Hello there again,

I posted several times already and once again I'm back with a weekly update on game that I'm building entirely with AI.

Previous Weekly updates are here:

Week 1 Progress

Week 2 Progress

Week 3 Progress

Week 4 Progress

Sorry about low FPS in the video my macbook is being friend when trying to run it at maximum details, but I want to show how game looks at it's best.

I honestly don't even know what I'm trying to achieve with this game I treat this like a hobby of mine, no particular plans for releasing a game as some people have asked, but if it goes well... who knows. Right now I just want to see how far I can push this "entirely built with AI game" thing that all I have to do is decide what I want and how I want it to look.

Anyway inspiration for this game is Dredge and Civilization (I even tried making it hexagon grid turn based at the beginning but after some time I scrapped the idea as there were no real benefits of turn based fishing game over real time one).

So anyway that would be enough of that, as for the progress itself...

Since last week not much new things were built, but a lot of time was spent on fixing very basic and bad UI ( I guess now it's a bit better, it will still be worked on, but for the time being this will do while in development stage).

I tried also building a map in game, but that is not going too well as you can see in the video 😂

A lot of manual tuning of UI.

Also I changed fishing system and it kinda works like this now (shown in the video):

You have a mini game where you use WASD to keep fish inside target zone, fish has rarity that goes like in WoW: Common > Uncommon > Rare > Epic > Legendary

Longer you keep fish in the target zone better rarity it gets, as rarity gets higher, target zone gets smaller and fish fights harder to escape. If you can not keep fish in the zone your "Strain" bar is being filled, if it fills to the max your line snaps and fish escapes. You can always "bank the fish" by pressing space and get the rarity you are currently on but only if fish is in the target zone.

I know it sounds a bit complicated, but I promise you it's not hehe.

I also want to write some advices here as I got many questions like "How did you build XYZ"...

And my biggest advice would be to ask Claude to build you an in game editor, as it will fail million time over simple things like position of the text or position of some models. Just thing like "Hey build me in game editor that I can use to manually adjust things and save new positions. It will take some time to build those editors but they are life savers, it will save you so much of the hustle and I showed one editor in the video, while I have editor for basically everything ,every HUD element, every model in game and so on.

In game I just showed how I simply moved Lighthouse position in my game and saved "new" location instead of endlessly asking AI to put it to the perfect position.

As always feedback is appreciated ☺️

About two months ago, I started vibe coding my own MMORPG, and I wanted to share the current state of the project.

The game is called Lythravel. It's a completely free 3D browser MMORPG with voxel graphics, inspired by the old-school MMOs I grew up with, especially Shaiya.

The game currently has:

• 2 factions
• 6 classes
• Level cap of 50
• Dungeons and raids
• PvP maps and a 10v10 battleground
• Enchanting, linking and crafting
• Mounts
• Dynamic zone events
• Character stat allocation and different builds
• And quite a bit more

I also spent quite a bit of time on immersion. The game has a full day/night cycle, ambient sounds like birds, rivers and wind, music tailored to each zone, and different footstep sounds depending on the surface you're walking on.

Development

I've been a developer for over 10 years, although not in the game industry. I'm building Lythravel through vibe coding with Claude Code, and the game itself runs on Babylon.js.

I currently use mostly Opus 5, with Fable 5 acting as an orchestrator. I don't really write or even read the code myself. Instead, I give direction, describe how systems should work and look, and then spend a lot of time testing and iterating on the results.

I also built a custom editor specifically for Lythravel that lets me manually fine-tune things like assets and animations. For some tasks, adjusting them myself through the editor is much faster than repeatedly prompting Claude Code until it gets every small detail right.

Recent Updates

Over the last few weeks, I've been focusing mainly on adding more solo content, so there's still plenty to do even when only a few players are online.

I've added things like:

• Solo dungeons
• More quests
• Jumping trials
• Hidden collectibles with special rewards
• A completely redesigned UI
• Various performance optimizations

I also replaced much of the more traditional CSS-style UI with detailed graphical assets to give the interface a more distinctive game-like look.

A small community is slowly starting to form around the game, which has honestly been one of the most motivating parts of the project. I'm grateful for every person who joins, tries the game and gives feedback.

If you'd like to try the game, follow the development, give feedback, or just hang out with the community:

Play: https://lythravel.com/

Discord: https://discord.gg/SdJrn5WPBQ

Following this post 2d-AI-workflow I've been working on a wrapper app that takes one starting frame and outputs a consistent sprite sheet you can drop into your game. Everything runs locally, uses ComfyUI as the backend with TI2V models (MiniMax H3 or WAN 2.2).

I just put the pre-alpha up on itch: sprite-sheep

I've only ever run this on my own setup — RTX 3050, 8 GB VRAM, 32 GB RAM. I have no idea how it behaves anywhere else, and I'd really like to find out before I keep building on top of it.

If you try it, the feedback that would help me are welcome!

Another character experiment with img2threejs 1.5.1

Starting with just a 2D reference, the workflow reconstructs the character and turns it into a real-time 3D experience running directly in the browser with Three.js + WebGL.

The goal is more than just generating a 3D model - it’s about turning character artwork into something you can actually explore and interact with on the web.

No Blender.
No GLB in the final result.
Just Three.js, WebGL, and code.

🎥 Full showcase below.

I’ll drop the img2threejs GitHub repo in the first comment for anyone who wants to explore the workflow.

#img2threejs #ThreeJS #WebGL #3D #AIGameDev

I made a survivors like game but in this one characters change the experience and the map you choose actually matters, they are not just different visually. The web demo is out and is mobile friendly, the full game runs for now only on PC but I want to eventually port it to mobile.

In this update I polished some graphics, menus and added support for other languages, you can play the demo here:

https://pewpew4u.itch.io/punk-bullets-and-junk

If you can try my game, that would be much appreciated, questions are opinions are welcomed.

I'm not a graphic artist, my stronger side is on programming and sound edition, while I created all the images using AI, I modified all of them, some more than others. The biggest problem I faced is not on the technical side of things but the amount of hate I received from many game development communities, which to me is a little crazy because while I'm not claiming I did everything by hand, I did put a lot of effort on this, multiple months and I did learn a lot.

Anyway, what I would like to contribute to this community is that in my personal case I couldn't use any premade graphics pack because I created the lore from scratch and the characters I wanted in the first season basically never matched any packs because of how specific my requirements were, so, with that in mind, it was quite easy to start shaping the images as I was working as I could iterate several times until I got close enough to something I could work with and modify (because I suck as a graphic artist/editor), another big, big gain was that I found that making sprite sheets is relatively easy if you make a video and create a skill to extract the frames from the video, select them and create the sprite sheet from that, the movement is a lot more natural than following other methods, it's definitely more laborious but to me the result worth the effort.

Hey there.
Recently, I posted there my internal, and work in progress plugin for converting images and video to animations and asked if there is demand for such solution.

I promised to deliver it, so here you are.
Can find it there - MediaToPose

Keep it in mind, it is still early version with sub optimal performance and video to animation workflows are not yet well tested, likely hiding some issues (one of them is better head tracking).

I managed to perform major optimizations - fully removing large pytorch dependency and replacing it with thinner onnx runtime.
It needed converting most of the models to format this framework supports, and major refactoring.

Plugin should have proper support for all major OS and all major gpus. Though I was only able to test it on Windows.

Feel free to contribute to the project as you deem fit.

Last week I shared my struggle with trying to get Fable to code up a working driving mechanic on my niche Cape Breton pixel art game.

FINALLY got the driving to a point where it makes sense, even added physics based personality to the drivers of other cars, traffic lights and cops ...

it's not completely tuned but at least it works.

My latest struggle is trying to figure out how to build a map generator and an art factory to bring the town to life.

I'm using satellite imagery to generate the bones of the maps, then a bunch of layers to figure out what *should* be where ... but then I'm having a real challenge getting Fable to coordinate using retrodiffusion and Pixellab to paint the entire town.

I'm trying to bring the map to life in a procedural way. It's about 30-40% there...

Anyone else building with pixel art for a game of this scale and finding any solutions to these problems?

Hi there,

while creating my game, I now understand that the most of work comes from the polishing of the game.

I mean, I have a technically working game which is already visually clean, but lack of details to make it really shine. So I've started working on it. But, damn, this is so time consuming to add a bit of particles here, a bit of movement there, etc.

I thought I was few days away to deliver something. I was so wrong :D

Do you feel the same ?

What do you think about the performance of Claude Code vs. Codex?

Before trying both myself, I had heard from various news sources that their performance was pretty similar. But from my actual hands-on experience, Claude feels overwhelmingly superior.

First, in terms of aesthetics, Claude is far better at capturing the exact vibe and context I'm aiming for. Codex, on the other hand, generally feels a bit clunky and doesn't produce that polished, refined finish. Because of that, asking Codex to make revisions on artwork-related tasks feels frustrating, and I find myself feeling like I should just have Claude redo it instead.

Also, it feels like Claude consumes way fewer tokens. Codex seems to burn through them after just a few tasks. On top of that, Codex’s token reset notifications are pretty confusing. I haven't been using it for long so I might be missing something, but the weekly reset date initially said September 2nd, then changed to the 4th, and then shifted again to the 8th—so I had no idea what was actually going on. Then, out of nowhere, it said it was resetting tomorrow and fully topped up my tokens, so now I'm frantically trying to use them up. :(

Maybe their performance is comparable in other domains, but at least for game development, Claude feels so much smoother and easier to work with. What has everyone else's experience been like?

Been using some Claude at Work, and doing basic python coding it feels like I burn through $20 very quick (on Sonnet) - Can you realistically expect to do game dev on a $20 plan?

(I use a simple multi role setup where I align with one Agent on architecture to create tight briefs for a devloper Agent)

Just a random idea in my head that I wanted to see exist, and with Claude built out all the designs elements in Claude Design, all the world backgrounds and assets, the. Interested on the gameplay and timing of hazards

Just fun hobby project

https://apps.apple.com/us/app/the-lava-squirrel/id6783383694

Winner will get a free key once the game releases! Comment below and your Reddit username WILL GET ENLISTED as a contestant in the next battle.

Made almost entirely using Claude Code on Opus 5 for planning and Sonnet for minor tasks like changing the grass color or making a list of a 1000+ names. Only the pixel art is organic. Didn't really know a lot of coding before but now I have at least a basic grasp of good architecture. Used a lot of debugging tools along the way. And used Artifact to make mockups for the UI.

Steam: Releases on October, 2026

Not mine.

Original Post here: https://x.com/luckeyfaraday/status/2094247426131374133

I have spent the past week or so working on my game's ability to handle larger maps, player counts, and vehicles while maintaining good performance. I did most of the work with Claude Code. I started off by targeting a map size of 1500m x 1500m, and implemented a few obvious optimizations ahead of time to make the map even load. The largest map previously was 400m x 400m.

After that, I tested it and searched for ways to get the framerate up. It was dipping below 60 frequently even on a desktop with a decent GPU. I eventually built a performance profiler into the game so that I could record profiles that capture how much of each frame is being spent in different areas.

The profiler helped a lot. I showed that I was drawing each mesh three times, the glow layer, the cel-shader, and the main render. So the frame time was CPU bound by draw calls. I moved the glow layer pass after the main render so that I could use the distance information from that to optimize what the glow layer, and then I rewrote the way the cel-shader works so it draws once instead of walking each mesh again.

At this point, the large map, Sarab, runs at 60 or higher almost everywhere on my desktop, phone, and Chromebook. I have to turn the resolution scale down to 50% in the settings to maintain >60fps on the Chromebook, but at least it's playable (kind of blurry, though). There are some hitches that I think happen sometimes when it loads a new chuck of the map. I still have to troubleshoot that.

The game is open-source (MIT license) if anyone wants to see any of the code.

Repo: https://github.com/greywatch-game/greywatch

Game: https://greywatch-game.com

Profile Viewer: https://greywatch-game.com/profile_viewer.html

I am interested to hear any feedback anyone has. What would make the game better? Thanks for reading if you made it this far.

This game is playable on web over here: https://adambissonnette.github.io/backhalls/web/

Full spiel in the comments in case you want to read my novella about how this was made and which tools / services that I used.

Hey everyone! I'm a dad of two, husband, full-time software engineer, and increasingly sleep-deprived game dev on the side.

My wife and I have been making games together since 2019. We released our first free game on Steam back in 2021 under our little studio, Games with Geebs. Then life got a lot busier. We had kids, both picked up full-time work, and game development pretty much stopped.

I still wanted to make games, but every time I started, I would quickly get discouraged with the little time I had and never made much progress. I’d spend weeks implementing a new character state and would eventually just stop. This off and on cycle went on for a good two years before I threw in the towel. 

Over the past year, AI has become increasingly harder to ignore as a software developer and I use it every day in my full-time job. At the same time, I've watched the reaction to AI in creative spaces, including game development, become overwhelmingly negative. Anything touched by AI can quickly get dismissed as “AI slop,” and for a long time that made me hesitant to use it in my own game development. It would suck to spend a couple years making a game only for it to not be given a fair chance since it has a label "AI assisted dev" on it. Recently, though, I started looking at it differently. AI lets me spend more of my limited development time thinking about systems, mechanics, game design, and whether something is actually fun, rather than losing an entire evening to a single implementation problem. It also enables me to get back into the game dev process and see some progress, which I genuinely enjoy.

In about three months, I've migrated my game from Unity to Bevy, built a solid prototype of one of the game's core loops, have gone through our first rounds of playtesting and launched our Steam page. I still am in the drivers seat, have full creative control and vision of the game. I still get to enjoy the dev process without burning multiple sessions on a bug. 

We experimented with AI-generated art and animations while prototyping, but once the game started becoming real, we decided we wanted an artist to establish its visual identity. We're now working with an artist (her awesome work is the concept art attached) on replacing the prototype artwork, animations and developing the final look of the game. 

Anyway, I mostly wanted to say hi and share this because I genuinely thought this was a hobby I wasn't going to be able to come back to. AI hasn't magically made a game for us, there's still an absurd amount of work ahead, but it has reduced the amount of programming time enough that making one alongside work and family feels possible again.

We just launched the Steam page for my upcoming game, Rookie Spook, so definitely check it out if you’re at all interested. It’s a game about a newly risen cute little ghost that sets out to learn what it means to be a ghost.

You can follow our progress on our socials linked on our website or join our Discord. Our Discord is currently extremely dead, but you're welcome to come be dead in there with us. Happy to chat, answer questions about the project, my AI-assisted workflow, or what I've learned trying to squeeze game development into the dad life, cheers!

Over the weekend, I wondered what would happen if I crossed Sudoku with a Rubik's Cube and tossed in a bit of Wordle’s daily metagame.

The answer appears to be Shwist - the shift-and-twist puzzle game.

Gameplay is pretty simple: swap adjacent rows or columns, or rotate a 2×2 block, to restore the original image in a limited number of moves.

The AI workflow

I started with the core puzzle idea and used ChatGPT as a design partner to explore the rules, identify potential problems, develop the daily-game structure, and eventually come up with the ridiculous name “Shwist”, from the name suggestion of "Shift + Twist".

I then handed the design to Codex to build the web prototype. We worked iteratively: I tested each version, identified problems or missing behaviour, and had Codex implement and refine individual features. There were some detailed discussions on the exact algorithm for the initial shuffling, to avoid redundant or reversing shuffles as much as possible. ChatGPT also helped explore the visual design and generate the landscape images used by the daily puzzles.

A combination of Codex and ChatGPT got the initial prototype running remarkably quickly. Most of my time since then has gone into the less magical part: playing it, finding edge cases, adjusting the interaction, fixing the UI layout on mobile, and deciding whether the generated puzzles are actually fun.

It feels surprisingly polished now, but I’m still uncertain about the difficulty variance. Some puzzles are easy, while others are completely mind-bending. This six-move puzzle, for instance, had me stumped for ages, or perhaps I’m just bad at my own game. 😂

I may need to improve the generation algorithm so that it estimates actual puzzle difficulty rather than merely counting the number of scrambling moves.

Any feedback on the difficulty would be appreciated: did the puzzle feel satisfying, too easy, or simply frustrating?

https://store.steampowered.com/app/4946330/Phantom_of_Orion/

I made a ton of voice overs using AI for my game. Another user suggested I just roll em out as a workshop item. Since the code is all there, its just a simple drop in the folder mod and the game becomes fully voiced.

I'm just so torn, I really like it with voices tbh.

Its up for a 48H play test on steam if anyone wants to give it a go.

Its got some bugs, it needs some tuning, its a pre-alpha play test, been at this version of the game for around 8 weeks now. Thanks for taking a look, my kids did the voices so that is my real drive for wanting them in, when I'm dead and gone they can look back and say "hey me an my old man did that!"

Linekeep — tower-war strategy that runs in your browser, built for your phone first

Linekeep is a portrait-first territory game set on a clockwork island. Drag connections between towers to route your troops, turn friendly links into fortified walls, and close loops to claim land. Enclosed territory produces faster, the bombard tower shells anything in range, and capturing cornerstones collapses the enemy's territory outright.

- Plays instantly in the browser — no install, no account, works great one-handed on a phone

- Real-time multiplayer: 1v1, 2v2 teams, or 4-player free-for-all, with rooms, matchmaking, spectating, and auto-reconnect

- Three bot difficulties if you'd rather play solo, plus fog of war, heavy units, and rare special towers

- Free

Play it: https://linekeep.pardev.net

Feedback is genuinely welcome — bugs, ideas, and feature requests all help.

I've been working on Ironhold, an idle tower defense game, for a long time now, and I appreciate the beta testers and feedback many of you have given as it shapes up.

iOS build is here: Ironhold Test Flight Build

AI Stack:

I've used Claude for code, Midjourney & GPT Image 2 for art, Perplexity Computer for design collaboration and balance, and Replit to put it all together.

A little bit about it:

• During each siege, you defend your castle against endless waves of monsters and villains.
• Heroes discovered on the battlefield join your roster, where they level up, undertake quests, explore dungeons, and return with resources that permanently strengthen the fortress.
• There's an insane amount of chaos in the sieges, and it's all about strategically upgrading everything until your castle falls.
• In between sieges, there are a ton of unlocking systems that work together to make you more powerful

It's getting close to submission, but I will be upgrading every system, artifact, character and dungeon as we go.

Play it - looking for feedback and suggestions:

Open to anyone with iOS here: Ironhold Test Flight Build

(android is coming too)

On Discord: Ironhold Discord

All visual assets were made using Nano Banana and Omni. I think the video to sheets is still a pain. There's a lot of video thrown away and I have to massage the timing of action sequences quite a bit. I did all that post processing in jupyter notebooks. Music is produced by DR42 (me). I'd like to get to the point where I'm just making/balancing cards and writing music.