Started a small merch biz, my monthly cost/return

Hey folks! Been a minute since I posted, usually hanging in the infra channels, but I've been deep in a side project that's a total departure from my usual Terraform loops. I started a small online merch store for a niche developer community I'm in, and I'm using **Leonardo AI** to generate most of the designs.

I figured some of you might be curious about the real numbers and workflow when using an AI image gen tool for a micro-business. This isn't a huge operation—it's just me, a Print-on-Demand (POD) service, and a Shopify Basic store.

Here's a rough breakdown of my monthly costs and returns after the first three months:

**Monthly Costs:**
* **Leonardo AI:** I'm on the **"Premium" plan at $30/month**. This gives me enough tokens for my volume (around 150-200 images a month). I mostly use Fine-Tuned Models (I trained one on a specific art style) and Alchemy.
* **Shopify:** $29/month for the basic plan.
* **POD Service Integration:** $10/month for the app subscription.
* **Product Costs:** This varies with sales, but as a cost of goods sold, it's roughly 40% of the sale price per item.

**Monthly Returns (Average):**
* **Gross Revenue:** ~$450
* **Net Profit (after all above costs):** ~$120

It's not quitting-my-day-job money, but it's a positive return. The big win for me is the speed. I can go from a concept in my head to a live product mock-up in under an hour.

My workflow is pretty systematic, almost like a CI/CD pipeline for t-shirts:

1. **Prompt Engineering:** I keep a structured prompt template in a note app to ensure brand consistency.
```text
Subject: [a cute robot, a minimalist mountain landscape]
Style: [vector graphic, screen print style]
Details: [clean lines, high contrast, no shading, on a white background]
Parameters: --style raw --ar 2:3
```
2. **Batch Generation:** I use Leonardo's prompt magic and alchemy to generate 10-15 variants of a concept in one go.
3. **Post-Processing:** Almost every image needs a quick pass through a raster graphic tool (like GIMP) to clean up edges, convert to true transparent PNGs, and optimize for the POD specs.
4. **Upload & List:** The POD app handles the rest.

The main pitfalls I've hit are the same as with any infra project: consistency and cost control. It's easy to burn through $30 of tokens in a day if you're not disciplined with your generations. Training a custom model was a game-changer for getting a cohesive look across all designs.

I'm curious if anyone else here is using Leonardo or Midjourney/DALL-E for a similar micro-SaaS or product biz. How do you manage your prompt libraries or handle asset versioning? Any tips on streamlining the image-to-mockup pipeline further?