Repository navigation
[FEATURE]: Add RapidOCR-json support — reading image text into the model when it can't see imagesI come from CHINA! I solution OPENCODE'S EYE problem, you can use RapidOCR-json, let bigpickle model (LLM) takes the picture's words!! #46785
Description
Problem
OpenCode refuses images for models that don't support image input, e.g. "Cannot read image.png ... this model does not support image input". Users who rely on text-only LLMs (like bigpickle) can't share screenshots of errors, logs, or documents.
Proposed solution
Integrate the local, offline RapidOCR-json engine as an optional image-to-text step:
- When a user attaches an image and the active model doesn't accept vision, run RapidOCR-json locally to extract the text.
- Replace the image part with that extracted text before sending to the model, so text-only models can still read image content.
- Fall back gracefully if no OCR engine is installed or recognition fails (vision models keep receiving the image unchanged).
Expected behavior
A user can paste a screenshot and, on a non-vision model, receive a reply based on the text read from the image — with no network and no image data sent to the provider.
Environment
- Windows, OCR engine: RapidOCR-json (offline, RapidOcrOnnx based)
- Region/backends: local text-only LLM (bigpickle)words!!
Metadata
Metadata
Assignees
Labels
No labels
Activity
27201904hui commentedon Sep 2, 2026
27201904hui commentedon Sep 2, 2026
27201904hui commentedon Sep 2, 2026
OC MODIFY BY ME! SET UP “eyes” HE can read the debug message from picture!
[-]I come from CHINA! I solution OPENCODE'S EYE problem, you can use RapidOCR-json, let bigpickle model (LLM) takes the picture's words!![/-][+][FEATURE]: Add RapidOCR-json support — reading image text into the model when it can't see imagesI come from CHINA! I solution OPENCODE'S EYE problem, you can use RapidOCR-json, let bigpickle model (LLM) takes the picture's words!![/+]github-actions commentedon Sep 2, 2026
Thanks for updating the issue! The title now has the
[FEATURE]:prefix and the description clearly explains the problem, proposed solution, and expected behavior. This issue is now compliant with our contributing guidelines.neriousy commentedon Oct 21, 2026
Thanks for sharing this, and nice work getting it running in your own build! We'd rather not bundle an OCR engine into opencode. You can get the same result without changing opencode: wrap RapidOCR-json in an MCP server that exposes an
ocrtool (https://opencode.ai/v2/docs/mcp-servers), save the screenshot to a file, and ask the text-only model to run OCR on that path. If you want images rewritten automatically before they reach the model, that's a job for a plugin (https://opencode.ai/v2/docs/plugins). Closing as not planned.