---

name: visual_prompt_scaling description: Build AutoVP frame-shaped visual prompt masks and apply bounded universal prompts to resized images.

Visual Prompt Scaling

Use this skill when implementing AutoVP-style input scaling and frame prompts. Do not use it for patch-token prompt tuning or model-internal prompts.

Inputs

Source image size, target image size, channel count, image array, and prompt parameters.

Outputs

Prompt width p, binary frame mask, centered image canvas, and prompted image.

Workflow

Compute p=(source_size-target_size)//2; build a frame mask; center the image; apply a sigmoid-bounded prompt only where mask is one. Reject target sizes larger than the source input. Record odd padding decisions.

Validation

Run python tests/test_visual_prompt_scaling.py or the Distiller skill-tree validator with tests.