Build an AI Agent/Pipeline to Batch-Convert Location Photos into Watercolor Illustrations.

About Little Outings Little Outings is a curated family-activity iOS app for the Twin Cities metro area, covering roughly 840 parks, museums, breweries, farms, libraries, and other family-friendly venues. The app already uses a locked watercolor illustration style (originally developed in GPT) for its venue artwork. This project extends that same visual identity to a new set of reference photos. Project Overview We have a large batch of reference photos (sourced from the public internet) depicting places families would want to visit — parks, breweries, museums, farms, etc. We need a repeatable pipeline, not a one-off manual job, that can: • Pull source images in bulk from a Google Drive folder • Detect and remove any people from each image • Re-render each image in our established watercolor illustration style (style reference to be provided) • Output a clean, consistent, app-ready image per source photo, saved back to a designated Drive/output folder with a predictable naming convention This should be delivered as a reusable tool (script, agent, or lightweight app) that our team can re-run ourselves whenever new photos are added — not a manual delivery of finished images alone. Scope of Work 1. Bulk Ingestion • Connect to a specified Google Drive folder and pull all images (JPEG/PNG/HEIC) • Handle a batch size of ~800+ images without manual intervention per file • Skip/flag corrupted, duplicate, or unusably low-resolution source files 2. Person Removal • Detect people (adults and children) in each source photo • Remove them cleanly, with the background/scene plausibly reconstructed (inpainting) rather than leaving visible artifacts • Flag (rather than silently guess on) images where person-removal would materially distort the scene, for manual review 3. Watercolor Style Transfer • Convert each cleaned photo into a watercolor illustration consistent with our existing locked style (reference examples and Midjourney style parameters will be provided at kickoff) • Consistency across the batch matters more than any single image being a masterpiece — the whole set needs to look like it belongs to one app 4. Output & Delivery • Consistent output resolution/aspect ratio suitable for mobile app display (spec to be finalized with dev team) • Consistent file naming tied back to source filename or venue ID • Batch saved to a designated Google Drive output folder • A simple log/report of what was processed, flagged, or failed Deliverables • A working, documented pipeline/agent (script-based or lightweight tool) we can re-run ourselves on future batches • Source code with setup instructions (what APIs/keys/accounts are needed, how to run it) • A first full run of the current photo batch, delivered as finished watercolor images • Brief documentation covering: tools/models used, how style consistency is enforced, and how to adjust settings (e.g., watercolor intensity, person-removal sensitivity) Ideal Skills & Experience • Experience with image generation / style transfer APIs (e.g., Midjourney, Stable Diffusion + ControlNet, Flux, or comparable) and prompt/parameter tuning for visual consistency • Experience with inpainting or object-removal techniques (e.g., person/subject removal with background reconstruction) • Google Drive API experience for bulk file read/write • Comfortable building a small automation pipeline (Python preferred) rather than a manual, click-through workflow • Bonus: experience building lightweight agents/orchestration for creative/media batch-processing tasks What We'll Provide at Kickoff • Style reference examples (existing GPT-generated venue illustrations) • Access to the source Google Drive photo folder • A small test batch (~10–20 images) to align on quality before the full run • Point of contact for review/approval at each phase