Integration of Mainstream Open-Source AIs

Compnay Name

INCUBIQ

Service Details

We will interact with the community for validating the final list of AIs that we will integrate, and for some basic minimum integration requirements.

Note: the Stable Diffusion AI (category image processing) pre-selection was dropped from this proposal after having received some community comments. Instead, it is proposed in Pool B as a smaller funding proposal for upgrading the existing service making use of Stable Diffusion.
See https://proposals.deepfunding.ai/collaborate/95bc667c-e3e6-4235-a595-5c455564c6c0

Risk and mitigation

To further ensure that we meet the project deadline, we have implemented a 15% buffer on each work package. This allows us to aim for a maximum delivery time of 2.5 months, with the possibility of completing the project within a more compressed duration of 1.5 months if all goes smoothly. By taking these precautions, we have effectively mitigated any risk of missing the deadline.

In terms of technical implementation risks, here is what we anticipate, and how we will mitigate them.

• One potential risk for this project is the integration of a first ping base service within SingularityNet, as it will be a new undertaking for us. However, the extensive documentation provided by SingularityNet (
  https://dev.singularitynet.io/tutorials/python/

  ) suggests that this risk may be very low, as it provides clear instructions on how to build this service.

• Another risk is the potential difficulty in matching the dockerized AI with various GPU hardware. While we have previous experience in building virtualization for NVIDIA hardware such as RTX 3090 on linux OS (as demonstrated in the TaChiKu project), it is uncertain at this time if we will need to create multiple virtual environments to support a range of hardware. To mitigate this risk, we will continuously monitor the situation and explore all available options, including the possibility of limiting hardware support as a last resort.
• Finally, there is a potential risk that some AIs may be challenging to virtualize due to compatibility issues with Python libraries, which could potentially extend the deadline. However, this risk has only been present in a small number of the dozen AIs we have tested thus far. In the event that this risk does arise, we believe it can be easily mitigated by selecting AIs that are less prone to conflicting libraries. Ultimately, we feel confident that this risk will not prevent the successful delivery of the project.

While all technology projects carry some level of risk, we believe that this particular project has a low risk profile and we are well-equipped to manage any potential risks. We have the resources and expertise to address any issues that may arise, and we are open to seeking assistance from the community if necessary. By proactively identifying and addressing potential risks, we are confident in our ability to successfully deliver this project.

Business Model

Our business model is to offer free API testing, similar to what is available on Google Collab. Small tests will be available at no cost, but for those who require significant GPU power, there will be a fee (which at this stage is still to be fully determined). We believe that the initial 25% "hosting cost", combined with revenue from paid API calls, will easily cover the cost of offering free testing for ever.

Our goal is to attract users to the SingularityNet platform by offering a simpler and superior alternative to Collab. Therefore, we see those 25% "hosting" of grant funding as a "marketing budget". It will indeed fund hardware and hosting expenses, but will be used as a great incentive to onboard more developers onto the SingularityNet platform.

Problem Description

Our goal is to significantly reduce the need for specialized knowledge, time, and cost associated with integrating popular open-source AI libraries into applications. Currently, developers often face a lengthy and costly process that involves configuring environments, virtualizing, integrating, and running the libraries on appropriate hardware, just to test them out. We want to streamline this process to facilitate the development of innovative applications.

In addition to simplifying the integration process, our proposal also addresses the issue of limited AI options within SingularityNet. By making it easier to bring mainstream AIs onto the platform, we can expand the available selection and provide an alternative to using Google Colab notebooks. This not only benefits developers seeking to use a wider range of AI tools, but it also helps to drive innovation within SingularityNet by providing access to a greater diversity of AI capabilities.

Solution Description

We will build one SingularityNet service per popular open-source AI that we integrate. Each service will provide clean, documented APIs and full virtualization (using Docker) of the selected AIs, allowing developers to easily access and integrate AIs in their projects through a SingularityNet access token.

Each AI will be dockerised and made accessible via REST APIs. Our goal is to integrate 6 top class AIs in the field of image processing, video processing, speech, music, and/or language. At this stage, we have pre-selected those listed below, with a preference to integrate those marked with an asterisk (the selection), but we are open to change of priorities and/or integrate others, so long as they are open-sources, can be virtualised, and are aligned with the values and ethics of SingularityNet. The final selection will be submitted to a community vote.

The list of current pre-selected and selected (*) AIs is:

- For image processing:

• * VQGAN-CLIP (
  https://github.com/nerdyrodent/VQGAN-CLIP

  )

• * Latent Diffusion (
  https://github.com/CompVis/latent-diffusion

  )

• * Super Resolution (
  https://github.com/idealo/image-super-resolution

  )

• Toonify (
  https://github.com/justinpinkney/toonify
   

  )

• Image colorization (
  https://github.com/baird-ai/image-colorization

  )

• Extend Image content (
  https://github.com/CompVis/imagebart

  )

- For video processing:

• * Face swap (
  https://github.com/ai-forever/ghost

  )

• * Make it Talk (
  https://github.com/yzhou359/MakeItTalk

  )

- For voice/music processing:

• * Real-time voice cloning (
  https://github.com/CorentinJ/Real-Time-Voice-Cloning

  )

• OpeAI Jukebox (
  https://github.com/openai/jukebox

  )

• OpenAI Whisper (
  https://github.com/openai/whisper

  )

- For text processing:

• Sentiment Analysis: (https://github.com/BenWiseman/sentiment.ai)

Milestone & Budget

Total cost = $39,500

Each work package, from 1 to 4, will be delivered sequentially, package 1 first, and package 4 last. The project is expected to be completed within 9 weeks. We are adding a 15% risk for slippage, therefore a final total of around 10.5 weeks, or around 2.5 months. If the decision to award the grant is given early March, as per our current understanding, it would mean a commitment to deliver the full project by end of May 2023.

25% hosting cost $9,875

Service License Info

Yes, the project will take open-source AI, and the configuration, virtualisation, documentation, and integration will also be released open-source.

Revenue Sharing

We will onboard our open-source AI access services on the platform. If the services cross the threshold of $1,000 revenue per month, 10% of the additional revenue will be fed back into the SNET/DeepFunding wallets. This condition will remain valid for 10 years after first onboarding the service and will be applicable to this service or any subsequent iteration of this service on the platform.

Marketing & Competition

Just having a look at how Google Collab notebooks have taken mainstream adoption, it is clear that if SingularityNet were to offer a similar safe environment for rapidly testing AIs in sandboxes, and then rapidly integrating the chosen and configured AIs into apps, it would be a tremendous value proposition and would greatly boost the profile of SingularityNet. The simple integration of major AIs into SingularityNet is a first necessary step towards this goal.

To help boost the profile, we will also deliver at least one high quality marketing video showcasing the integration of FaceSwap into SingularityNet.

Related Links

Those links below are provided as example of what can be expected from us in terms of delivery and quality:

API standard and documentation:

Sample REST API doc produced for Cardano fund 7: https://tachiku.com/doc/

Open source repo and code quality:

Sample open source repo for Cardano fund 8: https://github.com/incubiq/sign_in_with_wallet

Monthly video report on project progression (here was the final video report):

Sample video reports of what was delivered for Cardano (TaChiKu): https://youtu.be/OPSa7qocXQc

Long Description

I am seeking a grant to build a SingularityNet service that will make it easier for developers to use popular open-source AI libraries in their projects. Integrating those libraries can be time-consuming and costly, as it requires configuring environments, virtualizing the AI, building clean APIs, let alone running the AI on powerful enough machines. I experienced this firsthand when I received a small grant from Cardano Fund 7 (category Singularity Net) to use an AI for morphing user profile pictures and minting the result as an NFT on the Cardano blockchain. I spent 20% of my project time on AI research and configuration alone.

With the SingularityNet service I propose to build, developers will be able to access and integrate mainstream open-source AIs with minimal effort. By providing clean, documented APIs and full virtualization (docker) of these AIs, we aim to make it possible for developers to use them within minutes, rather than spending days or weeks on configuration and integration. Our service will save time and reduce costs for developers, enabling them to focus on building innovative applications that address real-world problems.

We believe we are uniquely qualified to deliver this service, as we have first-hand experience with the challenges of integrating open-source AI libraries and a track record of successfully delivering projects on the Cardano blockchain (project TaChiKu, fund 7 - delivered as per the schedule)

AI Services