search

Subscore

ControlTesting methodology

Control measures how much freedom you get beyond the basic buttons and presets in the character creator.

We check whether you can write your own instructions, preview the character before finishing, and make changes after the character has been created.

  • 3 evidence groups
  • 3 scored tests

Control is organized into 3 evidence groups. Each group contains one scored test: Custom Prompts, Editing, and Preview. All three tests count equally toward the final Control score.

How the Control score is calculated

Every scored test gets a score from 0 to 10.

All three tests count equally. We combine their scores to calculate the final Control score. Control makes up of the Customization score.

  • 33.33%Custom Prompts
  • 33.33%Editing
  • 33.34%Preview
Weighted evidence groups (combined 100%)Control score of Customization score
View exact calculation

Each test gets a score from 0 to 10. We multiply the score by how much the test counts. We then add the points together to get the final Control score.

Control makes up of the Customization score. Customization makes up 15% of the overall performance score.

Scored tests and weights

Scored testHow much it countsSee scoring
Custom Prompts33.33%View
Editing33.33%View
Preview33.33%View
Total100%

How the score is calculated

  1. Each test gets a score from 0–10

  2. Test score × how much it counts

    Example: 10 × 33.33% = 3.33 points

  3. We do this for every test

  4. We add all the points together

  5. Final Control score

    Counts for 33% of Customization

Example calculation

We multiply each test score by how much it counts. We then add all the points together.

Scored testTest scoreHow much it countsCalculationPoints added
Custom Prompts10.0033.33%10.00 × 33.33%3.33
Editing8.0033.33%8.00 × 33.33%2.67
Preview5.0033.33%5.00 × 33.33%1.67
Final Control score—100%Add all points7.67/10

Special cases

Not Applicable

If a test does not apply, we remove it and spread its weight across the remaining tests.

Unknown

If we cannot verify a result, the test receives a score of 0.

Manual adjustment

In rare cases, we may adjust a score when the calculated result is clearly unfair. We always record the reason.

Why Control matters

A creator can have hundreds of presets and still give you very little real control.

You may want a specific hairstyle, personality, relationship, or background that is not included in the standard choices. Custom prompts let you explain what you want instead of picking the closest preset and hoping for the best.

Editing is just as important. You may spend time creating a character and then notice that the voice, personality, or appearance is not right. You should not have to delete the character and start again because of one bad choice.

A preview also helps you avoid wasting credits. You should be able to see what you are creating before the app locks everything in or charges you for the final result.

How we test Control

We use a paid account and create five test characters.

We use the same five characters for every Control test.

First, we enter our own written instructions and check whether the creator accepts them.

We then check whether a visual or written preview appears before the character is finalized.

After creating the characters, we try to change five important areas: appearance, personality, relationship, voice, and name.

Using the same characters keeps the tests consistent and makes it easier to compare different platforms.

Limitations and edge cases

A custom prompt being accepted does not mean the final character will follow it perfectly. This test only checks whether written instructions are available and accepted. We test how accurately generated images follow prompts under Images.

Previews can also work differently. Some apps show a full image, while others only show a short written summary. We record what type of preview is available.

Editing may depend on the setting. An app might let you change the name and voice but lock the appearance and personality.

Some apps charge credits when creating or editing a character. Those costs are tested separately under Pricing.

Character creators change regularly. Our results show what was available through the paid test account on the date of testing.

Evidence groups

Control has 3 evidence groups made up of 3 scored tests.

33.33%

Custom Prompts

1 scored test

Custom Prompts measures whether you can use your own written instructions when creating a character.

This gives you more freedom than only choosing from preset buttons.

Custom Prompts

Whether you can use your own written instructions when creating a character.

How we test

We create five characters using our own written instructions. We check whether the creator lets us enter the instructions and accepts them for all five characters.

What counts
  • A free-text prompt box
  • Written appearance instructions
  • Written personality or background instructions
  • Custom instructions accepted during creation
What does not count
  • Choosing from preset buttons
  • Editing the character after creation
  • A profile description that does not affect creation
  • A prompt box that is shown but cannot be submitted
Result shown

Yes — custom instructions were accepted for all five characters

Scoring

We use the result categories below.

ResultScore
Yes — custom instructions are available and accepted10/10
Limited — prompts work but have major restrictions5/10
No — custom prompts are not available0/10
Unknown0/10

In this example, Custom Prompts scores 10/10.

Evidence group 2 of 3

33.33%

Editing

1 scored test

Editing measures what you can change after the character has been created.

This matters because you may not notice a bad choice until you see the finished character or start chatting.

Editing

What you can change after the character has been created.

How we test

After creating the five test characters, we try to change appearance, personality, relationship, voice, and name. Each editable area counts as one passed check.

What counts
  • Changes saved through the normal character settings
  • Appearance changes
  • Personality changes
  • Relationship changes
  • Voice changes
  • Name changes
What does not count
  • Deleting the character and creating a new one
  • Temporary changes that disappear after refreshing
  • Changes only available through support
  • Editing a chat message instead of the character
Result shown

4 of 5 areas can be edited

Editing result: 80%

Scoring

More editable areas means a higher score.

Editable areasScore
None (0%)0/10
1 of 5 (20%)2/10
2 of 5 (40%)4/10
3 of 5 (60%)6/10
4 of 5 (80%)8/10
5 of 5 (100%)10/10

A result of 4 editable areas scores 8/10.

Evidence group 3 of 3

33.34%

Preview

1 scored test

Preview measures whether you can see the character before finishing the creation process.

A useful preview helps you catch problems before the character is saved or credits are spent.

Preview

Whether you can see the character before finishing the creation process.

How we test

We create five test characters. Before confirming each character, we check whether the app shows a visual preview, a written preview, both, or no useful preview.

What counts
  • A character image shown before final confirmation
  • A written summary of the selected character
  • A preview that updates when settings change
  • A preview available without completing the full creation
What does not count
  • Seeing the character only after creation
  • Marketing example images
  • A blank placeholder
  • A preview that does not reflect the selected settings
Result shown

Limited — a written preview is available, but no final image is shown

Scoring

We use the result categories below.

ResultScore
Yes — a useful preview is available10/10
Limited — a preview exists but has important restrictions5/10
No — no useful preview before creation0/10
Unknown0/10

In this example, Preview scores 5/10.

Back to Customization