Every big AI release for game devs, the day it lands.
Unreal, Unity, Godot and the AI tools that matter: what changed, what it costs and which one is best for the job.
https://t.co/1WDgr3zRs7
@MeshyAI Tripo's rig API lists quadruped, bird, snake and fish rig types, but its default rig model is biped-only, so set the model version.
We compare both, from their own docs: https://t.co/h2K100AFgD
Meshy will auto-rig a four-legged character. It gets one animation: a walk.
@MeshyAI docs: humanoids get 600+ presets and Text to Motion. Smart Rig creatures can't use the library.
For custom skeletons, like tail physics, the docs say to rig by hand in Blender or Maya.
Jump pressed before landing? A 0.1 second buffer can save it.
Store the press. If jumping becomes allowed before the buffer expires, jump once and clear it.
Clear buffered input on pause or death too, so it can't fire later.
His testing split: agents run a full simulation of the gameplay to catch logic bugs, and he plays it himself to find the UI and UX ones.
AI tools for automated game playtesting:
https://t.co/jqgvlyK1Xe
@raedbahriworld is building Hearthroad, an idle kingdom builder, by directing a team of AI agents that write the code.
His devlog explains the 6-person stall. The tutorial's one gatherer brought in about six berries a minute, exactly what six people eat, so nobody new moved in.
Devlog #6. Building a kingdom game with a team of AI agents, from hamlet to empire.
We fixed the first hour. If you did what the game's own goal said, you sat at 6 people for an hour. Now the goal names the real next step, and the same hour ends at 22.
#indiedev#gamedev
The bow duel simulates you at 30 fps and your rival at 144 fps. Hold Space to step out and shoot. Release to hide.
Press M to switch between frames and seconds. Seconds mode gives both bows a 0.5 second cooldown.
https://t.co/H3gU3BXyIN
A cooldown of 30 rendered frames gives unequal waits.
At 30 fps it lasts 1 second. At 144 fps it lasts about 0.21 seconds.
On use, set readyAt = gameTime + 0.5 seconds. Allow the next use when gameTime reaches readyAt.
Same image or prompt for both, first result kept, no retries, run 7 Nov. Costs on https://t.co/6P8iTEv7QK were US$1.20 a run for Meshy and US$0.20 to US$0.30 for Tripo.
All 11 pairs side by side, where dragging one model turns both:
https://t.co/h2K100Bd6b
We gave @MeshyAI 7.1 and @tripoai H3.1 the same input 11 times on fal defaults.
Meshy never went above 31,269 triangles. Tripo never went below 828,015, but every Tripo model came with normal and roughness maps. Meshy's had colour only.
Check your triangle budget first.
@RaymondDuke Using cheat codes to record clips faster is a smart trick. A rough video you can share today beats a polished one you never get round to making.
What would you change first for the next cut?
@OlexGameDev Code signing is a good job to hand to AI: lots of tiny pitfalls, and none of it is the game.
What did it end up setting up for your builds?
@raedbahriworld Going from 6 to 22 people in the same hour just by changing what the goal says shows how closely new players follow the text on screen.
How did you spot that the first hour was stalling?