agentleFS
godot-fungodot-fun/gai/.agents/skills/image-to-text/SKILL.md

image-to-text

Skill210 starsChanged yesterdayMITView on GitHub

Converts one image into a faithful text description or OCR with the local 1.3B MiniCPM-V 4.6 GGUF model through llama.cpp, automatically preferring an available Vulkan GPU and falling back to CPU. Use when the user asks to describe, caption, transcribe, inspect, or extract visible details and text from PNG, JPEG, WebP, GIF, BMP, or TIFF images without a GPU or remote vision API.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run the official openbmb/MiniCPM-V-4.6-gguf Q4_K_M model locally with llama.cpp.

    line 13
  2. Distinctive:

    Read and follow dependency-manager before running commands.

    line 17
  3. Very specific:

    Run .ai/image-to-text/image_to_text.py through the default python manifest entry only.

    line 19
  4. Very specific:

    Before inference, check the llama-cpp-gpu runtime with --list-devices.

    line 21
  5. Very specific:

    Use --gpu or -GPU to require GPU inference and fail when unavailable.

    line 22
  6. Very specific:

    Use --cpu to bypass detection and force CPU inference for troubleshooting or comparison.

    line 23
  7. Very specific:

    Use --output description.md to also save UTF-8 text.

    line 41
  8. Distinctive:

    Never identify a real person from appearance alone or infer sensitive traits not explicitly visible as text.

    line 49

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Image to Text

Run the official openbmb/MiniCPM-V-4.6-gguf Q4_K_M model locally with llama.cpp. Do not send images to remote APIs.

Rules

Read and follow dependency-manager before running commands.

  • Run .ai/image-to-text/image_to_text.py through the default python manifest entry only.
  • Load the language model and multimodal projector only from .dependency/minicpm-v-4.6/model.
  • Before inference, check the llama-cpp-gpu runtime with --list-devices. Prefer it when a Vulkan device is reported; otherwise use llama-cpp-cpu.
  • Use --gpu or -GPU to require GPU inference and fail when unavailable.
  • Use --cpu to bypass detection and force CPU inference for troubleshooting or comparison.
  • Preserve observable facts and clearly distinguish uncertainty from fact.
  • Pass exactly one image file to --image per invocation.

Usage

.dependency/python/python.exe .ai/image-to-text/image_to_text.py --image C:\path\photo.png
.dependency/python/python.exe .ai/image-to-text/image_to_text.py --image C:\path\photo.png -GPU
.dependency/python/python.exe .ai/image-to-text/image_to_text.py --image C:\path\photo.png --cpu

Use --output description.md to also save UTF-8 text. See cli/image-to-text.md for copy-paste commands.

Output guidance

  • General description: mention scene, subjects, actions, composition, colors, and legible text.
  • OCR: preserve reading order and line breaks where practical; mark uncertain characters with [unclear].
  • Accessibility alt text: be concise and omit decorative speculation.
  • Structured extraction: request JSON explicitly and validate it before presenting it as machine-readable.
  • Never identify a real person from appearance alone or infer sensitive traits not explicitly visible as text.

Troubleshooting

  • Missing CPU runtime: verify .dependency/llama-cpp-cpu/llama-mtmd-cli.exe exists.
  • GPU unavailable: verify .dependency/llama-cpp-gpu/llama-mtmd-cli.exe exists and --list-devices reports a Vulkan device. Update the graphics driver if necessary.
  • Missing model: verify MiniCPM-V-4_6-Q4_K_M.gguf and mmproj-model-f16.gguf exist under .dependency/minicpm-v-4.6/model.
  • Out of memory: close other applications or use a smaller image; the official GGUF listing states approximately 2GB memory, but working memory varies with image size and context.
  • Slow inference: CPU speed, core count, memory bandwidth, and image resolution directly affect latency.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.