This week, we continued our research, discussed more about our deliverable goal, and prepared for the quarters presentation next week.
Work Preparation:
This week, we discussed our deliverable goal with instructors at our instructor meeting and got approved. We then worked on the quarters presentation deck, emphasizing clearly explaining the three different approaches we want to take, the storyboard we created, the project goal of comparing the tools and analyzing the pros and cons of each, the three different styles we wish to have for the three films, and the research result we have so far.
Based on the deliverable goal, we continued our research and looked more into the AI concepting, AI texturing, and AI rendering.
Progress Report:
- Continued researching on character concept art and chose a finalized version
- Continued researching on the environment concept art and chose the three different styles we want for the three films
- Tested the AI texturing on hand modeled mountains
- Had more discussion on the storyboard and AI generated the storyboard
- Finished the quarters presentation deck and rehearsed multiple times
- Finished our documentation for week1 research
- Had our team bonding dinner
- Finished our team logo
Research Results:
- AI Character Concept Art
With the Pangu design, we learned that there are significant diminishing returns to generative AI. It’s pretty easy to zero in on a generalized design after half dozen or so rounds of iteration with a dozen or more prompts in each. But the smaller the changes are that you need to make, the more difficult it becomes to direct the AI. The most fruitful methods involved crudely compositing parts of the images we liked in photoshop and resubmitting them back into the model.
- AI Environment Concept Art and Texturing Tool
Based on last week’s research result, we refined the prompts to get more precise results. We researched and played with the AI Texture tool to test their performance.
- Based on the results of two AI tools, we agreed that the results of Midjourney are more accurate.
- Poly and Texture Lab AI texture tools are pretty promising so far, but it needs to be tested on other AI generated models and manually made models. However, some textures it produced are problematic.
- The UV is also a mystery by now.
- AI Logo Generation
For one thing, the AI struggles to design a simple logo. Both of the models we tried repeatedly wanted to add more and more detail. Eventually we got a few things we liked, but about half of the work was just regular manual graphic design in Photoshop and Illustrator. The tools make graphics design a bit more accessible for someone like me who doesn’t have a background in it, and they might speed up iteration for a designer. But like with the concept art, there is a last mile problem when using generative AI to perfect an image.
Plan for next week:
- Have our quarters presentation on next Wednesday and reflect on the feedback
- Continue researching on the three pipelines
- Figure out our milestone goal before the half presentation
- Work on the storyboard more and figure out the scene transitions
- Start to greybox and set up camera angles
Challenge:
- Hard to estimate how long each approach will take and what potential problems we will meet throughout the semester.
- Hard to estimate the cost for AI tools, and how effective they will be.
- Need to make clear with the instructors what the expectations are, and what is the definition of ‘failure’ for our project.
- Need to convey our project goal clearly to faculty during the quarters presentation.