Civitai AI Video & Animation w/ Tyler
Entertainment
Introduction
We’re back! After a brief pause, we are live again at 1080p. I had a bit of technical trouble logging back into Restream where all my settings were reset. I needed to reconfigure everything, including the stream’s resolution, which defaulted back to 720p. But hey, we’re back in business now!
Today, we're diving into some exciting AI video and animation discussions. Lilly mentioned that Cog video could be used alongside image-to-video capabilities to maintain style. I spent last weekend getting my Mochi AI setup running smoothly, and thankfully, a member from the Banad Doo community hopped on a call to help me, as I faced multiple hurdles along the way.
Currently, each video I run takes about two minutes for just two seconds of footage on my local setup. I anticipate that improvements are in the pipeline thanks to people like Kiji who are at the forefront of optimizations for these processes. Nonetheless, the progress is exciting—we're gradually developing a high-quality, uncensored video model that functions locally!
We began by experimenting with video prompts by trying different combinations, such as "Donald Trump riding a skateboard with a fisheye lens." I am also considering creating vertical videos since I haven’t explored that possibility yet.
In training some models, I’ve specifically aimed for an Akira-style cyber bike since I believe it would be a fantastic addition to our resources—especially since there isn’t one available in the current library.
I ventured into organizing images of the cyber bike and deciding on their variations. However, I was met with a dilemma on whether the images should include a consistent color scheme. It's essential that the caption for the images includes important details, such as color, as it will influence the diversity in outputs when the model is trained.
The image acquisition process has been integrated into my workflow with “moodboarding” tools. I also have my collections on Cosmos where I gather inspirational images I come across. I believe moodboarding is beneficial for any creator to harness different influences and references.
My next steps included preparing and training a Laura model specifically for the Akira-style cyber bike. After curating 18 images and ensuring they include the right captions and descriptions, I zipped the files for submission into the Laura trainer. I’ve learned to define attributes like the color red, as it helps the learning process.
Throughout today’s live stream, we also touched upon past creations, including a surreal Halloween-themed video featuring cats, which required much effort to synchronize sound with the visual animation of cats. Halloween festivities have taken center stage, and it was a joy to see so many community members participating in the creative challenge.
Tomorrow, we’ll wrap up our training series with KF Tiger. We’ll cover how to effectively train character models and share tips for achieving quality results when working with AI-driven animation.
Keyword
Civitai, AI video, animation, Tyler, Mochi, training models, Akira-style, cyber bike, moodboarding, video prompt, character models, community engagement, Halloween.
FAQ
1. What is Civitai? Civitai is a platform focused on utilizing AI for video and animation creation, providing tools for users to experiment and create visually compelling content.
2. What is the purpose of the Mochi AI setup? The Mochi AI setup aims to enhance video generation capabilities by allowing for faster rendering and better quality outputs through local models.
3. How can I contribute to the community on Civitai? You can participate in community challenges, share your creations, and engage with other members through platforms like Discord or the gallery feature.
4. What will be discussed in the next livestream with KF Tiger? The next stream will focus on character modeling best practices, including how to train effective models for specific character types using AI tools.
5. Why is moodboarding important for creators? Moodboarding aids creators in gathering inspiration, keeping a cohesive visual language, and fostering creativity through the curation of diverse references and styles.