This is a submission for the Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady
Have you ever asked someone to take your photo, only to find out that you look awkward in every single shot? 😅
Or tried recreating a pose from Instagram, taken 15 pictures, and still wondered why none of them looked quite right?
Sometimes, the hardest part of taking a photo isn't the camera. It's knowing where to stand, how to pose, and how to frame yourself.
That's the problem I wanted to solve.
Meet Pose Coach — your personal AI photography director. 📸
The idea is simple: what if your phone didn't just capture the moment, but actually helped you get ready for it?
What I Built
Pose Coach is a mobile-first photography assistant designed to help people pose confidently and take better photos without needing someone else behind the camera.
Instead of guessing where to stand or repeatedly retaking pictures, the goal is to let your phone guide you in real time.
✨ The experience I'm building
- 📷 Smart framing: Help users understand whether they need to move closer, step back, or adjust their position in the frame.
- 🕺 Pose guidance: Use a reference pose to help users understand body alignment and positioning.
- 👗 Traditional poses: Explore pose ideas for sarees, lehengas, kurtas, and other ethnic outfits.
- 🧥 Western poses: Discover casual, fashion, street-style, and portrait poses.
- 🤳 Selfie mode: Get help with face positioning, angles, and composition.
- 🫶 Couple photography: Explore two-person poses and improve spacing and framing.
- 🔒 Privacy-first AI: Build around on-device computer vision so personal camera footage doesn't need to be uploaded to a cloud service.
The project currently has a mobile-first interface, camera experience, and pose categories. The more advanced live AI coaching and pose-matching experience is still being developed and tested.
My goal is to make taking a good photo less about trial and error and more about confidence.
Code
💻 GitHub repository: ShraddhaNigam19/pose-coach
The project is built with a mobile-first web approach, using React and TypeScript.
I'm building the application around a practical combination of on-device computer vision and photography-specific logic, rather than relying on a chatbot to tell users generic posing tips.
Feel free to explore the repository and share suggestions or improvements!
How I Built It
The most interesting part of Pose Coach isn't the interface. It's the idea of turning what a camera sees into useful, understandable instructions.
The technology behind it
- React + TypeScript + Vite: Application structure and mobile-first UI.
- MediaPipe Pose Landmarker: The intended open-source computer vision foundation for detecting human body landmarks.
-
@mediapipe/tasks-vision: Browser-based pose estimation. - Camera API: Accessing the camera and capturing photos.
- Canvas overlays: Rendering pose references, detected landmarks, and framing guides.
- Framing engine: TypeScript geometry and rules that translate landmark positions into practical feedback.
- Local persistence: Saving preferences and favorite poses on the device.
The challenge: from landmarks to useful guidance
A pose model can estimate where body landmarks are located, but raw coordinates don't tell someone what to do next.
Imagine selecting a full-body pose while your feet are outside the camera frame.
Pose Coach should recognize that the composition doesn't meet the selected framing target and tell you to step back.
Or imagine trying to match a reference pose. Instead of placing a static stick figure over the camera and calling it AI, the application should distinguish the target pose from the user's detected skeleton, compare their alignment, and suggest meaningful corrections.
That's the engineering challenge I'm most interested in: turning computer vision into practical, real-time coaching.
Why Does Open Innovation Matter?
I want Pose Coach to be useful without requiring users to pay for a proprietary AI API every time they open their camera.
Open-source AI makes it possible to experiment with local inference, inspect the underlying implementation, adapt the processing pipeline, and build a privacy-conscious experience.
With an appropriate on-device model, camera frames can be processed locally instead of being sent to a remote AI service. That can also reduce network dependency and avoid per-request inference costs.
It gives developers room to improve the product in ways that matter to the application: better landmark handling, more useful framing rules, alternative models, and improved performance on mobile devices.
Of course, open-source doesn't automatically mean every model is lightweight, offline-ready, or licensed for every use. Model assets, runtime compatibility, licensing, and real-device performance all need to be verified.
For me, open innovation is about having the freedom to understand the technology, adapt it, and build something useful around it.
My Agent Session
I used an AI coding assistant during development to help implement and refine the application.
DevRelay session: Add your session link here if available.
I also learned that getting a polished interface on screen is only one part of building an AI product. The underlying model must work, its output must be reliable, and the features must respond to real input.
That's why I'm focusing on making the AI functionality genuinely useful instead of treating the model as a decorative add-on.
The Lazy Payoff 🦥
Imagine setting up your phone for a full-body photograph.
You take a picture. Your feet are cropped out.
You move the phone. Now there's too much empty space above your head.
You try again. Your pose still doesn't look right.
We've all been there. 😅
Pose Coach aims to reduce that cycle by guiding the user before they press the shutter.
The payoff I'm working toward is simple:
Less trial and error. Fewer retakes. More photos worth keeping.
I haven't measured the time saved yet, so I won't claim a specific improvement. My next milestone is to test the live framing workflow on real mobile devices and compare the experience with taking photos without guidance.
One Last Thought ğŸ’
I started with a simple question:
What if your phone could be your personal photography director?
A camera that helps you find the right position, understand your pose, and feel more confident in front of the lens.
That's the experience I want Pose Coach to deliver.
I'd love to hear from the community:
If your camera could give you one instruction before taking a photo, what would you want it to say?
Would it be "move back," "fix your posture," "try a different angle," or something else entirely?
Let's build more useful things with open-source AI.
Top comments (1)
This project is genuinely useful, especially because it works completely offline! 🙌
One question: can it automatically capture a photo once it detects that I’m in the right position and have matched the correct pose? That would make it even more amazing! 📸