Coloring-page Generator (AI)

Idea

Coloring pages on specific themes (a “dragon with astronaut helmet”, “horse on the moon”, “favorite animal with a crown”) are popular but rarely available as stock images. Existing AI tools are usually cloud-bound, expensive per image, and hard to steer to the line-art format. This experiment shows how to do it locally.

Solution (local proof of concept)

A lean pipeline with a text prompt as input:

1. Prompt — user types e.g. “dragon with crown in front of castle”.
2. Style anchors added (coloring book line art, black and white, bold outlines, no shading, A4 portrait).
3. SDXL + line-art LoRA renders the image.
4. Post-processing — line sharpening, white-space cleanup, contrast.
5. PDF output, A4, borderless.

All local, no cloud upload.

Status today

• Local: runs, delivers good results on test images.
• Not deployed: no public demo yet. Planned next step: Hetzner GPU (GEX44, ~€200/month, reducible to €20-50/month effective with socket activation).
• No customers: this is a learning/showcase project, not a product for sale.

What this project shows

• AI integration without OpenAI — I built the full pipeline, not just an API wrapper.
• ControlNet understanding — preprocessing is where most quality comes from.
• Hetzner GPU economics — socket activation makes on-prem financially viable.
• Realistic scope of a solo project with AI assist.

Why interesting for you?

Because the same architecture — local workflow, no US cloud, predictable costs — works for other domains: document processing, contract analysis, patent search, internal knowledge bots with RAG.