Two ways to make a game with Claude
| Copy and change Saltline | Prompt your own game | |
|---|---|---|
| Starting point | The finished Saltline artifact | A blank conversation or project |
| Time to something playable | Minutes | Minutes for a simple game, hours for a polished one |
| Best for | Learning how a working 3D game is put together | A game that is actually yours |
| Can you publish it? | Not as your own: the original belongs to its creator | Yes |
Option 1: copy the Saltline artifact
Claude artifacts can be copied. Open the official Saltline page, click the arrow next to the title Saltline in the top-left corner, and choose the option to make a copy. You need to be signed in to Claude for the copy to be saved to your account. The copy opens in a conversation where you can ask Claude to change it.
Good first changes to try, one at a time:
- "Make it night with heavier rain and fog."
- "Add a minimap in the bottom-left corner that shows the three relays."
- "Give the player a shotgun on key 3."
- "Add a timer and show my best time on the title screen."
Each change teaches you how the game is organised: where the lighting lives, how the objective is tracked, how weapons are defined. That is the real value of the copy. Keep it private. Saltline is Felix Rieseberg's game, and a modified copy is still his work, so do not publish it or put it on your own site.
Option 2: make your own game with Claude
The more interesting route is to borrow the method, not the game. The Saltline promptworked because it did three things: it named a genre, set a visual quality bar, and kept the scope tiny. You can use the same structure for any small game.
Step 1: pick something small
One level, one mechanic, one place. "A first-person walk through a flooded subway station with a flashlight" is a good scope. "An open-world RPG" is not. The tighter the brief, the more effort the model can put into making it look and feel good.
Step 2: write the brief
A starting prompt in the Saltline style:
Build me a small browser game: [genre] set in [one specific place]. Make it look as good as you can: detailed lighting, materials and sound, all generated in code with no external files. Keep the scope tiny, one short level that takes about five minutes, with a clear goal and an ending.
Swap in your own genre and place. Ask for assets made in code, which keeps the whole game in one page that runs anywhere, the same way Saltline does.
Step 3: let it iterate
This is the step most people skip. Saltline did not come from one reply. It came from a long run in which the model built, tested and fixed its own work for about eight hours. In a chat, you can do a lighter version of this: after each version, play it and report what is wrong in plain words. With an agent tool that can run and test code on its own, you can give it the brief and let it go for longer.
Step 4: polish by hand
Even Saltline got a polish pass: Felix spent about two hours on weapon models, sound and gameplay after launch. Do the same. Play your game for ten minutes, write down the five things that bother you most, and ask for them one at a time. Specific requests ("the gun is too quiet", "enemies get stuck behind the red container") work far better than "make it better".
Step 5: share it
Publish your artifact so others can play it, or export the code and host it yourself. If it is good, post it with a short clip, the prompt you used and how long it took. That combination is exactly what made Saltline spread.
Tips for a better game with Claude
- Ask for a title screen and an ending. A start button and a clear finish make even a tiny game feel complete.
- Ask for loading feedback. Generated assets take time; a progress bar, like Saltline's "Waiting for first light", stops players from closing the tab.
- Ask for settings. A graphics quality option keeps the game playable on slower machines.
- List controls on screen. Saltline shows keyboard, controller and touch controls on its title screen. Players read it.
- Test on a phone early. If you want mobile players, ask for touch controls from the start rather than at the end.
What to expect
A first attempt will not look like Saltline, and that is normal. Saltline was a showcase run on a brand-new model, with a working day of compute behind it. Your first game will probably be simpler, but it will be yours, and a few rounds of focused feedback go a long way. For ideas, see what others built onFable 5.2 games, and read how Saltline was madeto understand the run behind it.