By 1990 Omnipage 3 and its successors were 'good enough' and with their compact dictionaries and letter form recognition were miracles of their time at ~300MB installed.
In 2025 LLMs can 'fake it' using Trilobites of memory and Petaflops. It's funny actually, like a supercomputer being emulated in real time on a really fast Jacquard loom. By 2027 even simple hand held calculator addition will be billed in kilowatt-hours.
privatelypublic 3 hours ago [-]
If you think 1990's ocr- even 2000's OCR is remotely as good as modern OCR... I`v3 g0ta bnedge to sell.
skygazer 2 hours ago [-]
I had an on-screen OCR app on my Amiga in the early 90s that was amazing, so long as the captured text image used a system font. Avoiding all the mess of reality like optics, perspective, sensors and physics and it could be basically perfect.
privatelypublic 1 hours ago [-]
If you want to go back to the start, look up MICR. Used to sort checks.
OCR'ing a fixed, monospaced, font from a pristine piece of paper really is "solved." It's all the nasties of tue real world that its an issue.
As I mockingly demonstrated- kerning, character similarity, grammar, lexing- all present large and hugely time consuming problems to solve in processes where OCR is the most useful.
- Optionally extract images/graphs within the page and include them in the same LLM call.
- Optionally add a bit of context from neighboring pages.
Then wrap everything with a clear prompt (structured output + how you want graphs handled), and you’re set.
At this point, models like GPT-5-nano/mini or Gemini 2.5 Flash are cheap and strong enough to make this practical.
Yeah, it’s a bit like using a rocket launcher on a mosquito, but this is actually very easy to implement and quite flexible and powerfuL. works across almost any format, Markdown is both AI and human friendly, and surprisingly maintainable.
GaggiX 2 hours ago [-]
>are cheap and strong enough to make this practical.
It all depends on the scale you need them, with the API it's easy to generate millions of tokens without thinking.
fcoury 5 hours ago [-]
I really wanted this to be good. Unfortunately it converted a page that contained a table that is usually very hard for converters to properly convert and I got a full page with "! Picture 1:" and nothing else. On top of that, it hung at page 17 of a 25 page document and never resumed.
nawazgafar 5 hours ago [-]
Author here, that sucks. I'd love to recreate this locally. Would you be willing to share the PDF?
“Turn images and diagrams into detailed text descriptions.”
I’d just prefer that any images and diagrams are copied over, and rendered into a popular format like markdown.
deepsquirrelnet 2 hours ago [-]
Give the nanonets-ocr-s model a try. It’s a fine tune of Qwen 2.5 vl which I’ve had good success with for markdown and latex with image captioning. It uses a simple tagging scheme for page numbers, captions and tables.
captainregex 2 hours ago [-]
I desperately wanted Qwen vl to work but it just unleashes rambling hallucinations off basic screencaps. going to try nanonet!
david_draco 6 hours ago [-]
Looking at the code, this converts PDF pages to images, then transcribes each image. I might have expected a pdftotext post-processor. The complexity of PDF I guess ...
firesteelrain 6 hours ago [-]
There is a very popular Python module called ocrmypdf. I used it to help my HOA and OCR’ing of old PDFs.
> annotateai automatically annotates papers using Large Language Models (LLMs). While LLMs can summarize papers, search papers and build generative text about papers, this project focuses on providing human readers with context as they read.
I’ve been trying to convert a dense 60 page paper document to Markdown today from photos taken on my iPhone. I know this is probably not the best way to do it but it’s still been surprising to find that even the latest cloud models are struggling to process many of the pages. Lots of hallucination and “I can’t see the text” (when the photo is perfectly clear). Lots of retrying different models, switching between LLMs and old fashioned OCR, reading and correcting mistakes myself. It’s still faster than doing the whole transcription manually but I thought the tech was further along.
> Ironically, Ollama likely is using Tesseract under the hood.
No, it isn’t.
roscas 6 hours ago [-]
Almost perfect, the PDF I tested it missed only a few symbols.
But that is something I will use for sure. Thank you.
nawazgafar 5 hours ago [-]
Glad to hear it! What types of symbols did it miss?
4 hours ago [-]
KnuthIsGod 3 hours ago [-]
Sub-2010 level OCR using LLM.
It is hype-compatible so it is good.
It is AI so it is good.
It is blockchain so it is good.
It is cloud so it is good.
It is virtual so it is good.
It is UML so it is good.
It is RPN so it is good.
It is a steam engine so it is good.
Yawn...
GaggiX 2 hours ago [-]
>Sub-2010 level OCR
It's not.
ahmedhawas123 4 hours ago [-]
This may be a bit of an irrelevant and at best imaginative rant, but there is no shortage of solutions that are mediocre or near perfect for specific use cases out there to parse PDFs. This is a great addition to that.
That said, over the last two years I've come across many use cases to parse PDFs and each has its own requirements (e.g., figuring out titles, removing page numbers, extracting specific sections, etc). And each require a different approach.
My point is, this is awesome, but I wonder if there needs to be a broader push / initiative to stop leveraging PDFs so much when things like HTML, XML, JSON and a million other formats exist. It's a hard undertaking I know, no doubt, but it's not unheard of to drop technologies (e.g., fax) for a better technology.
bm-rf 4 hours ago [-]
For the purposes of an llm "reading" a pdf, it just renders it as an image. The file format does not matter. Let's say you have documents that already exist, a robust ocr solution that can handle tables and diagrams could be very valuable.
mdaniel 4 hours ago [-]
That ship has sailed, and I'd guess the majority of the folks in these threads are in the same boat I am: one does not get to choose what files your customers send you, you have to meet them where they are
treetalker 5 hours ago [-]
I presume this doesn't handle handwriting.
Does anyone have a suggestion for locally converting PDFs of handwriting into text, say on a recent Mac? Use case would be converting handwritten journals and daily note-taking.
nawazgafar 5 hours ago [-]
Author here, I tested it with this PDF of a handwritten doc [1], and it converted both pages accurately.
FYI, your GitHub link tells me it's unable to render because the pdf is invalid.
simonw 5 hours ago [-]
This one should handle handwriting - it's using Qwen 2.5 VL which is a vision LLM that is very good at handwritten text.
password4321 5 hours ago [-]
I don't know re: handwriting so only barely relevant but here is a new contender for a CLI "OCR Tool using Apple's Vision Framework API": https://github.com/riddleling/macocr which I found while searching for this recent discussion:
My iPhone 8 Refuses to Die: Now It's a Solar-Powered Vision OCR Server
If you use Docling, you can set your OCR engine to OCRMac then set it to use LiveText. It’s a good arrangement. You can send these as command-line arguments, but I generally configure it from the Python API.
ntnsndr 5 hours ago [-]
+1. I have tried a bunch of local models (albeit the smaller end, b/c hardware limits), and I can't get handwriting recognition yet. But online Gemini and Claude do great. Hoping the local models catch up soon, as this is a wonderful LLM use case.
UPDATE: I just tried this with the default model on handwriting, and IT WORKED. Took about 5-10 minutes on my laptop, but it worked. I am so thrilled not to have to send my personal jottings into the cloud!
abnry 5 hours ago [-]
I would really like a tool to reliably get the title of PDF. It is not as easy as it seems. If the PDF exists online (say a paper or course notes) a bonus would be to find that or related metadata.
s0rce 5 hours ago [-]
Zotero does an ok job at this for papers.
leodip 5 hours ago [-]
Nice! I wonder what is the hardware required to run qwen2.5vl locally. A 6gb 2cpu VPS can do?
mdaniel 4 hours ago [-]
It does not appear that qwen2.5vl is one thing, so it would depend a great deal on the size you wish to use
In 2025 LLMs can 'fake it' using Trilobites of memory and Petaflops. It's funny actually, like a supercomputer being emulated in real time on a really fast Jacquard loom. By 2027 even simple hand held calculator addition will be billed in kilowatt-hours.
OCR'ing a fixed, monospaced, font from a pristine piece of paper really is "solved." It's all the nasties of tue real world that its an issue.
As I mockingly demonstrated- kerning, character similarity, grammar, lexing- all present large and hugely time consuming problems to solve in processes where OCR is the most useful.
Trilobites? Those were truly primitve computers.
For each page:
- Extract text as usual.
- Capture the whole page as an image (~200 DPI).
- Optionally extract images/graphs within the page and include them in the same LLM call.
- Optionally add a bit of context from neighboring pages.
Then wrap everything with a clear prompt (structured output + how you want graphs handled), and you’re set.
At this point, models like GPT-5-nano/mini or Gemini 2.5 Flash are cheap and strong enough to make this practical.
Yeah, it’s a bit like using a rocket launcher on a mosquito, but this is actually very easy to implement and quite flexible and powerfuL. works across almost any format, Markdown is both AI and human friendly, and surprisingly maintainable.
It all depends on the scale you need them, with the API it's easy to generate millions of tokens without thinking.
I’d just prefer that any images and diagrams are copied over, and rendered into a popular format like markdown.
https://github.com/ocrmypdf/OCRmyPDF
No LLMs required.
https://threadreaderapp.com/thread/1955355127818358929.html
Python: PyPdf2, PdfMiner.six, Grobid, PyMuPdf; pytesseract (C++)
paperetl is built on grobid: https://github.com/neuml/paperetl
annotateai: https://github.com/neuml/annotateai :
> annotateai automatically annotates papers using Large Language Models (LLMs). While LLMs can summarize papers, search papers and build generative text about papers, this project focuses on providing human readers with context as they read.
pdf.js-hypothes.is: https://github.com/hypothesis/pdf.js-hypothes.is:
> This is a copy of Mozilla's PDF.js viewer with Hypothesis annotation tools added
Hypothesis is built on the W3C Web Annotations spec.
dokieli implements W3C Web Annotations and many other Linked Data Specs: https://github.com/dokieli/dokieli :
> Implements versioning and has the notion of immutable resources.
> Embedding data blocks, e.g., Turtle, N-Triples, JSON-LD, TriG (Nanopublications).
A dokieli document interface to LLMs would be basically the anti-PDF.
Rust crates: rayon handles parallel processing, pdf-rs, tesseract (C++)
pdf-rs examples/src/bin/extract_page.rs: https://github.com/pdf-rs/pdf/blob/master/examples/src/bin/e...
https://github.com/rednote-hilab/dots.ocr
Seems to weigh about 6GB which feels reasonable to manage locally
No, it isn’t.
But that is something I will use for sure. Thank you.
It is hype-compatible so it is good.
It is AI so it is good.
It is blockchain so it is good.
It is cloud so it is good.
It is virtual so it is good.
It is UML so it is good.
It is RPN so it is good.
It is a steam engine so it is good.
Yawn...
It's not.
That said, over the last two years I've come across many use cases to parse PDFs and each has its own requirements (e.g., figuring out titles, removing page numbers, extracting specific sections, etc). And each require a different approach.
My point is, this is awesome, but I wonder if there needs to be a broader push / initiative to stop leveraging PDFs so much when things like HTML, XML, JSON and a million other formats exist. It's a hard undertaking I know, no doubt, but it's not unheard of to drop technologies (e.g., fax) for a better technology.
Does anyone have a suggestion for locally converting PDFs of handwriting into text, say on a recent Mac? Use case would be converting handwritten journals and daily note-taking.
1. https://github.com/pnshiralkar/text-to-handwriting/blob/mast...
FYI, your GitHub link tells me it's unable to render because the pdf is invalid.
My iPhone 8 Refuses to Die: Now It's a Solar-Powered Vision OCR Server
https://news.ycombinator.com/item?id=44310944
UPDATE: I just tried this with the default model on handwriting, and IT WORKED. Took about 5-10 minutes on my laptop, but it worked. I am so thrilled not to have to send my personal jottings into the cloud!
Also, watch out, it seems the weights do not carry a libre license https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct/blob/main...
LLMWhisperer(from Unstract), Docling(IBM), Marker(Surya OCR), Nougat(Facebook Research), Llamaparse.