Photo by Sam Pak on Unsplash. Illustrative, not game footage.
“Make the enemy 25% faster” is a useful AI game experiment—if you can still get back to the version you liked.
That is the part of prompt-driven creation I would design first: a reversible editing loop. A quick preview makes an idea tangible. A protected baseline lets you decide whether the next idea actually improves it.
Start with a change you can recognize
Here is a deliberately small prompt for a hypothetical top-down prototype:
Increase only the chaser's movement speed from 80 to 100 units per second. Keep player speed, enemy spawn count, collision size, damage and scoring unchanged. Explain which events or properties you intend to edit before applying changes.
These are example design values, not a tested balance recommendation. Use your engine's actual movement units.
The important feature is the boundary. “Make it more exciting” could reasonably change five systems. This request gives you one proposed cause and a short list of things that should stay the same.
Before sending it, save a recoverable copy and note the current behavior. Can you dodge the chaser? Does taking a hit remove the expected amount of health? Can you finish and restart the round?
That is your baseline, not a demand for a large automated test suite.
Separate the conversation from the edit
GDevelop's current AI documentation provides a concrete example of this distinction. It describes an Auto edit switch: with it off, the assistant can inspect the project and suggest changes but asks for confirmation before each modification. With it on, it applies changes as it works.
The same documentation says the assistant can create or modify events, objects and variables. That is useful power, but it makes scope worth checking. A request about movement can touch more than the one number you had in mind.
For the example above, review the proposed change first. If the plan also adjusts spawning or rewrites the score system, narrow the request before continuing. After the edit, inspect the relevant events or properties rather than treating the assistant's explanation as proof.
This is a recommended workflow, not a claim that a particular generated build has passed testing.
Know which restore mechanism you actually have
GDevelop documents automatic restore points for cloud projects before messages and AI requests. It also states that this automatic restoration is not available for local projects, where keeping a saved working copy is recommended.
That distinction matters. “The tool has undo” is not enough information to entrust your only good version to it.
For a cloud project, identify the relevant saved point and understand that restoring overwrites the current project state. Preserve any later work you want to keep before rolling back. For a local project, make a separate baseline copy before the experiment. If you already use version control, a named commit can make the boundary explicit.
Other generators may handle recovery differently. Check their documentation rather than assuming this behavior is universal.
Compare before asking for another change
Play the same small scenario in both versions. Keep the map, starting position and other settings consistent where possible. Then answer two different questions:
- Did only the intended behavior change?
- Do you prefer the new behavior?
The first checks implementation; the second is a design judgment. A correctly implemented faster enemy might still feel less enjoyable. That is useful information, not a wasted experiment.
Keep a tiny note: request, baseline, changed version, observation, decision. If you keep the change, save a new baseline. If you reject it, restore the old one. Only then start the next experiment.
AI-assisted creation can bring more people into the playful part of development. Reversibility helps them stay there: experimenting without needing every prompt to be the final answer.
What is the smallest change that has made one of your prototypes noticeably more fun?
Full KRI ZEK article: The Dawn of Playable Dreams.
Download Altered Brilliance: https://play.google.com/store/apps/details?id=tech.krizek.alteredbrilliance
Global website: https://global.krizek.tech
Pre-register for Arzenal Human Health: https://play.google.com/store/apps/details?id=tech.krizek.arzenal
Join The Power Of Gaming: https://discord.gg/sbYSPcCqJn
Top comments (0)