This content originally appeared on DEV Community and was authored by Brennan Gerle
AEC Compliance Image Fixer
Submission for the Google AI Studio Multimodal Challenge.
An AI-powered web app that analyzes construction-site photos for PPE issues and instantly generates a safety‑compliant version.
Demo
Live app: https://aec-compliance-image-fixer-111886031714.us-west1.run.app/
Table of Contents
-
What I Built
- App Overview
- The Problem It Solves
- The Experience
-
How I Used Google AI Studio
- How Google AI Studio Was Leveraged
- Multimodal Capabilities Implemented
-
Multimodal Features
- What I Built (Editing Pipeline)
- Why It Enhances User Experience
Tech Notes
Built By
What I Built
App Overview
This application is an AI-powered tool for the Architecture, Engineering, and Construction (AEC) industry. It automatically detects PPE (Personal Protective Equipment) issues in site photos and can apply compliant fixes—aligned with common standards such as OSHA—directly to the image.
The Problem It Solves
Marketing assets, training manuals, and project documentation must depict strict adherence to safety protocols. A photo might be perfect except for a minor violation (e.g., missing safety glasses, a non‑compliant sticker on a hard hat, or the wrong helmet type). Traditionally, the fix required a reshoot or painstaking manual editing. This app provides a fast, cost‑effective alternative.
The Experience
Upload & Analyze
The user uploads a jobsite photo and clicks Analyze Image. A vision model scans for common PPE issues and returns a summary (e.g., “Missing safety glasses”, “Non‑standard helmet detected”), pre‑selecting recommended fixes in the UI.Fix & Download
The user accepts the AI’s suggestions or manually selects additional fixes (e.g., Remove hat stickers, Add safety gloves) or adds a custom instruction. Clicking Fix Image triggers a generative model that edits the photo. The final, compliant image is produced and ready to download.
In short: an intelligent assistant that identifies safety problems and corrects them instantly, saving time, effort, and resources.
How I Used Google AI Studio
How Google AI Studio Was Leveraged
- Rapid prototyping: Core prompts for both analysis and editing were iterated in AI Studio, enabling quick A/B testing across models and prompt strategies.
-
Simplified API access: The app calls Gemini models via an
/api-proxy/path; AI Studio securely manages the API key in the background—no keys in client code. - Integrated environment: Code editor, live preview, and model access in one place created a seamless loop from writing to testing the AI UX.
Multimodal Capabilities Implemented
- AI‑Powered Image Analysis (Vision Understanding)
-
Multimodal input: The worker photo (image) + a detailed text prompt are sent to the
gemini-2.5-flashmodel. - Task: Act as a safety expert; visually inspect for PPE issues (e.g., missing glasses, non‑standard helmets, stickers).
- Multimodal output: A structured JSON response containing (a) a human‑readable summary and (b) boolean flags for recommended fixes.
- AI‑Powered Image Editing (Generative Vision)
- Multimodal input: The original photo (image) + a text prompt constructed from selected fixes and any custom instructions.
-
Task: Use
gemini-2.5-flash-image-preview(editing‑specialized) to apply realistic edits that respect pose, lighting, and style. - Multimodal output: A new, edited image—safety‑compliant and ready for use.
Multimodal Features
What I Built (Editing Pipeline)
The core Fix Image functionality leverages gemini-2.5-flash-image-preview:
- Image data: The user’s original photograph.
- Dynamic prompt: Built from UI toggles (e.g., Add Safety Glasses, Remove Hat Stickers) plus any custom text. This tells the AI exactly what to change.
- Result: A new image that integrates the requested edits naturally and believably.
Why It Enhances User Experience
- Natural‑language control: No pro photo software needed; users work with checkboxes or plain English (e.g., “Add a high‑visibility safety vest”).
- Context‑aware realism: Edits account for head angle, scene lighting, and photographic style—no obvious “sticker‑paste” artifacts.
- Flexible & fast: Common fixes are one‑click; the custom prompt field supports edge cases and creative requests.
Tech Notes
- Platform: Built and hosted in Google AI Studio.
-
Models:
gemini-2.5-flash(analysis) andgemini-2.5-flash-image-preview(editing). -
Security: API access routed through AI Studio’s
/api-proxy/, keeping keys out of the app code.
Built By
Author: https://dev.to/beardedbe4n
This content originally appeared on DEV Community and was authored by Brennan Gerle
Brennan Gerle | Sciencx (2025-09-11T17:11:25+00:00) AEC Compliance Image Fixer. Retrieved from https://www.scien.cx/2025/09/11/aec-compliance-image-fixer/
Please log in to upload a file.
There are no updates yet.
Click the Upload button above to add an update.