": 8192
  }
}

Supported MIME Types

Extension	MIME Type	Max Inline
.jpg/.jpeg	image/jpeg	20MB
.png	image/png	20MB
.webp	image/webp	20MB
.gif	image/gif	20MB
.mp4	video/mp4	20MB
.mov	video/quicktime	20MB
.webm	video/webm	20MB

Error Handling

• 400: Bad request (check prompt format)
• 403: Invalid API key
• 429: Rate limited (wait and retry)
• 500: Server error (retry with backoff)

Authentication

Environment variable RENOISE_API_KEY. Get one at: https://www.renoise.ai