Build AI Image Processing Agent for Transparent PNG Asset Extraction

We are looking for an AI image processing / computer vision developer to build a practical AI image extraction agent/workflow for preparing website visual assets. This is not simply a background-removal job and not just a basic automation. We need a repeatable system that can help identify, extract, clean, organize, and quality-check visual assets from website section images. We design website page mockups and page sections. These pages are later implemented in Wix by a Wix expert. To make the Wix implementation high quality, we need to provide the Wix expert with separate visual assets rather than flat full-page mockup images. These assets may include: buttons icons decorative ornaments banners frames badges character cutouts product cutouts section decorations small illustrations other reusable website design elements The final outputs must be clean transparent PNGs with real transparency. The most important requirement: The exported PNG files must have a proper alpha channel / real transparency. No baked-in white background. No baked-in gray checkerboard background. No fake transparency. No messy halos or dirty edges. Preferred Workflow We would like the system to work like this: We place website section images into an input folder, ideally in Google Drive. The agent/workflow processes those images. It identifies separate visual assets inside each section image. It extracts each asset individually. It removes the background properly. It exports each item as a clean transparent PNG. It places the finished files into an organized output folder. It creates a simple report/manifest showing what was extracted. Example input: page01_section01.png page01_section02.png page01_section03.png Example output: page01_section01_button_01.png page01_section01_icon_star_01.png page01_section01_ornament_left_01.png page01_section02_character_01.png page01_section03_badge_01.png Secondary Workflow Sometimes we may already have smaller cropped asset images. In those cases, the system should also be able to clean those images and remove any baked-in background, including fake checkerboard backgrounds, and export proper final transparent PNGs. What the System Should Be Able to Do The system should be able to: process images from an input folder detect or help identify individual visual elements extract visual elements separately remove backgrounds cleanly output real transparent PNG files preserve image quality as much as possible organize output files by page/section use clear file naming create a report/manifest of extracted assets flag uncertain or imperfect extractions for manual review be reusable for future website pages be simple enough for our team to run repeatedly Where the system is not confident, we prefer that it flags the result for review instead of pretending the output is perfect. Possible Tools You may use the tools you think are best, such as: Python OpenCV segmentation models background removal models AI image processing tools Google Drive automation Google Apps Script local scripts simple web interface other practical AI/image processing methods We are open to your recommended approach, but we do not want an overbuilt enterprise system. We want a practical, reliable workflow that solves this specific problem. Communication Requirements Good communication is very important for us. Please only apply if you are willing to: join Zoom calls when needed communicate clearly and frequently provide regular updates preferably use WeChat for faster communication if possible explain your process clearly be available for discussion during setup and testing Before we move forward, we will need a short call to confirm that you understand the project. Test / Proof of Ability Before committing to the full project, we need to confirm that you can actually produce clean transparent PNG extractions. Please be prepared to show either: relevant previous examples, or a small sample extraction using one of our provided images We need to see that the output has real transparency, clean edges, and no baked-in background. Final Delivery Must Include Final delivery should include: working image extraction agent/workflow organized input/output folder structure transparent PNG outputs report/manifest format written setup and usage instructions screen-recorded walkthrough video showing how to run the system troubleshooting notes all code/scripts/prompts/workflows transferred to us confirmation that everything is built in or transferred to our accounts confirmation that we own the final system Please do not build this only inside your private account. We need access to the full workflow and files. Timeline In your proposal, please provide your estimated timeline for: reviewing the requirements testing with a sample image building the first working version making improvements after review final handoff and documentation Long-Term Potential If this works well, this could become long-term work. We have many website pages, visual assets, product images, marketing images, and other creative workflows that may need similar extraction, cleanup, and automation support. Please Do Not Send a Generic Proposal To be considered, please reply with: The words “Transparent PNG Agent” at the top of your proposal. What technical approach/tools you would use. Whether you have built anything similar before. How you would ensure the PNGs have real transparency and no baked-in white/checkerboard background. How you would identify and separate individual visual elements from a section image. How you would handle uncertain or imperfect extractions. Your estimated timeline. What you would need from us to start. Whether you are willing to join Zoom calls. Whether you are able/willing to communicate through WeChat if needed. Whether you can provide a walkthrough video and handoff documentation. Generic replies such as “I can do this” or “I am interested” will not be considered. We are looking for someone practical, detail-oriented, and technically capable of building a clean, repeatable image extraction agent/workflow — not just someone who manually removes backgrounds one by one.