Civitai DevOps and Beyond ft. Sean // Civitai Team Session
TLDRWelcome to the special Civitai Team Stream featuring Sean, a devops engineer at Civitai. The live session aims to humanize the team behind the AI art community site, discussing Sean's role in maintaining servers, monitoring, and alerting. Sean shares his background in 3D graphics and devops, and they explore the challenges of scaling the site with the growing community. The conversation also delves into Sean's personal AI projects, including a unique 3D space visualization using stable diffusion, showcasing the potential of AI in creative realms.
Takeaways
- 😀 The video is a special 'Civitai Team Stream' featuring Sean, a devops engineer at Civitai, aimed at humanizing the operation behind the AI art community site.
- 👨💻 Sean has been with Civitai for nearly a year, working on maintaining the website, monitoring servers, and assisting developers with troubleshooting.
- 🛠️ Sean is also involved in maintaining the backend worker for the image generator and has contributed to the launch of new features like 'Comfy'.
- 🔧 Prior to Civitai, Sean had a background in 3D graphics and devops for about 10 years, working with midsize companies.
- 📈 The Civitai community and website have experienced rapid growth, requiring constant scaling and monitoring to handle the increasing load.
- 🤖 Sean prefers to run AI models locally for control and understanding, rather than relying on cloud services.
- 💡 The Civitai team is open to community feedback and quick to adapt, showing a commitment to building in tandem with the community.
- 🔄 The devops role involves balancing planned work with reactive tasks to maintain website performance and developer productivity.
- 🚀 Sean has been experimenting with integrating AI into 3D spaces, creating a workflow in Comfy that uses stable diffusion for 3D environments.
- 🎨 AI art allows Sean to realize his creative visions that were difficult to achieve through traditional art forms, due to the precision and control AI provides.
- 🔗 Sean's project on GitHub is open-source, inviting others to contribute and improve upon his work, reflecting the collaborative nature of the AI community.
Q & A
What is the main purpose of the video and the special Civitai Team Stream?
-The main purpose of the video and the special Civitai Team Stream is to introduce and humanize the people behind the Civitai site, particularly the devops engineers, and to give the community an insight into the work that goes into maintaining and improving the site.
Who is Sean and what is his role at Civitai?
-Sean is one of the devops engineers at Civitai. He has been with the company for close to a year, working on maintaining the website, monitoring servers, and helping developers identify and fix issues. He is also involved in the maintenance of the backend worker behind the image generator.
What does Sean's background include before joining Civitai?
-Sean has a background in 3D graphics spanning almost 20 years and has been doing devops for about 10 years. He has worked in midsize companies with a couple of hundred employees.
What challenges does Sean face in his role as a devops engineer at Civitai?
-Sean faces the challenge of scaling the website to keep up with the growing community and the increasing load on the site. He has to ensure the site is available, fast, and operational while also dealing with reactive work when issues arise.
What is the significance of the growth of the Civitai community and website from Sean's perspective?
-From Sean's perspective, the growth signifies the need for constant monitoring and scaling of the website infrastructure to handle the increasing number of users and the load they generate. It also means adapting to the growing number of images on the site and optimizing the site's performance.
How does Sean describe his experience working with AI and devops?
-Sean describes his experience as being involved in various aspects of devops and AI. He appreciates the ability to write scripts that automate tasks, which saves him time from manually pushing buttons. He also enjoys the control and understanding of what's happening inside these models by running everything locally where possible.
What is Sean's opinion on using cloud services for AI models?
-Sean does not rely heavily on cloud services. He prefers to run everything locally as much as possible for control and to understand what's happening inside the models. If he needs more resources, he might rent a GPU from a provider like Run OD.
What is the workflow that Sean has been working on in Comfy UI and what does it do?
-Sean has been working on a workflow in Comfy UI that integrates 3D space with stable diffusion, allowing the latents to be stored in 3D for inpainting purposes. This workflow can create a cohesive universe made of diffusion, with the ability to reproduce certain scenes and objects consistently.
What is the issue Sean is trying to solve with his workflow regarding the artifacts that appear in the generated images?
-Sean is trying to address the issue of artifacts that appear at the edges of the diffusion pads when generating images. These artifacts become more noticeable when using multipath models with short steps, which don't allow enough time for the prompts to integrate properly.
How does Sean's workflow handle the creation of prompts for the diffusion process?
-Sean's workflow uses prompt generators that utilize GPT2, trained from data on the Civitai website, to create interesting additions to the prompts. This allows for dynamic and varied prompts that can enhance the diffusion process.
What is the future of Sean's project and how does he plan to share it with the community?
-Sean plans to open-source his project and has already made the code available on GitHub. He hopes that others might be able to improve upon it, especially in areas where he has encountered difficulties, such as dealing with artifacts in the generated images.
Outlines
🎥 Introduction to the Citi Community and Special Guest
The script opens with a warm welcome to the Citi community on their website, highlighting the generative AI art aspect. The host encourages viewers to like and subscribe to support high-quality content. The video features a special guest, Sean, who is a devops engineer at Citi. The host expresses gratitude for Sean's participation and emphasizes the importance of community engagement with the team behind the site. Sean is introduced as someone who has been instrumental in maintaining the servers and monitoring systems, as well as contributing to the development of the image generator. The host invites viewers to get to know Sean better and asks him to introduce himself.
👨💻 Sean's Background and Experience in DevOps and AI
Sean introduces himself as a devops engineer at Citi, having been with the company for nearly a year. He talks about his responsibilities, which include maintaining servers, monitoring, and assisting developers. Sean also discusses his previous experiences in 3D graphics and devops for midsize companies. The conversation touches on the rapid growth of Citi and the challenges of scaling up the website to accommodate an increasing number of users and images. Sean shares insights into the autoscaling processes and scripts that help manage the site's growth and mentions the importance of scripts in reducing manual workload.
🚀 The Role of DevOps in Supporting a Growing Community
The script delves into the day-to-day responsibilities of a devops engineer at Citi, emphasizing the role in ensuring the operational efficiency of the website and the productivity of developers. Sean explains that despite the potential for perfect code and design, the website's uptime is crucial for the success of the service. The conversation highlights the balance between planned work and reactive tasks, such as addressing issues as they arise. Sean also discusses the growing pains of the site, including the need to adapt and optimize for the increasing number of users and the community's feedback.
🤖 Open Source Contributions and Personal AI Projects
The script explores the importance of open-source contributions to the rapid advancement of AI, mentioning how big companies like Google and Facebook have influenced the field. Sean shares his personal AI projects, including his work on the Onyx project, which aimed to provide hardware acceleration for AI models on AMD cards. The discussion also covers Sean's journey into AI, starting with a rock pie device for real-time computer vision and leading to his current work with AI art and generative models.
🎨 Combining AI Art and 3D Modeling for Creative Workflows
Sean presents his latest project, which involves using AI to create 3D spaces with stable diffusion run over it, allowing for inpainting and consistency when revisiting scenes. He discusses the technical aspects of his workflow, which includes using depth control nets and multi-diffusion techniques to achieve a cohesive 3D environment generated by AI. The script also covers the challenges faced, such as artifacts appearing due to the storage of latents and the limitations of real-time rendering for video games.
🔧 The Future of AI Integration in Citi's Services
The script concludes with a look towards the future, discussing upcoming features and improvements at Citi. Sean mentions the recent addition of Comfy support in the image generator, which enables new capabilities like face fix and high-resolution outputs. He also hints at potential video generation features on the site, inspired by his own workflows. The conversation underscores the team's commitment to leveraging open-source tools and community contributions to enhance the platform's offerings.
📅 Upcoming Events and Closing Remarks
In the final paragraph, the host announces upcoming events, including a guest creator stream featuring Emma Catnip, known for her work with anime diffusion. The host also mentions the Twitter AI community space and encourages participation. Sean is thanked for his time and contributions, and the host expresses hope for a restful weekend. The script ends with a reminder of the next week's schedule and a farewell to the viewers.
Mindmap
Keywords
💡Generative AI art community
💡DevOps engineer
💡Autoscaling
💡Stable diffusion
💡Latents
💡Comfy
💡GitHub
💡Crunch time
💡Prompt engineering
💡Onyx
Highlights
Introduction of a special Civitai Team Stream with Sean, a devops engineer.
Sean's role in maintaining Civitai's servers and monitoring systems.
Sean's background in 3D graphics and devops, spanning nearly 20 years.
The importance of community engagement and humanizing the operation of Civitai.
Civitai's growth and the challenges of scaling the website to meet user demand.
Autoscaling strategies and the use of scripts to manage server load dynamically.
Sean's personal preference for local AI model deployment over cloud services.
Discussion on the adaptability required in a devops role, especially in a startup.
The significance of open-source contributions to the advancement of AI and tech tools.
Sean's self-taught journey into AI, devops, and 3D graphics.
The impact of community feedback on product development at Civitai.
Sean's exploration into integrating AI with 3D spaces and the potential applications.
Technical deep dive into Sean's workflow for 3D diffusion and inpainting.
The process and challenges of turning 2D images into 3D spaces using AI.
Sean's motivation behind creating AI art and the capabilities it enables.
Upcoming features and improvements at Civitai, including support for new models and workflows.
The potential for real-time generative environments in video games using AI.
Closing thoughts on the importance of community and open-source in AI development.