Why do the generated results sometimes not meet your expectations?

1. Situations You may encounter

When using combos.fun to generate games, you may encounter situations such as these.

• describes the desired game, but the generated result deviates significantly from the expectation

• The first generation was quite good, but subsequent adjustments have become increasingly difficult to control

• The content that was clearly emphasized was weakened or ignored in the new round of generation

  These situations are not uncommon, nor do they mean there is something wrong with your ideas. In most cases, ​​ the problem lies in the "way of expression" rather than the "creativity itself" ​​.

1. Isn't guesswork but executive description

The generation process of Combos.fun is based on the information you currently provide for continuous creation.

This means:

• Boo will automatically complete unspecified rules and settings during the generation process to ensure the results are executable

• Boo does not judge which requirements are more important and which are merely for reference

• Boo does not retain by default the designs that you think "should of course exist" in your mind

  When there are ambiguities, conflicts, or omissions in the description, the system can only continue generating based on the existing information, and the results may naturally differ from your expectations.

2. Many clarifications are actually not specific enough

From a human perspective, the following expression is clear:

• "The gameplay should be a bit addictive"

• "The overall feeling should be more sophisticated"

• "Create a complete and formal mini-game"

  However, these expressions cannot be directly translated into​rules, visual or interaction logic​. Deviation in generation often occurs not because you haven't written enough, but because​the information at key levels cannot be executed​.

2. How should game prompts be written?

In Combos.fun, ​​ the first round of prompts is not a trial run ​​, but rather setting a ​ long-term effective boundary ​ for the entire generation process.

A large number of successful cases indicate that:

Whether the initial prompt is clear directly determines whether subsequent revisions will be "fine-tuning" or "repeated overhauls".

The core purpose of the initial prompt is not to "write comprehensively" but to "set the direction" ​

There are only three things that the first round of prompts really need to accomplish:

• Identify what kind of game experience this is

• Tell Boo which directions are certain

• Exclude clearly undesired outcomes

  If these three points are not clear, even if details are supplemented later, it will be difficult to achieve stable results.

3. 5 Key Levels of Effective Prompts

4. What - Game Positioning Layer

What kind of work is this?

This is ​​ the highest priority ​​, and also the most easily overlooked level. If this level is not clearly defined, the system will default to generating a "general gameplay game", which is the reason why many results "seem okay but feel off".

Boo will first determine whether this is a game that requires a complete gameplay loop or an interactive experience focused on atmosphere, expression, and emotion.

If this layer is not clear, Boo will ​​ default to completing it as a general gameplay-style game ​​, automatically adding structures such as victory and defeat, levels, challenges, etc. This is precisely the root cause of many results being "seemingly okay but always feeling something is off".

The function of this layer is to ​​ tell Boo in advance: what kind of thing you are doing ​​.

Example**: ​**

• "Develop an experiential mini-game themed on 'Hatsuyume' targeting the Japanese market"
• "Develop a light strategy tower defense game that emphasizes strategic depth rather than pure thrill"

2. Why - Core Experience Goal Level

What feeling should players ultimately be left with?

When designing all details, which kind of experience should Boo prioritize?​What is the main feeling left by players after completing this game?​When generating gameplay, rhythm, and feedback, Boo will constantly make trade-offs.

The clearer this layer is, the less likely it is to "become very good at playing but feel off" in subsequent steps.

Examples: - "Allow players to complete a dream-like interactive experience of 'being heard and responded to' within an extremely short period of time" - "Each level in the game negates the player's habitual puzzle-solving experience"

3. Where - World and Scene Context Level

What specific environment does the game take place in?

When generating content, Boo automatically searches for a "reasonable occurrence environment" for the game.

The role of this layer is ​​ to allow all elements to have a common world premise ​​. If this layer is not clear, it is easy to have: the gameplay is correct, but the scene, narrative, and visual style conflict with each other.

Examples: - "Before dawn, you arrived at the First Dream Shrine at the foot of the mountain" - "The game scene is set in a space base station" - "Single-screen level, no complex scene transitions"

4. How-Rules and Player Behavior Layer

What exactly can players do? What can't they do? Where are the boundaries? ​

This layer is not writing about "system implementation", but describing: ​ what operations players can perform, the number of operations, the rhythm of operations, and whether actions can be repeatedly executed. ​

By clarifying this level, Boo will not automatically add redundant interactions or complex mechanisms.

Example:

• "Players can choose at most 3 symbols, and cannot continue clicking after the selection is completed."
• Each level contains 3–6 interactive elements, with only 1 being the real solution
• "Players can build defensive towers on fixed tower positions and use gold coins to upgrade them."

5. Style & Boundaries - Style and Boundary Layer

What kind of temperament should the overall presentation have? Which directions must be excluded? ​

The role of style and boundary layer is not to add details, but ​​ to reduce the possibility of errors ​​.

You need to clarify two things in this layer:

• Which direction should the overall content style lean towards?

• Which directions should not appear

  This layer is usually expressed through "rather than / do not / avoid / not appear", which can both constrain the visual and language style, as well as gameplay, rhythm, and feedback methods.

Examples: - Emphasize "brief interactive experience" rather than clearance or victory/defeat. - Avoid the conventional puzzle-solving structure of traditional key → door, password → number. - Feature minimalist black-and-white line drawings, pure black-and-white color scheme, and no grayscale. - Have a unified style, leaning towards hardcore tech rather than cartoonishness.

4. Common workflows for advanced users

When you are already able to consistently write well-structured prompts and gradually refine the generated results through multiple rounds of dialogue, what truly affects efficiency next is no longer "whether it is detailed enough", but ​​ whether a reusable method of use has been established ​​.

1. First confirm the generation target, then delve into the details

Before starting the generation, it is recommended to first clarify the core objective of this generation.

• First, describe the "core value" of this generation in one sentence
• All subsequent modifications will be judged as necessary based on this goal

2. Advance in stages rather than complete all at once

In the early stage, "ugliness" is allowed, but the structure must be correct, and visual design should not be repeatedly refined before the gameplay stabilizes ​

Breaking the generation process into several stages is usually more stable than completing it all at once.

Advanced users tend to break down the generation process into stages:

• Confirm whether the gameplay and structure are valid
• Reunify the art style
• Finally, supplement UI, feedback, and details

3. Accept the existence of intermediate states

It is normal to have incomplete or temporarily less-than-ideal results during the generation process. As long as the direction is clear and the structure is adjustable, these intermediate states are themselves part of the generation process, and there is no need to rush to achieve the final effect in one go.

Advanced users are more concerned about:

• Is this direction "correctable"?
• Is there room for further optimization?

4. Consciously accumulate reusable structures

Advanced users often consciously think during the generation process:

• Can this structure be reused next time?

• Which descriptions are universal?

• Which ones belong only to the current project?

  Over the long term, this will gradually form a set of personal prompt asset.

5. A Complete Advanced Usage Mental Model

Treat Combos.fun as a partner for continuous collaboration, rather than a one-time output tool.

When you start:

• Focus on structure rather than outcome

• Focus on the process rather than achieving it all at once

• Focus on collaboration rather than instruction

  You will find that improvement in generation quality occurs naturally