Launch spectacle in capture region mode and send the catpured screenshot to openai-compatible endpoint to OCR and possibly translate

screenshot_llm.sh

#!/usr/bin/env bash

## Launch spectacle in capture region mode and send the catpured screenshot to openai-compatible endpoint to OCR and possibly translate
## Example usage: bind 'notify-send img "$(./screenshot_llm.sh 2>&1)"' to a hotkey
## Environment variables to set: OPENAI_API_KEY, optionally: OPENAI_API_BASE, OPENAI_MODEL, PROMPT

set -euo pipefail

[[ $# -ge 1 && $1 == --help ]] && exec sed -n 's/^## //p' -- "$0"
[[ $# -ge 1 && ( $1 == --verbose || $1 == -v ) ]] && set -x

# launch spectacle and allow use to select a region to screenshot, then encode png to b64
screenshot="$(spectacle -o /dev/stdout -r -b -n | base64 -w 0)"


[[ -z $screenshot ]] && exit 1 # user pressed esc

request='{
  "model": "'"${OPENAI_MODEL-gpt-4o}"'",
  "messages": [
    {
      "role": "user",
      "content": [
        {
          "type": "text",
          "text": "'"${PROMPT-What is in this image? Translate the text.}"'"
        },
        {
          "type": "image_url",
          "image_url": {
            "url": "data:image/png;base64,'"$screenshot"'"
          }
        }
      ]
    }
  ]
}'

# using curl -d @- <<< "$request" instead of curl -d $request because $request can be longer than argument max length
curl --no-progress-meter --fail -X POST \
  "${OPENAI_API_BASE-https://openai.com/v1}/chat/completions" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d @- <<< "$request" | jq -r '.choices[0].message.content'