I'm back! (with more AI power!)

Greeting

Wow, I can't believe I did't write anything over 6 months! Well, actually I can see how that happened as I was swamped at work, as well as (Japanese) baseball season kick off, you know, I can't really do much at home so nothing to write about.

So what brought you back?

Few things. For one, my team went to play off and got eliminated rather quickly ... so that would open up few hours per day for me and at the same time, I have a reason to check my game (to prepare for internal tech conference)

6 months later ... AI is WAY more capable

I remember, 6 months ago when I ask Antigravity (with low-end model) to change something on the biggest file, it often mess things up, like put closing braces somewhere else and break the game - I thought that's just the way it is. Now - it almost never mess things up. I'm using Antigravity still for development so low-end model for me is "Gemini 3.8 Flash-Low" but it's so capable (compared to old Flash model anyway)

I myself learned a bit more on how SKILL.md works and that certainly helped to speed my development but I feel even without all that, it can perform at least twice if not more things in the same amount of time - it's really insane how quickly AI improve!!!

My multi-agent approach

Whole wack of people talk about how they can leverage multi-agents to supercharge the development process. I myself thought about it and probably could/should look into that but ... with this improved AI, I can't even keep up with making one agent busy lol I feel I'm the bottleneck here - but one thing I started to is to leverage Codex as a very clean place to do all sort of experiments - like creating various effects, blush up scenes and even look into 3D capability with Three.js (but I don't think I would use that in my game - at least in the first one that I'm builiding)

So here is how I do - I use Antigravity to do coding task. While it is in progress, I then ask Codex for some experiment - I know, that's not really how "multi-agents" works but ... this works! Only problem is, because I like doing things with low-end model, they are rather responsive and it come back pretty fast lol

I built utility before and posted here

I don't think not too many people follow me or read my past posting but I did talked about how I built utility to help me create a hotspots (link to article) It was pretty useful, I use it, I love it - it serve my purpose well.

As I'm going back to game development, first thing I notice is the how i was placing my character in dialogue - it was awful as I was using standing image no matter what the background image is, making character looks like floating when they are indoor lol So I fixed it (well, Antigravity did) by moving character forward, and that would make the rather poor image resolution very obvious. I thought, if AI for coding agent is this much improved, I'm sure image generation capability of Gemini must improved too.

Remaster images

So I took one image for test and ask it to remaster ... and OMG, it was done so well and so fast! (I set "Flash-Lite" as well) At first, I thought it just paste back the image I provided but when I look at the output, it is so much clearer and much better resolution. AI, you never disappoint me (well, never is too strong of a word but hopefully you get a gist of it)

Remaster images means ... I need to do THAT again ...

For those of you who tried to build game knows this but game development has so many repetitious and tedious tasks. One of which is to take image and convert it to PNG with proper transparent zone so that you can use it. In the past, I've done this, with brute force, using Preview of Mac ... needless to say this was painful process. So knowing we can build utility, I decided to give that a shot.

Codex, you are good too!

Sorry, this might sounds stupid but I had Chat GPT for a while and I keep my paid subscription because ... I feel it helped me so much to refine my story, it's like a buddy of mine (I know, I might look weird to say that but that's how I feel so can't help it) Anyway, so I knew I have access to Codex. I heard a lot about it, I knew it's probably as capable (or more) than Antigravity so I gave it a shot.

One hour is all it need

Honestly, I didn't keep track of time - it initially did what I want with essentially a single prompt but since they I've extend its capability few times. First, I ask it to be accessible via my tablet. The idea behind this is so that I can use tablet to do the small, fine area editing - this way I don't have to buy drawing tablet (which I bought once in my life and quickly gathered a dust lol)

At first I thought I need to somehow mirror my PC and tablet - Gemini suggested me application (Duet something) but after going back and force, I realize:

Hey, I have a tablet - maybe I can just use that?

Once I got that idea, rest were rather easy. I realize I have a tablet (Pixel) that can work with SCI pen so I ordered that. But while I'm waiting on that, I thought I can work on another, arguably bigger pain points which is to create "facial expression images".

Here you go - meet Alpha Studio

So once again, I decided to expand utility - idea is simple - load image, automatically detect background (often white but could be any color) and first automatically remove it. Then potentially use small tweak (I can do this per pixel - that is the reason why I wanted this tool) as needed and ... here comes another pain points. The way I set up, I have 4x2 grid of images, all having specific facial expression so wouldn't it be nice if I can just simply assign area that I want to crop and then use pre-register filenames to generate all files a once? Yeah, Codex please make it happen.

I thought this is too wishful thinking but Codex one shot my requests. I was deeply impressed (and honestly got scared how powerful this thing is)

Anyway, I recorded my entire session of

• load image
• select background and remove it
• do small tweak (here, boundary is clean enough I didn't need to go to pixel by pixel but if I wanted to, I can)
• use multi-select to frame each facial expression images
• select all frame and move frame a bit up so that I can capture up to what I wanted to be
• export as zip, with filenames that is ready to be loaded into my game!

I can't believe what you can do with AI these days ... and best part is, all of this (remaster images and build utility) is not even full morning task! Honestly I feel I'm spending more time typing this (yeah, I know, I can use AI for this too but I'd rather use my own words when I put some article out, though nothing with using AI too if quality of information is high, mine post has questionable in that sense so I should at least type this out :p)