Axis Vapix
LANCER
Lancer :
Ideal for quick, focused engagements
Team: ≤ 2 resources
Duration: ≤ 3 weeks
Indicative Budget: ≤ $3,000 USD*
Use Cases: MVPs, prototypes, simple websites, bug-fix sprints
Indicative Pricing: ~$15–$25/hr per resource*
Growth :
Best suited for growing products or structured builds
Team: 2–5 resources
Duration: Up to 4 months
Indicative Budget: Up to $30,000 USD*
Use Cases: Feature-rich web apps, integrations, custom dashboards
Indicative Pricing: ~$15–$35/hr per resource*
Enterprise :
Tailored for high-scale, complex or long-term initiatives
Team: 5+ resources
Duration: More than 4 months
Indicative Budget: Starts from $30,000+ USD*
Use Cases: End-to-end platforms, SaaS products, compliance-heavy systems
Indicative Pricing: ~$25–$50/hr depending on roles*
(Development)
Project Description
An AI-first media startup experimenting in synthetic content creation partnered with Oodles to build a minimal video generation platform. The client aimed to replicate talking-head videos using deepfake technology, driven by synthetic audio from AWS Polly. They sought a backend MVP to run experiments—without UI—focusing on real-time rendering precision and infrastructure readiness.
Scope Of Work
The project aimed to develop an experimental video rendering pipeline using deepfake models synced with AWS Polly speechmarks. The client required a backend MVP with the help of Oodles that processes a source video, aligns it with generated audio, and produces a lifelike talking-head output. Core areas of work included audio-visual synchronization, face animation, infrastructure setup, and system automation.
Our Solution
To help the client achieve their goal, a modular cloud-ready backend was architected, focused entirely on processing speed and deepfake accuracy.
Key Features Implemented:
- Facial Landmark Detection & Motion Transfer: Used pre-trained models (based on First Order Motion Model) to animate facial expressions from a static source video.
- Speechmark-Driven Syncing: Integrated AWS Polly to convert text to speech, and used its speechmarks to time mouth and facial movement.
- Video Processing Pipeline: Built using OpenCV and FFmpeg for frame-by-frame video generation and stitching.
- Cloud-Based Execution: Deployed on AWS EC2 with input/output handling via Flask endpoints, enabling remote API access.
- No-UI Architecture: Built intentionally without front-end layers, simplifying integration for ML experiments and rapid iterations.
Share Your Requirements