green rose systems

“growth, forever”
A Software Studio · Established 2026

cascade-img

A visual asset generation pipeline an LLM can run.

01  ·  description

cascade-img lets you go from an idea to a folder of finished images in a single conversation. You tell an AI assistant what you want — by voice or text — and it composes the prompts, generates the images or video, picks the good ones, cleans them up, and files them away. Instead of making one picture at a time on a paid service, then saving each result in the web UI, downloading it, building folders, and tracking it all by hand, you generate by conversation and let the assistant manage the whole process. You start with a rough idea and end up with a coherent, related set of assets.

You work as the critic or director, not the operator. Set a moodboard and a reference image, then move through batches fast — “yes, I like that”; “no, wrong direction”; “let’s go back two”; “make that the reference style now” — conversing with the assistant instead of typing prompts yourself. And everything is recorded: each attempt — the prompt, the result, why it was kept or discarded — goes into a log the assistant reads back next time, so it builds on what has already been tried instead of starting over.

The generator it drives today is Midjourney — one of the strongest models for stylized, art-directed work, and the hardest one to reach, which is why we started there. Midjourney has no public API; the only way in is its Discord bot, where you type /imagine and it replies with a 2×2 grid of four candidates to pick from and upscale. cascade-img automates that whole Discord flow for you, so you never have to touch Discord or learn the prompt syntax. It splits the prompt into composable parts you set independently and exposes the loop through an MCP server — the protocol Claude, Cursor, and others use to call tools — so the agent can compose, generate, curate, and log without your input on every generation. Other backends (Flux, DALL·E, Imagen) slot in behind the same interface.

To use it, just open the repo, point your assistant at it, and tell it to read AGENTS.md — from there your Claude can drive the whole thing over MCP at your direction. There is a CLI and a Python API for scripting and embedding too, but the agent loop is the main point: you converse, it generates, and you walk away with a written trail of exactly what produced each final image.

02  ·  repository

github.com/laffeyp/cascade-img

Contact

We’re open to a small number of new engagements. Write with a short note about the work and the timeline.

info@greenrosesystems.com