Inspiration
We've all been in situations where we find a great photo on Pinterest or Instagram but have no idea how to recreate it. Getting the right angle, framing, lighting, and pose usually takes a lot of trial and error, especially when asking someone else to take the picture.
We also noticed the opposite problem: sometimes you're in a great location but don't know what kind of photos to take.
We wanted to make photography more intuitive by building an AI assistant that helps people find inspiration based on their surroundings and guides them through recreating the perfect shot.
What it does
Poser is an AI-powered photography assistant that helps users discover, plan, and recreate photos.
Users can upload a reference image, import inspiration from Pinterest or Instagram, or scan their surroundings to discover photo ideas that match their environment.
Poser analyzes reference images to understand composition, camera angles, subject positioning, poses, and lighting. It then translates that information into actionable instructions.
Through a live camera interface, Poser guides users on where to stand, how high to hold their phone, how to angle the camera, and how to position their subject. Visual overlays, composition grids, and pose guides help users align their shots with the reference.
Rather than just showing users what a good photo looks like, Poser helps them actually take it.
How we built it
Poser is designed as a mobile application using SwiftUI and AVFoundation for the camera interface.
Our technical approach combines several components:
- Vision-language models: Analyze uploaded reference images and environmental photos to identify composition, lighting, subject positioning, and camera perspective.
- Computer vision: Use MediaPipe Pose and OpenCV to detect subjects, estimate body landmarks, and compare the live camera feed against a reference.
- Scene-based inspiration: Use image embeddings and semantic similarity to retrieve photo references that match the user's surroundings.
- Real-time guidance: Combine camera frames, device orientation data, and geometric calculations to generate instructions for adjusting framing, angle, and subject placement.
- Visual overlays: Display composition grids, pose landmarks, and reference alignment guides directly over the camera preview.
The goal is to connect AI-powered image understanding with real-time camera feedback in a simple, intuitive experience.
Challenges we ran into
One of the biggest technical challenges is translating a static reference image into useful real-world camera instructions. A single photograph doesn't provide exact information about camera height, distance, focal length, or the original shooting environment.
Another challenge is providing accurate feedback in real time. Continuously processing camera frames with large AI models would introduce latency, so the system needs to balance more expensive image analysis with lightweight computer vision techniques.
Finding relevant inspiration based on a user's surroundings is also difficult. Two locations might look visually similar but require completely different camera positioning or lighting to recreate the same composition.
Finally, designing guidance that is genuinely helpful without overwhelming the camera interface requires careful consideration of the user experience.
Accomplishments that we're proud of
Our biggest goal with Poser is to make photography guidance actionable rather than purely informational.
The aspects of the project we're most excited about are:
- Connecting reference-image understanding with real-time camera guidance.
- Designing a workflow that supports both recreating existing photos and discovering new inspiration.
- Using computer vision to translate photography concepts like framing, perspective, and subject placement into understandable instructions.
- Making photography more accessible to people without technical photography experience.
- Bringing together computer vision, multimodal AI, and mobile development in a practical consumer application.
What we learned
Working on Poser has helped us explore how multimodal AI can interact with the physical world rather than simply analyze images.
We've learned about the challenges of extracting spatial information from 2D photographs, using pose estimation for real-time applications, and combining AI models with traditional computer vision techniques.
We've also learned that building a useful AI product isn't just about generating accurate predictions. It's about turning those predictions into clear, actionable feedback that users can understand immediately.
What's next for Poser
We envision Poser becoming a complete AI photography companion.
Some features we'd like to explore include:
- Personalized video tutorials: Generate short visual demonstrations showing how to position the camera and subject.
- Smarter inspiration discovery: Find photo references based on location, surroundings, lighting, and personal aesthetic preferences.
- Real-time pose coaching: Provide more detailed guidance on body positioning, facial orientation, and group photos.
- Automatic capture: Take the photo when the framing and composition closely match the reference.
- Photography education: Teach users composition, lighting, and perspective through interactive guidance.
- Creator tools: Help content creators reproduce consistent visual styles across photos and videos.
Ultimately, we want Poser to make taking a great photo as simple as finding one you like.
