image-gen

Backend-agnostic image generation workflow: single images, series with anchor-chain consistency, and batch pipelines. Two backends: Gemini (API) and Nano Banana (local scripts with post-processing).

Reference Loading Table

Signal	Load These Files	Why
Every request (always load)	`references/series-consistency.md`	Anchor-chain and prompt-file-first rules apply to all generation
Every request (always load)	`references/backend-selection.md`	Mode decision required before every generation
Script output `gemini`	`references/backends/gemini.md`	Gemini API models, env vars, flags
Script output `nano-banana`	`references/backends/nano-banana.md`	Nano Banana subcommands, flags, aspect ratios

Phase 1: Detect Mode and Load References

Run the backend detection script — it reads environment variables and outputs a single word: