AI Video Generation - Week 10 (tri 2)
- Julia Toczyska
- Apr 9, 2024
- 3 min read
Today, we had a look at video generation. We started the lecture with a group task - exploring and reflecting on Ai video generated art.
Group Activity One
The Metamorphosis by Roope Rainisto
For this task, I was paired with my classmate Leela. We found the whole video to be "uncanny" looking.
The use of Ai generated scenes made each character very inconsistent, but in my opinion it really added to the element of horror that the video tried to achieve. It also allowed supernatural characters to blend into the scenes seamlessly, for example in traditional movies, typically you can tell when a movie uses CGI - whereas here, everything is unnatural in a way. However, this could also be seen as a disadvantage as Ai has its own recognisable "art style" of sorts that makes it easy to tell when something is Ai.
Additionally, Ai allowed the 'bug actors' to actually act! I really enjoyed the use of extreme close ups and how detailed Ai made them look.
Inspiration
What are the current best video-gen tools?
Runway is a versatile video generation tool which offers a wide variety of features beyond video generation, including image editing, motion tracking, etc. It is a beginner friendly tool, however, it might not be best for advanced customization.
Pika Labs is known for fast video generation, as well as a affordable, pay as you go, payment model. Additionally, it's paid plans now allow you to generate sound effects to add to videos. Despite it's strengths, Pika Labs offers less variety and control compared to other Ai tools.
Deforum is a top choice for animation focused projects. It has a vast potential through community-developed scripts and extensions. On top of that, there is no cost barrier! However, it requires more technical understanding to set up effectively, AND may need a powerful dedicated computer for optimal results.
OpenAI has a history of pushing AI model capabilities. Compared to other video-gen models, it has a potential to integrate smoothly with other OpenAI tools (like DALL-E and ChatGPT). HOWEVER, it is important to note that OpenAI is not yet publicly available...
Group Activity Two
For this task, we were asked to experiment with animating dream scenes using Deforum Studio, starting from a static image generated though Ai tools like MidJourney, DALLE, or Stable Diffusion.
MidJourney
This time, I have used MidJourney to generate two images; a dream like fairy land, and a horrifying monster coming from underneath a bed. The goal was to create two drastically different animations in order to explore vast capabilities of the Ai tools.
Deforum
I was positively surprised with the outcome! Below I have uploaded a YouTube video of the animations. The process was simple, I have uploaded the images from above onto Deforum, and asked it to animate it. I have used the 'classic zoom out' animation type and the style was kept as 'Dreamscape'.
Group Activity Three
The objective this time was to create a line of wearable art (clothing) inspired by the imagery, themes, and emotions of dreams. After brainstorming with my task partner Savina, we decided to explore nightmarish fashion.
I found this activity extremely useful as it allowed us to combine a bunch of Ai tools we have been exploring over the past few weeks. In this scenario, I found ChatGPT a great source of information and advice, eg how to make different type of clothing and how to advertise to my niche target audience. Interestingly, it's suggestions were extremely up to date and accurate! Additionally, MidJourney did a great job designing extremely detailed concept images off of my brief description of the clothing line. Overall, I am extremely happy with the outcome and I feel inspired to experiment with combining a different type of Ai tools in my own project!
Comments