Writing & MediaFree
Handwritten Draft Transcription for Authors
Send photos or scans of your handwritten draft, and get back the transcribed text with every unclear word and passage flagged for you to check.
River reads the photos or scans in the order you send them and produces a transcript with your chapter and page breaks preserved exactly where your notebook has them. Every word or passage the model cannot confirm gets marked inline, using the same bracket convention archives and editors already use: a probable reading gets a question mark, an unreadable stretch gets [illegible]. Nothing gets a confident, silent guess. You also get a page register, one row per page, so you know before you open a single photo which pages actually need a second look.
Unlike the handwriting-recognition platforms built for archives and libraries, this is not aimed at somebody else's century-old cursive. Transkribus is explicit that its own model produces a first draft for a crowd of volunteers to correct, built for institutions digitizing wills and letters nobody alive can clarify. Your draft has exactly one person who can resolve an ambiguous word from memory, and it is you. So there is no volunteer queue, no public archive, no XML export. There is a transcript and a page register, shaped to feed straight into your own revision instead.
Built for novelists, memoirists and nonfiction authors who draft in longhand before anything is typed. Jeffrey Archer has handwritten every word of more than thirty novels on lined pads before his office types the draft up, and he is far from alone in the habit. Reach for this the day a notebook is full, not partway through a chapter, so the page structure stays intact. Nonfiction authors whose notebooks hold verbatim quotes to clear later can carry those straight into a research and citation register.
A transcription accuracy number hides exactly where it's wrong
Handwriting-recognition vendors publish an accuracy number and rarely publish what happens next. Transkribus states plainly that its own model produces a first draft at 70 to 95 percent accuracy, then hands that draft to a queue of volunteers to correct. That is still a correction pass, done by other people, sized for an institution digitizing thousands of pages at once. A novelist finishing one notebook has neither the queue nor the spare week that implies, so the real question was never whether the draft needs checking. It is which handful of pages actually need it.
Take a real-shaped draft: Ottoline Reyes-Fairweather's novel Ember and Salt, 340 handwritten pages across 28 chapters, roughly 82,600 words. Score every line for confidence, and 597 of 9,180 lines come back flagged, 6.5 percent, spread across 151 pages, 44.4 percent of the manuscript. The other 189 pages need no second look at all. One chapter breaks the pattern: chapter 19, drafted over nine pages during a wildfire evacuation, holds 43 of those flags, 7.2 percent of the book's total from 2.6 percent of its pages, 2.7 times the manuscript's own average density.
The notation follows a convention archives already settled on, not one invented for this page. The National Archives' own transcription guidance writes [illegible] over a word nobody can make out, and the word itself plus a question mark for a confident guess. Reusing it means any editor or writing partner who opens the transcript already knows what the brackets mean. From there the work is narrow: reopen your own photo for the 151 flagged pages, correct each marked word from memory, and leave the other 189 exactly as transcribed.
How it works
Send your pages
Attach the photos or scans in the order they were written, and say how legible you think the handwriting runs.
It transcribes and scores
Every page is read line by line, and anything the model cannot confirm is flagged rather than smoothed over.
Check the flagged pages
The register sorts flagged pages first, so you reopen your own photos only where a check is actually needed.
Keep working in chat
Ask for a chapter rechecked, a name searched across every page, or the transcript exported for your next draft.
What you get
- A transcribed Doc with your original chapter and page breaks preserved, exactly where your notebook has them
- Every uncertain word or passage marked inline as [word?] or [illegible], never silently guessed at
- A page-by-page register in a Sheet, flagged pages sorted first, so you know exactly what to recheck
- Character, place and other proper names checked for spelling consistency across every page
- A flag count and share for every chapter, so a rough patch of handwriting stands out
- Nothing rewritten, corrected or polished. What you get is what the page says, marked where it is unsure
Common questions
Will it just guess at words it can't read?
No, and that is the entire point of the brackets. A word it cannot confirm becomes [word?] if it has a probable reading or [illegible] if it does not, following the same convention archives use for damaged manuscripts. A confident sentence written over unclear handwriting is how a transcription error survives into a manuscript nobody rereads against the original page.
How accurate is a handwriting transcription, honestly?
Handwriting-recognition vendors report ranges rather than a single figure, and Transkribus states its own model produces a first draft at 70 to 95 percent accuracy before any correction. Neat, consistent handwriting sits at the top of that range and a rushed page sits lower. The honest answer is that some pages will need a second look, which is exactly what the flags are for.
Does it clean up my prose while it transcribes?
No. It transcribes what is on the page, misspellings, crossed-out words, invented character names and all, because that is a transcription and not a copyedit. Cross-outs are recorded rather than removed, so you can still see what an earlier version of a sentence said. If you want the prose itself improved once it is typed up, that is a separate, later pass.
What happens if a whole page is too far gone to read?
It marks the whole stretch [illegible] rather than forcing a guess across it, and the page register carries a status for that page so it does not read as clean by accident. If a page is bad enough, the register will tell you to reshoot it in better light rather than transcribe a page that was never going to resolve.
My handwriting is bad. Should I even bother?
Rough handwriting is the actual reason to use flags rather than a plain transcript. On a draft this page walks through, 44.4 percent of pages held at least one flagged line, which sounds high until you notice it also means the other 55.6 percent needed nothing at all. Bad handwriting raises the flag count. It does not raise the number of pages you have to open blind.
What do I do with the transcript once I have it?
That depends on where the book is. If a developmental letter is coming back on this draft, the revision plan sequences the notes once it arrives. If the book is not sold yet, the proposal comes first. If it is finished and going to agents or publishers, the submission pack formats it to each recipient's own rules.
What if what I have isn't handwritten, like a recording or somebody else's scanned records?
Then this is the wrong tool, on purpose. A recording gets its own confidence scoring on audio rather than ink, which the interview transcript workup handles. Somebody else's scanned records, arriving by the thousand rather than as your own notebook, are a search-and-index problem that the document dump index is built for instead.
Handwritten Draft Transcription for Authors
Fill in the form and your workspace opens with the work already underway.