I didn't plan to share this one.
I was in the middle of launching Vizora — a database documentation tool — and I needed a hero image for the launch post. Not a placeholder. An actual creative brief I could hand to a designer or drop into Midjourney.
I typed into #koko-content on Discord:
Create a detailed image concept for the Vizora launch.
Modern SaaS hero image. Professional.
Koko is my content agent. It runs locally via OpenClaw. I wasn't expecting much — I thought I'd get a vague paragraph and have to do the real work myself.
This is what came back:
![Koko agent in #koko-content Discord channel responding with a full image concept for Vizora launch — style, scene, color palette, key elements, dimensions, and tools listed]
Read that response.
It didn't just say "make it blue and professional." It gave me:
A specific scene — messy database diagram on the left transitioning into clean documentation on the right
An exact color palette — blue-to-purple gradient with a rationale (tech, trustworthy)
Key elements with layout logic — the central arrow showing transformation, where the logo sits, what the overlay text should say
Exact dimensions — 1792x1024, landscape, with the reasoning (social media headers)
A list of specific tools to actually build it — Canva template name included, Midjourney, Adobe Firefly, even a Python/OpenAI Images API prompt if I wanted to generate it myself
That last part. It wrote the generation prompt for me.
A sleek modern SaaS landing page hero image for Vizora,
a database documentation tool. Abstract glowing database
schema with interconnected nodes and tables transforming
into a clean interface. Blue and purple gradient background,
tech aesthetic, professional, 8k resolution.
I copied that prompt. Pasted it into Midjourney. Had a usable hero image in four minutes.
The thing I keep noticing about Koko specifically is that it doesn't just complete the task — it thinks about downstream use. The dimensions came with an explanation. The tool list came with a ranked order. The prompt came pre-formatted for the API.
I didn't ask for any of that. It just included it.
For context on the system: APES OS is a local multi-agent setup I built on OpenClaw. Seven agents, each with one job, each with their own Discord channel and a workspace folder containing a SOUL.md file that defines who they are and what they hand off to others.
Koko's job is content. Blog posts, newsletters, creative briefs, copy. The #koko-content channel is where I drop anything that needs writing or concepting. The response comes back in Discord and gets logged to Supabase.
The whole stack runs locally. Free OpenRouter model tier. Supabase free tier. Discord is free. OpenClaw is open source.
I shipped Vizora's launch post with that hero image. The image concept Koko wrote became the actual brief.
That's a weird sentence to type. My AI agent wrote the creative brief for my product launch. But that's what happened.
If you want the full setup — all seven agents, the Discord routing, the Supabase schema, the SOUL files — I documented it here: nexflowai.gumroad.com/l/npzufj
Or poke around the open source side first at docs.openclaw.ai.
Drop a comment if you want to know more about how Koko's SOUL file is structured — that's what shapes the quality of the output.
