Is there a word count per image for transcriptions? I have some manuscripts written in a very small italic hand, sometimes maximizing all the space on a page (see image). For especially full pages the Leo transcriptions stop before the end of the text, in a systematic way, implying there’s a word count limit (red line in image is where the transcription stops). This isn’t necessarily a problem – images can be cropped and split in parts as needed – but it would be useful for the user to know these limitations upfront, so they can plan accordingly and make best use of each image credit. I couldn’t find anything about word count limits per image in the guidance, but may be missing it
Hi Taylor, thanks for reporting this. Yes there is a output length limit (due to memory requirements) which we recently increased.
Would you mind sharing the URL for this image? It should look something like: https://www.tryleo.ai/document/67c8cfd7-bc98-4114-8a64-4787452a89e6?imageId=e999344b-24ca-46bd-acf0-5ac975ebaa09, but with different long identifiers. Only me and Jon will be able to open the URL.
I’ll take a look at whether this was transcribed before or after we increased the limit, and I’ll see if it’s technically possible to increase the limit enough to comfortably cover documents of this length.
As a workaround, you can split this image into 2 using the cropping tool in the image viewer, but obviously less credit-efficient than being able to do it all in one go.
Let me know if that doesn’t work. I tried these on 5 May, so with the earlier v0, and ended up cropping the images to get full transcriptions. So it wasn’t a problem, just confusing at first. This is a rather extreme example - this guy’s handwriting is abnormally small. But if in the end, there is a limitation on output length (understandable), I think it should be enough to say this somewhere so users can plan their image credit use accordingly. Cheers!
Thanks Taylor! And yes it looks like the updated version can transcribe a whole page of yours, though it may be sensitive to repetitive hallucinations when there’s so much text (we’re working on this).
And it sounds like we need some kind of informative message when the output hits the limit to avoid future confusion – we’ll get this added too.
