Skip to content

MODEL-POLICY.md

Latest commit

 

History

History
43 lines (30 loc) · 1.4 KB

MODEL-POLICY.md

File metadata and controls

43 lines (30 loc) · 1.4 KB
Raw

Model Policy

Rafiki's v1 default image model is gemini-2.5-flash-image.

That default is intentionally local-first and iteration-first: it works well for fast prompt development, supports the existing Gemini path in the CLI, portal, and MCP tools, and keeps setup simple for users who only have a Google API key.

Defaults

  • CLI: generate.py and index.js default to gemini-2.5-flash-image.
  • Portal: Prompt Studio uses gemini-2.5-flash-image when the request does not include a model.
  • MCP: rafiki_generate and rafiki_batch default to gemini-2.5-flash-image.
  • Prompt files may override the default with **Model:** ... per prompt.

When To Choose Gemini

Use Gemini when you want fast local iteration, style exploration, reference image workflows, or higher-resolution Gemini Pro outputs via gemini-3-pro-image-preview.

Useful aliases:

  • flash, nano, gemini: gemini-2.5-flash-image
  • pro: gemini-3-pro-image-preview

When To Choose OpenAI

Use OpenAI when your project depends on OpenAI-specific behavior, when a prompt set was authored and approved against OpenAI outputs, or when you are using the separate gpt-image-1 Streamlit workflow documented in the operator guides.

Useful aliases:

  • gpt, gpt2: gpt-image-2
  • gpt1: gpt-image-1
  • dalle3: dall-e-3

Always pass OpenAI models explicitly with --model; do not rely on old roadmap examples as the default policy.