[Proposal] DOB Studio - AI Proposal

GPU Support Proposals
#1

Title: [Proposal] DOB Studio: AI-Based Video Transformation App Service

Author: Cho Young-jae / discord id : toscakarajan
Date posted: 2024/04/23

Summary

DOB Studio: https://www.dob.world
We are developing an AI-based video transformation service that automatically converts users’ videos into various unique styles such as cartoon, paper texture, and plaster texture. This technology is based on the latest AI algorithms and deep learning models and works by analyzing each frame of the video in real time to reconstruct it into a visual similar to the selected style.

Requirement

  • Description: Advanced GPU servers for AI model development
  • Resource type: H100 p5.x48xlarge
  • Amount: 8
  • Period: 2 months
  • Purpose of use: The machine will be used to provide a video-to-video service utilizing AI technology.

Background

In the era of the metaverse, humans are spending increasingly more time in the digital world. For Generation Z, who are digital natives, experiences on various digital platforms are as significant as those in the real world.

DOB Studio develops advanced AI graphics technology to provide realistic and enjoyable digital experiences that can substitute for the real world.

The virtual faces created by DOB Studio are faster and more precise than any existing technology, making them indistinguishable from real human faces.

DOB Studio envisions a fun world in the metaverse where everyone can experience a second or third life with any face they desire.

Scope of Work

The scope of this proposal primarily involves utilizing the requested GPU resources for massive video content training essential for the launch of AI services. Specifically, we plan to use Stable Diffusion technology to process and learn various styles and themes of video data. Currently, we are facing a shortage of GPU resources, which is expected to extend the model training period to over two months, significantly hindering project progress.

Securing additional computing resources through this program is expected to significantly reduce the training period, allow experimentation with more models, and produce higher quality results. This enhancement will expand the scope of AI services we plan to launch in the market and strengthen our competitiveness. Moreover, it will improve user experience and enable us to offer better services, ultimately maintaining our technological lead in the field.

Timeline

  • GPU Usage: Utilized for large-scale training of videos/images to create specific style ‘transformation filters’
  • Completed Transformation Filters: Once created, these filters are integrated into the app to transform users’ videos into the desired results (Video to Video).
  • Example of Video with Transformation Filter Applied: View Sample Video

Here’s a detailed 8-week schedule for the development and training of transformation filters:

Week 1

  • Generate training data for Style #1, Style #2, Style #3.

Week 2

  • Begin model training for Style #1, Style #2, Style #3 (1/3 of the process).
  • Conduct market research for Style #4, Style #5. Each style is expected to take 2-3 weeks.

Week 3

  • Continue model training for Style #1, Style #2, Style #3 (2/3 of the process).
  • Continue market research and prompt validation for Style #4, Style #5.

Week 4

  • Complete model training for Style #1, Style #2, Style #3 (final phase).
  • Generate training data for Style #4, Style #5.

Week 5

  • Modeling and stabilization for Style #1, Style #2, Style #3.
  • Begin model training for Style #4, Style #5 (1/3 of the process).
  • Conduct market research for Style #6, Style #7.

Week 6

  • Continue model training for Style #4, Style #5 (2/3 of the process).
  • Continue market research and prompt validation for Style #6, Style #7.
  • Generate training data for Style #6, Style #7.

Week 7

  • Complete model training for Style #4, Style #5.
  • Begin model training for Style #6, Style #7 (1/3 of the process).

Week 8

  • Continue and possibly complete model training for Style #6, Style #7 (2/3 to 3/3 of the process).

This schedule outlines a progressive and structured approach to developing and refining multiple style models, ensuring each step from data generation to model stabilization is carefully planned and executed.

Note: The timeline above can be adjusted based on the stability of the results from the current training of [Style #1, Style #2, Style #3]. For example, if stability is high, the goal for additional style training may be increased; if stability is insufficient, the retraining process may be repeated.

Benefits for Resource Providers (Runo Holders)

● Incentives for Node Participants and Early Users: We plan to promote the project and its sponsors extensively, particularly by offering vouchers for our service or similar benefits to AI network node participants, to secure an initial user base and foster ecosystem growth. These benefits will encourage participation and help collect initial feedback about the service.

  • Provision of 100,000 credits available within the app (* 1,500 credits provided to regular new users)
  • Equivalent to about 83 minutes of paid video conversion.

Contribution to AI Network Ecosystem

● AI-Related Workshops and Seminars In addition to generative AI tools, we will hold workshops and seminars on various topics related to the AI field to create a platform for knowledge sharing. Through these events, we will share the latest trends in AI technology, the technical background of our service, and use cases, thereby enhancing understanding of AI and fostering the exchange of new ideas.

  • DOB Studio is the inaugural chair company of the Korea Virtual Human Industry Association (KOVHIA). KOVHIA collaborates closely in four main areas: aligning with government policies, enhancing global competitiveness, supporting member companies, and expanding the industrial base. It actively supports MICE (Meetings, Incentives, Conferences, Exhibitions) organization and participation, investor and client networking, promotional support, and forms expert groups for business consulting. (Reference Article: 가상인간·버튜버가 함께 그리는 미래, 한국버추얼휴먼산업협회(KOVHIA) 출범 - 글로벌이코노믹)

  • (Planned) Overview of the seminar linked to the AWS GPU support program through AI Network
    Topic: Expert Perspectives on AI Technology Trends and an Introduction to the deVut.ai App Service
    Title: <DOB Studio, Talking AI in Everyday Life> May Coffee Chat
    Location: Uncommon Gallery (TBD)
    Attendance: Maximum 30 people (* 10-15 people from the creator and generative AI community are expected to attend)

● Promotion Through Social Media and Marketing Activities We will share information related to the project through various channels such as Instagram, X (formerly Twitter), and Discord, and increase interest in the AI ecosystem. These activities will not only target users interested in AI but also the general public, enhancing the visibility of the project and contributing to improving the social perception and acceptance of AI technology.

1. Press Release

  • Location: Uncommon Gallery (https://maps.app.goo.gl/JCW31pQHes5un9V2A)
  • Photos and videos (around 10 seconds) of representatives from both companies will be shot > transformed using devut.ai technology
  • Content production (using results transformed through the devut app) and press release content discussion
  • Additional material > Variations to be uploaded on official accounts of both companies on social media

2. Column article writing and contribution

  • CEO: Oh Jae-wook’s column contribution (Includes how the delivery of GPU resources from AI Network contributes to the deVut.ai app service)

Specification

Unique Value Delivered by the AI-Based Video Transformation App Service:

A Creative Tool for Creators:
Our developed app allows users to quickly and easily transform their video content in unique and creative ways. This expansion into new creative realms is expected to create synergistic effects across the industry.

Improved Accessibility to Generative AI Tools:
Our app features a simple operation mode where users only need to select the desired AI style and press transform. Compared to many generative AI tools that require prompt input to create images and videos, our app is being developed to be incredibly user-friendly, allowing even those without advanced video editing skills to easily create cartoon-style videos.

Time/Cost Convenience for Users:
Our service allows users to produce cartoon-style videos much faster and cheaper than professional video editing services or software, thus saving time and costs.

Goals

● GPU Usage KPI: Within 2 months, integrate 5 new AI transformation filters into the app.
● deVut.ai App Service KPI: Officially launch the OBT and secure initial test users within 2 months, followed by the collection of app feedback.

  • Test User Group (1): High-engagement group with generative AI tools (Verification of USP for Video to Video transformation quality and speed)
  • Test User Group (2): Reels/Shorts creators group (Immediate application and verification of convenience in account operation)

more detailed

  • deVut.ai App - OBT Launch: April (Android launch) / May (iOS launch)
  • Quantitative KPI: OBT KPI: 3,000 downloads / 300 feedback submissions
  • Qualitative KPI 1: From 3,000 downloads, analyze new user acquisition through tracking codes, group app user activities, and validate hypotheses about target groups.
  • Qualitative KPI 2: From 300 feedback submissions, prioritize enhancements for deVut.ai 2.0, and initiate development and updates.

Team members

● Cho Young-jae (Leader of the Core Research Team): Leads the entire process of app development and AI style transformation modeling.
● Kim Hyung-jun (Core Research Team): Handles practical AI style transformation modeling and system infrastructure setup.
● Lee Jung-min (Brand Communication Leader): Leads service planning and brand communication across all channels.
● Yoon Hyun-jin (Content): Responsible for content planning and production for app services and overall brand communication.
● Ji Hye-eun (Design): Leads app UI/UX, AI style creation tasks, and content design for overall brand communication.
● Kim Chae-young (Design): Manages generative AI prompter, AI style creation tasks, and content design for overall brand communication.