Roqer (@RoqerDev) header
Roqer (@RoqerDev) profile picture

An open-source AI agent for Roblox Studio. It builds, scripts, and playtests in your place using your ChatGPT or Claude subscription.

Joined May 2022
Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)Photo by Roqer (@RoqerDev)
Free week of Claude Code referral. claude.ai/referral/F6L8ZKG5Y…
1
1
3
4,126
All used!
2
1
305
Roqer retweeted
Free week of Claude Code referral. claude.ai/referral/F6L8ZKG5Y…
1
1
3
4,126
Roqer retweeted
Roqer now supports Antigravity (Gemini) and Rojo, plus a bunch of other changes and fixes. Thanks to all our contributors! github.com/S4US/Roqer
2
2
9
1,405
Roqer now supports Antigravity (Gemini) and Rojo, plus a bunch of other changes and fixes. Thanks to all our contributors! github.com/S4US/Roqer
2
2
9
1,405
Roqer retweeted
Replying to @RoqerDev
made using Gemini 4 Argon shadow drop
1
4
295
Roqer retweeted
"Arcane Gatling" VFX made by Opus 5.5 with Roqer github.com/S4US/Roqer
8
5
87
5,035
"Arcane Gatling" VFX made by Opus 5.5 with Roqer github.com/S4US/Roqer
8
5
87
5,035
Roqer retweeted
Anyone interested in macOS and Linux support for Roqer?
3
1
4
1,204
Anyone interested in macOS and Linux support for Roqer?
3
1
4
1,204
Roqer retweeted
Magic Missile VFX, made entirely by Haiku 5.5 on max effort. Custom textures from Blender, built in Roblox Studio. Not bad for the small model. github.com/S4US/Roqer
10
3
79
6,727
Magic Missile VFX, made entirely by Haiku 5.5 on max effort. Custom textures from Blender, built in Roblox Studio. Not bad for the small model. github.com/S4US/Roqer
10
3
79
6,727
Roqer retweeted
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released. On average, it costs around 75% less to run than Claude Haiku 4.5.
1,605
3,252
2,431
45,131
6,837,700
Roqer retweeted
Opus 5.5 high vs GPT-6 Astra extra high. Same prompt. Which one is better? github.com/S4US/Roqer
34
60
2
1,508
90,648
Opus 5.5 high vs GPT-6 Astra extra high. Same prompt. Which one is better? github.com/S4US/Roqer
34
60
2
1,508
90,648
Prompt: "Make an arcane tarot skill for an anime battlegrounds game. A deck of ornate cards appears floating in front of the player's chest and fans out into a wide arc, each card spinning in place with a faint glint along its edge. Three cards slide out of the fan one at a time and flip face-up with a sharp flash, each showing a different glowing arcane symbol, while the rest of the deck dissolves into sparks. After a short beat, the three cards launch toward a target point about 45 studs away, spinning edge-first like blades and curving in on slightly different paths. They land one after another, each one stabbing upright into the ground and projecting its symbol as a ring of light beneath it, and once all three are planted, lines of light connect them into a triangle that collapses inward and detonates in a sharp burst of runes and shredded card fragments. It should feel mysterious, elegant and precise, with a deep violet, midnight black and burnished gold palette with sharp white highlights on the symbols and card edges, not soft or whimsical. Draw custom textures in Blender (the card back, three card faces with distinct symbols, the edge-glint trails, the symbol rings, and the burst fragments) and preview them in Studio before uploading anything; only upload the final set. Put it under ReplicatedStorage.VFX and tell me how to play it."
1
97
10,181
Roqer retweeted
What if we put the same level of detail, into every prompt, that we used to put into making the asset by hand in blender? What if we spent the same amount of time curating how it should look in words, as we did making how it looks in 3D models) This is the question that I’m asking myself, because seeing people who prompt in detail like this (his prompt is in the replies) remind me of the contrast to my own forceful One-or-Two Shotting entire games. I use custom instructions, I use subagents, I use a detailed first prompt. I often avoid curating dozens of small interactions like this. But this could be avoidance of quality itself. It makes me ask whether another way would yield better results, even though it’s Hardworkmaxxing the prompt, and would take more time. Worth testing. Goes against my patience. But valuable. I think the end game right now is not black and white. I’m sure it’s some sort of hybrid between one shotting the base game and loop, then hyperfixating on promoting high fidelity models, VFX, animations, and systems that are closest tied to that loop and player experience.
Same prompt. Astra vs. Opus 5.5. Two models, two very different results. Which one did it better? github.com/S4US/Roqer
3
1
1
53
5,655