First Indian Company to Adapt
Chat GPT 4.0
Creating 2000 videos in 3 months
Text - Video Generator
Bhanzu
NOV 2023 - DEC 2023
A.I Video Generator
A hackathon project, which received the winning award, emerged as a potential product to assist our marketing team. This project, leveraging Chat GPT 4.0 AI (text-based), was developed to generate solutions for math-related questions using our unique teaching methods. Given the resource-intensive nature of our marketing team's task of posting math-related short clips online, we opted to leverage the hackathon project. We enhanced its capabilities from "text-to-text" to "text-to-image" and "text-to-audio", enabling the creation of videos using our proprietary database and distinctive teaching techniques.
Role
Product designer
Responsibilities
System design
User interface design
Research
Team
Designer (Myself)
Product manager
Engineer
Curriculum team
2 Months duration
2000+
Unique videos created within first 3 months
New teaching techniques
A.I starting suggesting new content creations
Cost reduction
Reduction in manual labor costs resulted in the best ROI
The marketing team struggled to create math-related short clips online, which demanded a lot of resources, time, and effort to maintain quality. To streamline this process, we aimed to build an AI system that could generate math solutions aligning with our teaching methods.
Our Goal 🎯
Key metrics
Reduce man power and cost to company
Create efficiency and uniqueness to content
Mass production of quality content in shorter time period
What we have to achieve? 🏆
Where it all began? 🎬
Buzz Buddy
World’s first A.I math bot to teach math problems step by step
Will this project yield results? 🤔
Qualitative research, field research, competitor research
Let’s ask the users 🕵🏻
To ensure the application's viability among video editors, we reached out to various professionals, gathering feedback on usability and productivity within tight deadlines. While popular tools like After Effects, Premiere Pro, and DaVinci Resolve dominate the industry, we aimed to create a product catering to diverse sectors, focusing on simplicity while leveraging our AI technology.
(User Interviews, Google form surveys, Linked in polls
Questions
Starting with open ended question and directing users to close ended questions
How video editors manage tight deadlines ?
How do they maintain quality with the content while also having a high production rate ?
How do they come up with unique content content during mass production?
Introducing a basic wireframe to ask users whether the app will increase their productivity
#Deadlines
#Quality content
#Creativity
#Ease of explanation
#Efficiency
Our current backend requires users to provide a prompt and select from over 50 values to create a video, which is unconventional for an AI-backed application. As designers, we strongly oppose this feature. We understand that users are unlikely to use our application if they must input numerous values before creating a video. Instead, they are more inclined to opt for applications like After Effects, Premiere Pro, or DaVinci Resolve for content creation.
✋🏻Wait, there’s a catch!
What we found out? 👁️🗨️
Major findings from all forms of survey
Compromising content for quality
Maintaining quality alongside a high production rate is challenging, often resulting in compromised content due to time constraints
Maintaining consistent quality across thousands of hours of video content requires a lot of time and work.
Compromising quality for cost
Generating unique content during mass production is difficult, leading to reliance on familiar formats and ideas
Marketing teams frequently prioritize creativity over timely hiring, resulting in missed opportunities to onboard the most suitable talent when needed
Feature change
While we introduced the features, users were not intrigued due to the substantial workload required before obtaining the final video output
Difficult to scale
The creation of a video includes many steps and stakeholders, requiring a lot of effort to make content creation scaleable.
Startup struggles
Video editors struggle to meet tight deadlines in startups due to the pressure to deliver quickly with limited resources
A glance at our competitors 🥇🥈🥉
Understanding the common practices utilised in designing video editing software and integrating AI within it.
Synthesia
AI-generated video with human-like presenters from scratch
Not ideal for teaching
Runway
Removing video background and objects without a green screen
Not beginner-friendly
Descript
Automatic transcription and voice dubbing
Imperfect transcriptions
Wondershare Filmora
Motion tracking
Steep learning curve
Fliki
Blog to video creation
Limited functionality
Peech
Marketing videos
Not ideal for teaching
Opus Clip
Social media videos
Requires existing video clips
A.I Video editing tools
Best for
Limitations
Most AI video editing apps lack the ability to perform step-by-step script edits or replacements, indicating a deficiency in teaching capabilities
There are very limited functionalities available when it comes to editing the video after its creation
There is a lack of functionality for creating animations using scripts
There is a pattern 🧬
How does this A.I work? 🕵🏻
We explored using AI, specifically Chat-GPT 4.0, to enhance our product. Engineers integrated the API, aligning the AI with our math teaching style. Research led to discovering "Private AI (Closed loop)" tech like LLM, which works offline. Platforms like Ollama.com offer LLMs like Llama 2 for local installation, empowering users with basic computer skills to train AI as personal assistants using their data.
Diverging into the world of AI
BigQuery ML
BigQuery ML democratises the use of ML and AI by empowering data analysts, the primary data warehouse users, to build and run models using existing business intelligence tools and spreadsheets. Predictive analytics can guide business decision-making across the organisation
Our Tech-stack 👾
Technologies we used
Neural network
A neural network is a method in artificial intelligence that teaches computers to process data in a way that is inspired by the human brain. It’s a type of machine learning process, called deep learning, that uses interconnected nodes or neuron's in a layered structure that resembles the human brain
Neelakantha Bhanu Prakash-CEO
(World’s fastest human calculator)
Neural network
Yes, there will be more than 100 editing parameters 🥳
Product schema
"durations": {
"fill_duration": 1.0,
"twoD_duration": {
"fadein_duration": 1.0,
"shape_duration": 1.0,
"text_duration": 1.0,
"move_along_duration": 1.0
"type": "shapes",
"color_enable": false,
"image_path": "multiply",
"image_applicable": true,
"move_along": true,
"move_along_shape": "1",
"three_d_shapes_applicable": false,
"shape_type": "7",
"sides": 1,
"custom_shape": {
"line_inputs":{
"start":0,
"end":10,
"step":2
"transform_3d_duration": {
"transform_duration": 1.0,
"arrange_duration": 1.0,
"rotate_duration": 1.0
},
"transform_2d_duration": 1.0,
"multiple_shapes_durations": {
"fadein_duration": 1.0,
"shape_duration": 1.0,
"text_duration": 1.0,
"arrange_duration": 1.0,
"downward_arrange_duration": 1.0,
"left_arrange_duration": 1.0,
"grid_arrange_duration": 1.0
"threeDDurations": {
"arrange_duration": 1.0,
"shape_duration": 1.0,
"rotate_duration": 1.0,
"rotate_duration_x": 1.0,
"rotate_duration_y": 1.0,
"rotate_duration_z": 1.0
"text": "Hello World!",
"arrangement_type": "1",
"image_multiple_factor": 0.05,
"durations": {
"fill_duration": 1.0,
"twoD_duration": {
"fadein_duration": 1.0,
"shape_duration": 1.0,
"text_duration": 1.0,
"move_along_duration": 1.0
"transform_3d_duration": {
"transform_duration": 1.0,
"arrange_duration": 1.0,
"rotate_duration": 1.0
"divide_shape_enable": true,
"num_of_parts": 3,
"part_color": "3",
"fill_shape_enable": false,
"fill_shape": "1",
"distance_to_center": 1,
"fill_multiple_factor": 1,
"angle_of_triangle": 20,
"outline_color": "1",
"fill_color": "1",
"threeDDurations": {
"arrange_duration": 1.0,
"shape_duration": 1.0,
"rotate_duration": 1.0,
"rotate_duration_x": 1.0,
"rotate_duration_y": 1.0,
"rotate_duration_z": 1.0
"transform_applicable": false,
"transform_to_threeD": false,
"transform_multiple_factor": 1,
"number_of_transformed_shapes": 2,
"transformed_shape_type": "1",
"transformed_outline_color": "1",
"transformed_fill_color": "1",
"number_of_shapes": 1,
"position_input": "1",
"selected_direction": "1",
"transform_2d_duration": 1.0,
"multiple_shapes_durations": {
"fadein_duration": 1.0,
"shape_duration": 1.0,
"text_duration": 1.0,
"arrange_duration": 1.0,
"downward_arrange_duration": 1.0,
"left_arrange_duration": 1.0,
"grid_arrange_duration": 1.0
We got some insights 🔍
Understanding and uncovering the myth
Complexity of Video Creation Process: Users expressed concerns about the substantial workload involved in creating videos, highlighting the numerous steps and stakeholders involved. This indicates that making content creation scalable requires significant effort and resources. The tech work flow needs to be modified
Challenges in Meeting Tight Deadlines: Video editors face challenges in meeting tight deadlines, especially in startup environments where there's pressure to deliver quickly with limited resources. This suggests the need for more efficient workflows and tools to streamline the production process.
Balancing Quality and Production Rate: Maintaining quality alongside a high production rate is a significant challenge for video editors. The time constraints often lead to compromised content, highlighting the importance of finding a balance between speed and quality.
Time-Intensive Quality Assurance: Ensuring consistent quality across extensive video content requires a considerable amount of time and effort. This underscores the importance of investing resources in quality assurance processes to uphold standards.
Talent Acquisition Challenges: Marketing teams tend to prioritize creativity over timely hiring, leading to missed opportunities to onboard suitable talent when needed. This indicates the importance of aligning hiring practices with project timelines to ensure timely delivery.
Struggles with Generating Unique Content: Generating unique content during mass production is a common challenge, resulting in reliance on familiar formats and ideas. This highlights the need for innovation and creativity to break away from repetitive content creation patterns.
Deciding based on our understanding 🎯
The research findings indicate that developing this product will indeed yield productivity and innovation, ultimately resulting in the best ROI
We leveraged Chat GPT 4.0 AI technology, renowned for its text-based capabilities, to develop a solution that automates the creation of math-related content. Our system generates solutions for math-related questions using our proprietary teaching methods, thereby ensuring the content's relevance and educational value.
To address the resource-intensive nature of the marketing team's task, we enhanced the system's capabilities from "text-to-text" to "text-to-image" and "text-to-audio." This upgrade allowed us to create engaging videos incorporating visual and auditory elements, thereby enhancing the content's appeal and effectiveness.
Integrating AI inspired by an AI library (Ollama) to enhance editing capabilities.
Introducing over 100 editing options, including custom vector shapes and a pen tool (Under construction as of APR 2024).
Allowing breakdown of scripts and editing or replacement of each line.
Enabling creation of custom animations using scripts and adjustment of their values.
Future versions will include built-in plugins such as Unsplash, Vecteazy illustrations, and a background remover.
Providing the ability to edit videos anytime post-creation, facilitating adjustments for future curriculum changes.
Final design UI
UI-Design
Users will now enjoy the ability to edit videos seamlessly, incorporating custom vector shapes, and much more
Introducing more than 100 editing capabilities
The data behind our AI is built according to our math curriculum.
Users have to set up the basic information from which the data is pulled.
Grade group, module, and topic are managed by the curriculum team.
The duration of the video type is adjusted according to media platforms.
Pre-requisites might be required for some topics, and users will have to choose accordingly.
Once the context is set up, users can proceed to "Generate script."
Provide AI with a context
Once the details are provided, users must provide a prompt based on the suggestions from the curriculum team.
The prompt is then fed to the AI, which breaks down the entire story into 3-7 groups.
Each group includes narration and animation, both of which are created using a script and can be modified for better results.
Custom narration and animation can be added to provide a unique touch to the video output.
Generate and breakdown scripts
As the user creates the animation using a script, the values utilised for creating the video are extracted and displayed to the user.
Users can then experiment with over 100 editing options, adjusting them down to milliseconds.
Additionally, they can incorporate custom shapes and graphs into their animations.
The pen tool, essential for creating custom shapes, is currently under development for Version 2.0.
Export the file or directly post on social media
Over 100+ editing options
Implementation of timeline now gives the user to edit that millisecond twist
Audio frequency shown in decibel level and the sound levels can be tweaked
Merge between groups to make the video more unique
Create that millisecond touch
View, share & publish reels right after creation
Our journey from hackathon project to AI-powered marketing solution exemplifies our commitment to innovation and excellence. By harnessing the power of AI technology, we have revolutionised our marketing efforts, paving the way for more efficient, engaging, and effective content creation processes.
At last 😇
The implementation of our AI-powered math solutions project resulted in a significant transformation of our marketing efforts. By automating the content creation process, we achieved:
Efficiency: The time and resources required to produce math-related content were drastically reduced, allowing the marketing team to focus on other strategic initiatives.
Complex concepts are explained in the easiest form resulting into at least 25% improvement in student engagement
Engagement: The introduction of multimedia elements such as images and audio enhanced the content's engagement levels, leading to increased viewer interest and interaction.
Effectiveness: The use of our unique teaching methods ensured that the generated content was not only informative but also aligned with our educational objectives, resulting in improved learning outcomes for our audience.
Improved time to market: With A.I reels, Bhanzu was able to cut down video production from several months down to 2 weeks.
Happy students: First surveys showed great results in terms of student satisfaction since its implementation.
Video at scale: Since adding A.I reels to the content creation toolbox, the team has implemented AI video in around 50 courses in less than a 3 months.
Flexibility: once created, a video can be easily updated at any time, without having to start from the beginning.
Localisation: we can create videos in over 120 languages and accents, enabling you to expand your reach to new countries without leaving your desk
The results below are from version 1.0 (MVP)
What we achieved in 3 months 🎯 ?
Building on the success of our AI-powered math solutions project, we aim to further refine and optimise the system's capabilities. Future iterations may include enhancements to the AI algorithms, integration with additional multimedia formats, Introducing to teachers for creating content and expansion to cater to a broader range of educational topics.
Our Future plans 🔮
To find a balance between user needs, tech-feasibility and business needs
To make decisions which make positive impact on all sides
To collaborate with tech, product, curriculum, marketing and other stakeholders
To build a versatile MVP in a shorter time period
To understand AI and machine learning
The limitations of technology and understand a product schema
This project taught me
Personal take aways from this project 🤠
Inventory of reels created with existing curriculum
