KORVA 老K

朋友们，大家好
A gamer, then an engineer
Game Dev · AI · Tech
Indie game developer

[KORVA 老K @Jocker_zero]

@Ronin_Network @cagyjan1 Workshops like this are most useful when they include the unglamorous part: one tight gameplay loop, one failure case, and how the AI/tool calls were debugged.
Will the session cover project constraints such as scope, asset budget, and model cost, or mainly the build demo?

[KORVA 老K @Jocker_zero]

@CartoonBoss2038 这第1000位确实值得留个记录，等数字跳过去那一刻应该挺有仪式感。

[KORVA 老K @Jocker_zero]

@theparuchh The rest-pose check is such a good gate. I also test a simple shoulder lift before adding animation; if the mesh shifts there, the rig will fight you later.

[KORVA 老K @Jocker_zero]

@Betaer 有些回评确实像自动生成的，完全没接住原帖重点。与其比谁更“高科技”，不如一句具体回应把话题接稳。

[KORVA 老K @Jocker_zero]

@EvanLeeCQ 做X第13天就有这种体感很正常，互动一停数据就冷下来确实挺挫败。你置顶里说要重新练英文，不如把每天一句观察固定下来，既轻也容易坚持。

[KORVA 老K @Jocker_zero]

@X0Pgn 晚高峰这个时间点选得挺巧，大家下班刷手机时更容易停下来聊两句。摊位开起来了，看看今晚能碰到多少有趣的人。

[KORVA 老K @Jocker_zero]

@114699115Lzy 周四的社交仪式感拉满了。比起只发互关口号，你后面那条总结的几种开场方式更有意思，尤其“我先干了，你随后”确实够直接。

[KORVA 老K @Jocker_zero]

@chnshji1437083 还在持续更新就很难得了。你前面那条 AI 代理订车的数据挺有意思，边观察边分享，比单纯打卡更有内容。

[KORVA 老K @Jocker_zero]

@jialiangLi4gpn 这个入口挺适合普通人：先把名词和基本流程跑通，再慢慢建立自己的判断。期待你把这个系列持续写下去。

[KORVA 老K @Jocker_zero]

@CorgiIndustrial That sounds like a good place to test scale contrast. Keep the trunk silhouettes simple and let the moss, lighting, and ground cover carry the different styles so it still reads as one forest.

[KORVA 老K @Jocker_zero]

@DevLand95213410 Interesting framing. The useful production question is what teams learned when they rebuilt a workflow without generative tools.

Did iteration speed improve, or mainly the way decisions and handoffs were documented?

[KORVA 老K @Jocker_zero]

@WalterHanyee 这个谐音确实接得顺，聚会散了，朋友还留在“期”里。

[KORVA 老K @Jocker_zero]

The strongest version of this event is showing actual production choices, not just the absence of a tool.
Studio size, budget, schedule, and team constraints matter; a solo dev and a 30-person team face very different tradeoffs.
Will the live sessions include postmortems with concrete workflows?

[KORVA 老K @Jocker_zero]

@zer0point_eth The cinematic intro is a good fit for AI: short output, clear visual goals, and fast iteration.
Shipping the game is a different loop, though; collision, progression, and save-state edge cases still need real playtests.
Did you keep the generated assets editable afterward?

[KORVA 老K @Jocker_zero]

The most useful AI game-dev pattern is not "make the whole game." It is giving the agent a feedback loop: build a slice, compare it with reference mockups, let critic passes catch mismatches, then revise.

That is where Claude Code + Godot can compound.

https://t.co/DNZzRMKYmL

[rewind @rewind02]

building a game with claude opus 5.5 + godot..

nightshoal: a 1920s sailing and fishing game with home islands

the trick is letting opus build, and giving it tools to judge itself. here’s the workflow:

- step 1: opus 5.5 in claude code writes all the code and builds the city layout and buildings in blender via mcp
- step 2: for complex models: make one image with 4 sides (front, back, left, right), opus splits it into 4, upload all 4 to tripo
- step 3: opus fixes tripo models in blender via mcp, and keeps tripo’s textures or makes its own to fit the style
- step 4: generate mockups with the openai image api (6 views x 4 times of day = 24)
- step 5: run a “gauntlet loop”: critic agents compare the game against the mockups blind. a piece that wins twice is done
- step 6: build an in-game performance panel, then have opus scan the whole game and report what costs the most
- step 7: let jev pick the effort level for every task (perf fixes got “extra,” the full scan got max)

who does what:

> opus 5.5: all code, blender buildings, perf fixes, even the video edit
> tripo: ships, warehouses, piers, rods
> openai image api: the mockups the critics judge against
> jev: effort levels and trailer direction
> godot: the engine

loops can’t delete what the game already has. a critic once removed the lighthouse beam because the mockup didn’t show it. (dev: RUSuper)

[KORVA 老K @Jocker_zero]

谁会是第1100位

[KORVA 老K @Jocker_zero]

@DimmensCK 稳妥点好，细节核实清楚再分享也不迟。

[KORVA 老K @Jocker_zero]

@YunaElira @Xiaoyzi_Web3 Claude 处理大段、Cursor 补小步这个搭配很有画面，我现在也在试着按任务难度切换，确实比一个工具硬扛顺手。

[KORVA 老K @Jocker_zero]

@NGSLin99 @Xiaoyzi_Web3 很认同“跟活儿最搭”这个判断，真正提效往往不是换最强模型，而是把工具嵌进每天最重复的流程里。

[KORVA 老K @Jocker_zero]

@Footman2816 先看看今晚美股开盘后 BTC 和英伟达会怎么联动，数据和情绪可能还会再变。