Anthropic

Mode: text · Models: 3

Vendor: Anthropic · Official API docs: docs.anthropic.com

Anthropic's Claude models are large language models for analysis, not media generation: give them an image (and an optional question) and they return text — captions, OCR, classification, or a description. They power the gen-ai describe command. Three tiers trade capability for speed and cost: Opus (most capable), Sonnet (balanced, the default), and Haiku (fastest).

Models

id	Name	Input type
claude-opus-4-8	Claude Opus 4.8	i2t
claude-sonnet-4-6	Claude Sonnet 4.6	i2t
claude-haiku-4-5	Claude Haiku 4.5	i2t

CLI

# describe an image (Sonnet is the default model)
gen-ai describe -i photo.jpg

# ask a specific question with the most capable model
gen-ai describe -m claude-opus-4-8 -i receipt.jpg -p "extract the total and tax"

# fast, high-volume captioning — pipe clean text out
gen-ai describe -m claude-haiku-4-5 -i product.jpg --script | pbcopy

These models return text, so they run through gen-ai describe (not generate) and print to stdout — no download or Drive save.

MCP

{ "name": "picsart_generate",
  "arguments": {
    "model": "claude-sonnet-4-6",
    "prompt": "What brand is the shoe?",
    "imageUrls": ["https://example.com/photo.jpg"]
  } }

Parameters

Full parameter surface, sourced from gen-ai models info <id> --json. CLI flags show the primary short form; the canonical --kebab-case long form always works too.

claude-opus-4-8 — Claude Opus 4.8

Try claude-opus-4-8 in Playground ↗

Input type: i2t

Param	CLI flag	Type	Values
prompt	-p	text	required
imageUrls	-i	file	image (up to 8)

claude-sonnet-4-6 — Claude Sonnet 4.6

Try claude-sonnet-4-6 in Playground ↗

Input type: i2t

Param	CLI flag	Type	Values
prompt	-p	text	required
imageUrls	-i	file	image (up to 8)

claude-haiku-4-5 — Claude Haiku 4.5

Try claude-haiku-4-5 in Playground ↗

Input type: i2t

Param	CLI flag	Type	Values
prompt	-p	text	required
imageUrls	-i	file	image (up to 8)

Pricing

gen-ai pricing claude-sonnet-4-6

Claude analysis is priced per generation; cost is resolved per modelId via the backend /options call.