Reflections on "Best Game app"

Gemini API Developer Competition
1

Winner of “Best Game App” here (well one of the “lesser” prizes, but still, one is allowed to reflect!). Some random thoughts:

Positives:

  1. Strong artistic vision: I am to make an indie/doujin game. It should not look out-of-place in Comiket. This is who I am and what I am good at. I should use a retro game engine – either pico8 or pyxel. Given the AI things the Python-based pyxel it is then. Btw pyxel is awesome, and you should check it out.
  2. Beelined strategy: I am only here to win the best game category – I aim for no more, no less. I have decent ROI if I aim for that – two months for possibly $50k – not shabby. This is “easy mode” compared to Ludum Dare.

Things that did not go too well:

  1. Feature creep (as always). So much stuff didn’t make it into the final video or game. I brainstormed too much, implemented too much, and burned out a lot. Many boba teas died in the making of this video.
  2. The LLM fun times. Each step would break with like a 5-10% chance because… LLMs being LLMs. You can optimize prompts somewhat but cascading failures killed some ambitious ideas. Thank god for error catching - at least the whole game world didn’t explode.
  3. Mid-dev requirement surprises. Found out halfway through that we needed to check boxes for “accessibility, environmental sustainability” - well, I guess my game is better than Disco Elysium now in that my game promotes more on environmental sustainability…?

Random remarks:

  1. This is strangely like indie game making. Good games die all the time, and good trailers can get people excited like crazy. Unfortunately, hype sells. That’s why so much money is spent on trailers for many games. So I woke up my old cynical self and started producing videos – this is the inescapable fate of indie game devs.
  2. It is very very good that this competition has a game category. Games you sell in Comiket tend not to fare well in the south in hackathons. I am glad that this competition format still allowed a very indie-looking doujin-inspired game to win. I do feel validated. Doujin makers from the world unite!
  3. I regretted the name “Pen Apple” – for perhaps obvious reasons – but welp I am bad at naming things and I do like how hilariously “pineapple”-sounding the name is.

AI thoughts:

  1. I tried using LLMs as a sort of “world model” - they’re impressively great at that, but sometimes hilariously bad. Got to say though, Gemini Flash really impressed me! I use gpt4o-mini more nowadays, but Gemini Flash was (as far as I know) the first “small very-capable LLM” that was very cheap yet more than capable.
2

Did u create the assets yourself or downloaded from somewhere else? If you made them yourself that is crazy tbh. Congratulations.

3

Mostly made by (the extended mind of) “myself” with Imagen and tons of aseprite modifications + filters + dithering programs. Unfortunately/fortunately it is hard to not make assets oneself when one is using pico8 or pyxel.

The fonts and icons were not self-made. Sorry I forgot which icon-pack I used on-top-of-my-head (terribly sorry – but I think it was a paid asset so I guess I am fine :upside_down_face:).

4

Please don’t apologize, nothing wrong with using the tools at your disposal. It looks good and game evaluations are subjective by nature. Good luck.

1 Like
5

Oh no I was mostly feeling bad for the fact that I forgot which icon-pack I used. I know first-hand sometimes it feels a tad-bit hurting that one’s assets is (legitimately, IIRC I paid for the asset according to the license) used but one is not formally credited.

1 Like
6

Congrats on the win! Totally agree.

I was gunning for best game app as well, so I did a pretty thorough “competitor analysis”. I did consider PenApple one of my favorite submissions (didn’t spot Outdraw at the time). I was hoping to outrun you on the scoring; I put in a day or so into accessibility lol.

My strategy was casting a wider net, maximizing score, and playing the odds. Creative prize, game, android, firebase. Technically, I might have one of the better task list apps in the competition, arguably better than Vite Vere, but the video did not sell that. Congrats to Guido and his team for nailing that. But I was basically pulling in the direction of both impact and games and not quite scoring the #1 in either. Can’t play the odds in a global competition with a large prize pool.

100% agree on the trailer strategy. This was mine:

I think it was pretty catchy, but in retrospect, I wish I spent three days on it instead of one lol.

I brainstormed too much, implemented too much, and burned out a lot.

IMO, this might have been why you did well. I joined a bunch of other hackathons after Gemini, and often played it safe and delivered everything as planned. But often lost to more inexperienced people who went absurdly over scope. I do think subtle animations in the demo count for a lot of points.

I tried using LLMs as a sort of “world model” - they’re impressively great at that, but sometimes hilariously bad. Got to say though, Gemini Flash really impressed me!

Gemini Flash is incredibly good on “rendering” tasks - writing descriptions, dialogues, etc. I might have been a little obsessed with showcasing its writing ability. I spent the first two months on ‘superhero high school’ and it did a pretty good job handling things like relationships and superpowers. But I scrapped that because “impact” and because text-based games don’t make for good demos.

Anyway, congrats! Let me know if you want to make AI games together; it’s so hard to find people who enjoy making both games and working with LLMs.