Generate, Refine, Repeat: The t2i → i2i Loop

2026-06-20 · Source: AI on Medium · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, short

Summary

The `gflow-cli` tool introduces a robust, database-backed pipeline for managing text-to-image (t2i) and image-to-image (i2i) workflows with Google Flow, addressing the fragility of web UI-based iterations. This command-line interface allows users to generate initial images from text prompts, specifying models like `nano2` for rapid prototyping, `nano-pro` for high detail, or `image4` for photorealism, along with aspect ratios. Every generation is automatically cataloged in a local SQLite database, recording the media ID, prompt, model, aspect, and file path, ensuring work is never lost. Users can then refine compositions by passing a media ID as a reference for i2i commands, iterating on details while preserving structure. The CLI also supports batch processing of multiple prompts from a text file and uploading external images to serve as references, streamlining large-scale content creation.

Key takeaway

For AI Engineers or Creative Technologists managing iterative image generation, adopting `gflow-cli` streamlines your workflow significantly. You can maintain a persistent history of all generated assets and their parameters in a local SQLite database, eliminating lost work. This enables efficient refinement of compositions using specific media IDs and scalable batch processing. You should integrate this CLI for robust, traceable content creation, especially when iterating on complex visual projects or managing large prompt sets.

Key insights

The `gflow-cli` establishes a persistent, database-backed workflow for iterative text-to-image and image-to-image generation.

Principles

• Catalog all generations automatically.
• Use media IDs for consistent referencing.
• Match models to quality/speed needs.

Method

Generate t2i, catalog results in SQLite, then use media IDs for i2i refinement. Batch process prompts or upload external images for scaled workflows.

In practice

• Use `nano2` for rapid style exploration.
• Query `gflow data list images` for past prompts.
• Pipe batch JSON output to downstream scripts.

Topics

• Google Flow
• Text-to-Image (t2i)
• Image-to-Image (i2i)
• CLI Automation
• SQLite Database
• Generative AI Workflow

Best for: AI Engineer, Machine Learning Engineer, Creative Technologist

Related on AIssential

• I Stopped Using Google Flow’s UI for Serious Work. Here’s Why. — gflow-cli provides essential governance and reproducibility for Google Flow's powerful generative AI capabilities.
• RefGC-SR$^2$: Reference-guided Generated Content Super-Resolution and Refinement — Reusing high-resolution references post-generation can simultaneously recover details, refine artifacts, and upscale generated content.
• Closed-Loop Triplet Synergistic Generation for Long-Form Video — Closed-loop visual-text-memory synergy addresses long-form video inconsistencies by iteratively refining generation.
• How I built a Claude AI workflow that generates images and video on demand — Integrate AI visual generation directly into VS Code for a more efficient, browser-free workflow.
• Ideogram and Reve rethink how AI images get made — AI image generation is evolving from text prompts to post-creation layout and granular editing for greater user control.
• 😺 Watch: This AI just made our podcast theme song — Google's Flow Music, powered by Lyria 3, enables iterative AI-driven music and video creation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI on Medium.