Image Gen Cli

• gpt-image
• gemini-image
• imagen
• flux
• kling

Providers and routes

• Direct providers expose their own native model endpoint: openai, google.
• Aggregated providers broker another vendor's model: together, replicate.
• Canonical models can map to multiple routes. Provider choice does not change the canonical model; it changes the concrete route used to execute it.
• If routes list --model <model> shows more than one provider, pass --provider explicitly to generate.

These currently require --provider together or --provider replicate:

• flux-1-schnell
• flux-1-kontext-pro
• flux-2-pro
• flux-2-dev
• flux-2-flex

Common command patterns

pnpm dev -- providers list
pnpm dev -- models list
pnpm dev -- models list --family flux
pnpm dev -- models list --provider google
pnpm dev -- routes list --model imagen-4-fast
pnpm dev -- routes list --model flux-2-pro --provider replicate
pnpm dev -- generate --model imagen-4-fast --prompt "A studio product shot"
pnpm dev -- generate --model flux-2-pro --provider together --prompt "A studio product shot"

Output mode guidance

• Default text output is best for human-readable confirmation in the terminal.
• Use --json when another tool, agent step, or test needs deterministic structured output, including provider, canonical model, provider model, raw response, and output references.
• Use --output-dir <dir> when generated assets must be persisted to deterministic local file paths for later steps.
• --json and --output-dir can be combined when both machine-readable metadata and saved files are needed.

Operating rules

• generate always requires --model and --prompt.
• Prefer routes list over guessing when a provider route may be ambiguous.
• If a prompt fails on one provider route, keep the canonical model fixed unless the user asked to change capabilities; first verify whether a different route for that same canonical model is intended.
• Do not invent provider-specific model ids in commands when the registry already exposes a canonical id or alias.