Build Is Not What I Wanted
It built something — just not the thing in your head. Here is how to steer it back without throwing away the work.
Quick checks
- Read the "Heads up" block at the end of the summary. Anything left as a placeholder or unverified is listed there, and it often explains the gap.
- Say what is wrong in one sentence and send it. You do not need to repeat the original request — the chat carries it.
- Ask for the smallest possible change, not a rebuild. "The close button is on the wrong side" beats "redo the menu".
- If it is far off, roll back one turn with Restore Checkpoint (below), then restate the request with the missing detail.
- Press Play before deciding it is broken. Anything that only happens for a live player was never run — see the last section.
Correcting course without starting over
The chat is one continuous session. Revix is told to change the minimum — to fix the wrong property in place rather than delete and rebuild — so a follow-up is almost always cheaper and better than a restart.
Say what is wrong, concretely:
Close. Three things: the shop panel is too wide, the coin counter should sit top-right not top-left, and the buy button needs to be green. Change only those.Naming your own direction always wins over its taste. If you have a look in mind, say it up front and it will stop guessing:
Use a dark navy panel with orange accents and rounded corners, like a mobile simulator game. Keep that palette for everything you add from now on.Both of those are ordinary messages. There is no special "edit mode" to enter.
Why broad prompts drift
"Make me a tycoon" contains about forty decisions. Revix will make every one of them, quickly and confidently, and the odds that all forty match your mental picture are low.
The single highest-value habit is naming the outcome and one or two constraints, then letting it fill in the rest:
Build a dropper tycoon plot: one 60x60 stud plot, a conveyor that carries parts to a collector, and a cash counter on the wall. Server-side money only. No shop yet.Compare that with "make a tycoon". Same amount of typing, ten times the control.
Two smaller habits that help:
- Say what you do not want. "No round system", "do not touch the existing spawn", "leave my leaderstats script alone".
- Say where. "Put it under
Workspace/Map/Lobby" removes a whole class of surprise.
Undoing just the last turn
You have three separate ways back, from most surgical to most blunt.
1. Restore Checkpoint — reverts one turn's changes in Studio and removes that exchange from the chat.
- Hover in the gap just above the message you want to roll back. A small Restore Checkpoint pill fades in.
- Click it. A toast reads "Reverting turn changes...", then "Checkpoint restored successfully".
- Rewrite your request with the detail that was missing, and send it.
The plugin must be connected for this to reach Studio. It is only offered on turns that could have changed your place — never on an image or a 3D model.
2. Just ask. Plain language works, and it drives Studio's own undo:
Undo that last change.3. Ctrl+Z in Studio. Every change Revix makes plants a normal Studio undo step, so undo behaves exactly as it does for your own edits. Press it repeatedly to walk back through a build.
One honest caveat: a single operation that placed dozens of parts — a curved wall, a dome, a whole tower — comes back as one undo step, not one per part. A single Ctrl+Z can remove more than you expect.
The one thing you cannot undo: map generation
Generating a map wipes the existing terrain first, then builds the new world. Do not treat that as reversible.
The rule is simple: generate the world before you build anything on it. If you already have a build you care about and you want new terrain, save a copy of the place first.
Revix also plants its own trees, rocks and foliage as part of generation, so there is no need to ask it to scatter vegetation afterwards.
Splitting a big ask into reviewable pieces
If a request keeps coming back not-quite-right, it is usually too big to review in one go. Break it along natural seams and check each piece:
- Shape first. "Build the shop UI in StarterGui — panel, item rows, close button. No scripts yet." Look at it.
- Then behaviour. "Now wire it up: clicking an item spends coins, server-side." Press Play and try it.
- Then polish. "Make it feel better — hover states on the rows and a slide-in when it opens."
Each step is small enough that a correction is one sentence, and a bad step costs you one rollback instead of an afternoon.
The three suggestion chips under a finished turn are built for exactly this — they are the sensible next slice, phrased as a message you can send with one tap.
"It looks wrong" that is actually "it was never run"
Revix measures, it does not look. It cannot see your viewport, and it cannot press Play.
That means anything that only exists for a live player — a button click, camera behaviour, input, respawn — is checked by reading the code back against what was built, not by running it. It is required to tell you which parts it could not confirm, in the Heads up block and, when the internal check still objects, in a line beginning:
⚠ I could not fully verify this: …
So before reporting a build as broken:
- Press Play in Studio.
- Try the thing.
- If it misbehaves, come back and say what happened, and ask it to read the Output window:
I pressed Play and clicking Buy does nothing. Check the Output window and fix it.That is a much stronger bug report than "the shop doesn't work", and it is the loop Revix is designed to run.
Still stuck?
Post the prompt you sent, one screenshot of the result, and one sentence describing the difference from what you wanted. That is enough for someone in the Revix Discord to give you a better prompt: discord.gg/SDtaPGkUP4.