Newly optimized and doing great things, Journal Recognizer OCR - Document Ready!

[tribal sage]

https://chat.openai.com/g/g-T7bW2qVzx-journal-recognizer-ocr
W've been refining Journal Recognizer OCR, an efficient tool optimized for speed and practicality. This GPT excels in processing multiple images, adeptly recognizing words and maintaining paragraph integrity. Unlike other OCR tools, it avoids unnecessary line breaks in paragraph-style text, adding them only when encountering distinct, short lines. This feature significantly improves the practicality of transcribing large volumes of notebook content.
The output is streamlined into a single plaintext code block for easy copying and pasting into any application. You'll first see this code block, followed by a concise summary of the content, and, if applicable, descriptions of significant graphics or features in the scans.
An added advantage is the ability to customize recognition, particularly useful for writers. For instance, as a sci-fi author, I can instruct the GPT to recognize specific terms like 'Lamen' as a planet, not a misspelling, and 'Tem' instead of common names like Tom or Tim.
Journal Recognizer also handles printed text, incorporating style information into the output markdown within the plaintext block. It aligns graphical descriptions with the relevant text, enclosed in brackets, offering a cohesive reading experience.
Capable of processing up to 10 images at once (due to OpenAI's limits), this GPT is your go-to for converting handwritten or printed material into editable text. Whether for standard text or unique cases, just ask, and this tool will deliver efficiently.

[high belfry]

Can't you just use GPT Vision?

[tribal sage]

It is GPT-4V that is doing the heavy lifting. But while capable, it lacks consistency, until you explain exactly how you want it to capture the text and what to do with it. For years I've wanted to just take photos of a bunch of pages and have them consolidated into a single text block without questions and interventions. For this GPT, you don't have to give any text in the prompt at all. Up to 10 images, with handwritten words will be paragraph captured and output in a plain text code block, ready to be pasted into your document. Paragraphs that span pages are seamlessly stitched together. After the first prompt, if you want to customize some of the default behaviors, you can for example turn down the temperature on letter-perfect fidelity, to allow more common words that may not be the exact word used.