| description | Guidelines for evaluating the addition or modification of text in images. |
| name | evaluating-text-additions |
| type | skill |
Evaluating Text Additions
When a prompt asks to add or change text (e.g., "Add the word 'begging'", "Add the letters 'CBS'"):
- Exact Spelling: Verify that the text in the image matches the prompt EXACTLY. Misspellings (e.g., "beging" instead of "begging") are severe instruction-following failures. Use visual QA tools to read the exact text.
- Legibility and Clarity: The text must be readable. Distorted, stretched, or blurry text reduces both instruction following and visual quality scores.
- Integration: The text should look like it belongs in the scene. It should match the perspective, lighting, and texture of the surface it is placed on, rather than looking like a flat digital overlay.
- Unintended Characters: Check for extra characters, punctuation, or gibberish added around the requested text.