VLM Image Helper

Treat this skill as a visual aid for the model, not as a general image editor.

Use scripts/image_helper.py to create a clearer intermediate image, then re-run analysis on that result.

Core Workflow

1. 1. Start from the original image path, a raw base64 string, or a data URI.
2. Apply the smallest transformation that is likely to remove the ambiguity.
3. Prefer semantic crop presets over manual coordinates unless the exact box is already known.
4. Return the processed image as a file or base64, then re-read that result.
5. If the image is still unclear, iterate once with a tighter crop or stronger zoom instead of stacking many edits at once.

Quick Commands

CODEBLOCK0

Choose the Next Action

• - Text is sideways or upside down: use --rotate.
• Only one region matters: use --crop-preset first, then add --scale-preset.
• Small text or icons are hard to read: use --scale-preset x2 or x3.
• Contrast is weak or edges are fuzzy: use --auto-enhance, or manually tune --contrast and --sharpness.
• Another tool needs inline image data instead of a file path: add --base64.
• The source image arrives as raw base64 or a data URI: use --input-mode auto or force --input-mode base64 / data-uri.

Input and Output Rules

• - Accept a file path, raw base64 string, or data URI as input.
• Return a file with -o or return inline base64 with --base64.
• Allow passthrough output with no edits when the only goal is format conversion or path-to-base64 conversion.

References

• - Full CLI reference: INLINECODE15
• Crop and scale preset table: INLINECODE16

Prerequisite

Install Pillow if it is missing:

CODEBLOCK1