20250702 Supervision Meeting

Here is the Speech draft.

P1. As you can see, here is the refined custom workflow for AI-integrated digital woodblock print animation, which was shown in my transfer meeting, probably Eirini can remember it.

The reason I need to show this is that I need to tell you my current progress and what I have done this month.

This month, I mainly focused on these three processes, from Integrating AI into Traditional Workflows to Splitting images. And of course, I already resolved these technical challenges before integrating AI into Traditional Workflows.

P2.

So, let's move to the next page, LoRA. This page is just for a reminder that LoRA can transfer style in a transitional way.

P3.

Ok, why do I need to train multiple custom, stylish LoRA? Because I want to do an animation with a specific style and character. On the left side, as you can see, here is an image generated by the prompt only with no LoRA.

In the middle, here is my main character for the Qigong animation; the right side is the style distilled from an ancient printed book.

So, middle is what we call character LoRA to ensure that the character retains specific features, such as clothing, pants, and appearance.

P4.

So, actually, we can combine multiple LoRa to generate a certain character with a specific style. like Addition. But the question is, how can I let this character do a pose that I want him to do?

P5.

So, we need to adopt a way to figure this out. So, the idea or approach is to generate a T-pose image by using my character LoRA.......After this process, I can basically get a reference image to help AI understand what I want it to do.

P6.

Here is a screenshot that shows AI is generating the character with a certain pose by using multiple reference images.

P7.

Finally, I can get two satisfied images after a large number of attempts. A start frame and an end frame, which AI will automatically fill in the missing parts between the two images based on the prompt. This is also the most popular and mature solution at the current stage.

P8.

The short animation on the right side is one of the shots that I took from the animatic. I used a specific prompt to achieve this animation after several attempts.

P9.

Here is the basic principle for making woodblock prints.

P10.

If you noticed, AI-generated images don't have some characteristics such as misregistration, impression or trace. So, here is the problem I need to solve for transferring AI-generated images to images with an authentic print-like appearance.

P11.

I need to create a tool or script that separates image sequences into different monochromatic colour layers, based on the basic principles of woodblock printing. This will allow me to achieve an authentic print-like appearance for my animations, whether I'm working digitally or using traditional physical methods.

P12.

But the last script is not good enough to achieve my intention. We can find that the colours of these separated images are not clean. These separated image sequences will mix with other colours, and are not monochromatic images. That's why I spend lots of time refining my script with AI's cooperation.

P13.

Let's compare these three separate images. It's a gradual process of refinement. which improved a lot.

P14.

Here is the final result. The illustration with flat colour at the right is composed of 3 separate images.

P15.

On the very left side is the original image, on the very right side is the recombined image. The images in the middle are separated into monochromatic images with different colours.

P16.

Also, I found a very interesting point is that the image sequence can be adjusted into a carved line appearance when I adjust the settings of my script.

Questions:

I couldn’t find the book Eirini recommended. Which one is it exactly? Can you just tell me the name? Do I need to read these two PhD books?

How should I turn my practice notes into organised writing? Can I just follow the style of those PhD theses?

Is it possible to include the journals I published in as a chapter in my dissertation?

I feel like I already have a lot of writing and practice material. Can I submit a paper now?

Right now, I want to focus all my energy on finishing this animation and then submitting it for awards. I think making it will be really fun. Is it okay if I put most of my effort into this?

others： 1. The cost of laser cutting wood boards / the printing cost here / the composition rules for laser-cut wood prints.

Then we need the standards to be the same across different studios.

So why are there some jagged edges on the wooden boards?

If we finish the animation, can we print the whole thing? And is the cost affordable?If we use Wong Kar-wai’s low frame rate style, let’s say 4 frames per second at minimum.

Option 1: From 0:15 to 1:50, that’s 1 minute 35 seconds — so about 380 wood boards. We could maybe adjust the animation pace to make it shorter and save costs.

Option 2: Or we can pick one part of the story and make a printed book from the wood boards. From 10s to 1min12s is around 1 minute — about 240 wood boards.

Option 3: Just print one shot. From 45s to 57s is 12 seconds — so 48 wood boards.

Option 4: Go fully digital and only print keyframes.