AI Workflows that Accelerate Production & Respect Artists' Process (GDC 2024)

Invoke
15 Apr 202435:57

TLDRKint KY, CEO and founder of Invoke, discusses the role of generative AI in creative processes at GDC 2024. He emphasizes Invoke's commitment to empowering artists with control over AI generation, rather than replacing them. KY shares Invoke's vision for integrating AI into creative pipelines, including game studios and VFX studios, and highlights the importance of owning creative power. He demonstrates how AI can be fine-tuned to an artist's style, leading to more consistent and controllable results. The talk also covers the potential of AI in concept art generation, asset texturing, and the future of 3D model creation from 2D images. KY concludes with a call to view AI as a tool that respects human creativity and accelerates the creative process.

Takeaways

  • 🎨 **Generative AI Integration**: Invoke is an open-source interface designed to integrate generative AI into the creative process, aiming to accelerate the creative spirit of humanity.
  • 🤖 **Artist Empowerment**: The tool is built for creatives, focusing on giving artists control over the generation process and incorporating AI into their workflow without replacing them.
  • 📈 **Industry Adoption**: Game and VFX studios are increasingly incorporating generative AI into their pipelines, considering control and consistency as key factors for professional use.
  • 🛠️ **Creative Power Ownership**: Invoke emphasizes owning your creative power by training AI models on artists' styles and IP, creating a new visual language that they can control.
  • 🔍 **AI for Professionals**: Professional AI tools differ from mainstream by offering a more controlled and consistent output, trained on specific visual languages to meet professional standards.
  • 🚀 **Workflow Efficiency**: The ultimate goal is to use AI to reduce time spent on generating assets like character concept art, allowing for more varied and immersive game development.
  • 🎭 **Artistic Process as a Tool**: AI is treated as a part of the artistic process, not a replacement, with artists using it to iterate and enhance their work, maintaining their creative vision.
  • 📚 **Training AI Models**: Invoke offers both community-shared models and the option for individuals to train their models using their own assets, ensuring a personalized creative output.
  • 📉 **Legal and Ethical Considerations**: The platform respects copyright laws and encourages the use of clean, openly licensed data sets for training AI models.
  • 🔗 **Open Source and Community**: Invoke is part of the open-source ecosystem, offering flexibility and control to users, and discourages the use of non-commercial code.
  • 📱 **Future Developments**: The company is working on scalable solutions, including potential API integration for invoking specific workflows, and anticipates advancements in image-to-3D and image-to-video models.

Q & A

  • What is the main focus of Invoke as a company?

    -Invoke is focused on building tools for creatives and artists, specifically utilizing generative AI to accelerate the creative process without replacing the artist. They aim to give artists more control over the generation process and incorporate AI into their creative pipeline.

  • What does Kint KY mean by 'owning your creative power' in the context of AI tools?

    -Owning your creative power means that artists and creatives can incorporate their style, concepts, and characters into an AI model that they own. This involves training their content into the system, creating a new visual language that they can use, rather than contributing to a centralized model owned by someone else.

  • How does Invoke differentiate its AI models from mainstream AI tools?

    -Invoke differentiates its AI models by training them on the user's visual language and assets, providing more control and consistency. This is in contrast to mainstream AI tools that often generate repetitive patterns and lack the quality needed for professional use.

  • What are some of the considerations Invoke has when advising customers on incorporating AI tools?

    -Invoke considers where in the creative process the AI tool should be incorporated, how it can be used effectively without replacing the artist, and ensuring that the AI model understands the team's visual language and intellectual property.

  • How does Invoke's approach to AI in creative processes aim to save time for artists?

    -Invoke aims to save time for artists by reducing the time it takes to generate character concept art or other creative assets. This allows artists to explore more concepts and ideas, leading to a more varied and immersive final product.

  • What role does Invoke's interface play in deploying AI models to teams?

    -Invoke's interface allows users to codify and define their creative workflows using a visual editor. This enables teams to deploy AI models in a simple UI, making it easier for them to incorporate AI into their existing processes without the need for extensive retraining.

  • How does the process of training an AI model on an artist's style work?

    -The artist creates a dataset of their work and describes the commonalities among these images. The AI model is then trained on this dataset, learning to recognize the artist's style. When prompted, the model can generate new artwork in that style based on the artist's input.

  • What are the potential legal implications of using AI models trained on scraped internet data?

    -There are concerns about copyright infringement and the legality of using models trained on data without explicit permission. However, many legal scholars believe that training AI models may fall under fair use. To mitigate risks, Invoke is moving towards using clean, openly licensed datasets.

  • How does Invoke ensure that its AI models respect the creative process and the artists' originality?

    -Invoke ensures this by allowing artists to train models on their own assets and intellectual property. This way, the AI is an extension of the artist's creativity, generating new work that aligns with their unique style and vision.

  • What is the Invoke interface's approach to handling the iterative nature of creative work with AI?

    -Invoke's interface is designed to be iterative, allowing artists to make adjustments, composite elements, and modify outputs to fit their creative vision. It is not a one-size-fits-all solution but a flexible tool that adapts to the artist's process.

  • How does Invoke's workflow editor help in automating and streamlining the creative process?

    -The workflow editor in Invoke allows users to create automated processes for specific generation tasks, reducing manual and repetitive work. It connects various inputs and models to form a pipeline that can be saved and deployed for consistent results.

  • What are some of the creative applications Invoke is targeting with its AI technology?

    -Invoke is targeting applications such as concept exploration, asset texturing, seamless tiling and textures, 2D visual assets like icons and marketing materials, and eventually, 3D modeling and video generation.

Outlines

00:00

🎉 Introduction to Invoke and Generative AI

Kint KY, CEO of Invoke, introduces the topic of generative AI, noting a shift from hype to more thoughtful questions about its role in the creative process. Invoke is presented as an early open-source interface for generative AI, designed for creatives and artists to have more control over the generation process. The company's vision is to accelerate the creative spirit of humanity, and they emphasize the importance of owning creative power by training AI models with one's own content and style.

05:02

🚀 AI in the Creative Workflow

The paragraph discusses how generative AI can be incorporated into the creative workflow, emphasizing the need for control and consistency in professional settings. It contrasts mainstream AI tools, which often produce repetitive patterns, with professional AI tools that are trained on a team's visual language. The goal is to reduce the time taken to generate assets like character concept art, without compromising on quality, and to use AI as a tool that complements the artist's process rather than replacing it.

10:04

🤖 Training AI Models and Creative Process Customization

This section delves into the process of training AI models on an artist's style and how it can be integrated into the creative process. It highlights the ability to define where in the workflow AI is used and the importance of having a base model that understands the team's visual language. It also touches on the potential for 3D asset generation from 2D images and the iterative nature of using AI in the creative process.

15:04

🎨 Artistic Control and Iterative AI Image Generation

The speaker discusses the importance of artistic control when using AI for image generation. They emphasize that the process is not a one-size-fits-all solution but an iterative one where artists can set the fidelity to their sketches and have the freedom to modify outputs. The paragraph also mentions Invoke's tools for in-painting and the ability to focus the AI on certain regions for detailed generation.

20:05

🛠️ Workflows and AI-Assisted Creative Pipelines

The paragraph outlines various ways AI can be used in creative workflows, such as concept exploration, asset texturing, and generating 2D visual assets. It also mentions the potential for 3D model generation from 2D images and the anticipation of image-to-video models. The speaker stresses that AI is a tool to accelerate the creative pipeline rather than replace human creativity.

25:07

📚 Training AI on Art Style and Legal Considerations

The speaker addresses how to train AI on an artist's style, mentioning both individual efforts using open-source scripts and professional services for more extensive datasets. They also discuss legal considerations regarding the use of scraped art and the move towards using clean, openly licensed datasets. The paragraph emphasizes Invoke's commitment to respecting artists' rights and the importance of fair use in training AI models.

30:07

🌐 Open Source and Community Models

The final paragraph explores the open-source nature of Invoke, its distinction from other open-source projects, and its commitment to being commercial-friendly. It also mentions the availability of an API for local products and the potential for a workflow library that can be invoked through an API, catering to a broader range of use cases beyond just artists.

Mindmap

Keywords

💡Generative AI

Generative AI refers to artificial intelligence systems that are capable of creating new content, such as images, music, or text. In the context of the video, generative AI is used to produce artwork and accelerate the creative process for artists and designers. It is a technology that is no longer seen as 'magic' but is being thoughtfully integrated into creative workflows.

💡Invoke

Invoke is mentioned as an early open-source interface for utilizing generative AI on a local machine. It is designed to give artists more control over the AI generation process and to incorporate AI into their creative pipeline. Invoke is positioned as a tool that can accelerate the creative spirit of humanity while respecting the artist's process.

💡Control and Consistency

These are two categories of concerns that professionals have with mainstream AI models. Control refers to the ability to direct the AI to create specific styles or characters, while consistency means the AI can reliably produce similar outputs. In the video, it is emphasized that professionals require AI models that offer both control over the creative output and consistent results.

💡Fine-tuned Model

A fine-tuned model in the context of the video is an AI model that has been trained on specific assets or intellectual property (IP) to generate content in a particular style. This is contrasted with mainstream AI models that may not capture the desired style as effectively. Fine-tuning allows artists to create a new visual language that aligns with their unique creative vision.

💡Workflow Integration

Workflow integration involves incorporating AI tools into the existing creative process of a team or individual. The video discusses how AI can be deployed within a team's workflow, such as using AI-generated assets for concept art or texturing in 3D models. The goal is to enhance productivity without replacing the human creative element.

💡AI for Mainstream

This term refers to the common perception of AI as a push-button solution that instantly generates perfect images. The video points out the limitations of such mainstream AI tools, which often produce repetitive patterns and lack the quality needed for professional use. The speaker argues for a more nuanced understanding and use of AI in creative processes.

💡Data Privacy and Legal Risks

The video touches on the importance of data privacy and the legal implications of using AI, especially when it comes to training AI models on proprietary data. The speaker advocates for owning creative power by training AI models on one's own assets and IP, which helps to mitigate legal risks associated with using third-party data.

💡Peter Morbach

Peter Morbach is a well-known artist who has worked on projects like Magic the Gathering and has a distinct style. He is used as an example in the video to illustrate the difference between mainstream AI's interpretation of an artist's style versus a fine-tuned model trained on the artist's own work. The latter is shown to be more successful in capturing the artist's unique style.

💡Creative Process

The creative process is a series of steps that artists and designers go through to create new work. The video emphasizes that AI should be a tool that enhances this process rather than replacing it. It discusses how AI can be used iteratively to refine concepts and generate assets, ultimately saving time and allowing for more creativity.

💡AI Training

AI training involves feeding an AI system a dataset of images or other content so that it can learn to generate similar content. The video explains that the amount of data needed can vary depending on the complexity of the subject matter but that a fine-tuned model requires a curated dataset that reflects the specific style or subject the artist wants the AI to learn.

💡Open Source

Open source refers to software whose source code is made available to the public, allowing anyone to view, use, modify, and distribute it. Invoke is described as an open-source tool, which means it can be freely downloaded and used, offering flexibility and control to the users. This aligns with the video's theme of giving artists ownership over their creative tools.

Highlights

Kint KY, CEO and founder of Invoke, discusses the role of generative AI in creative processes.

Invoke is an early open-source interface for using generative AI on a local machine.

The company focuses on giving artists more control over the AI generation process.

Generative AI is seen as a tool to accelerate the creative spirit of humanity.

Invoke's vision is to build tools for artists, respecting their creative power and process.

AI models can be trained on artists' assets and IP to create new visual languages.

Professional AI tools differ from mainstream by offering control and consistency in creative outputs.

Invoke offers a workflow editor for customizing AI image generation processes.

The company provides a simple UI for teams to deploy generative AI workflows.

Generative AI can reduce the time taken to generate character concept art.

Invoke's tool is used as a creative part of the process, not a replacement for the artist.

AI-generated images can be used as a starting point for 3D models and textures.

Invoke is working on improving the transition from 2D images to 3D models.

The future of generative AI includes image-to-video models and more immersive content creation.

Generative AI is seen as the next tool in the creative process, respecting human creativity.

Invoke's approach to AI training involves creating a dataset and teaching the model through descriptions.

Legal considerations for AI training include fair use and terms of service agreements.

Invoke's open-source model aims to provide a clean, commercial-friendly base for generative AI.

The Invoke platform is designed to be artist-friendly and supports custom nodes for flexibility.

Invoke plans to offer an API for invoking specific AI workflows in a scalable manner.