I hate taking notes when playing D&D, so I made a website called Scrybe Quill (https://www.scrybequill.com/) that automatically generates immersive, narrated recaps of your TTRPG sessions using just a recording. Think Witcher 3 loading screens if you’ve ever played it.
As of recently, I updated the site with some super handy features like a notes section with descriptions of all NPCs, locations, and items that you came across during the session. There’s also an outline section that breaks down the major parts of the session into point form notes (really helpful for GMs). Maybe less practical, but super fun and often hilarious is that the site highlights the most iconic quotes spoken during the session.
The idea started out as a fun side project where I made the recaps manually for my group and I ended up turning it into the website.
All the music on the site is either creative commons, or made by real composers (I even commissioned a real composer and hope to again) with all of the composers credited fully at Scrybe. The TTS narrators also get a cut every time their voice is used too.
I’m the only one working on the site, so if you checked it out and let me know what you think it would mean the world to me.
Thank a bunch!!
I’m also a video editor and whipped up this explainer video if you’re interested:
Yes, I also want to give an overview of the tech behind Scrybe:
All of this is possible due to structured outputs offered by certain LLM APIs. I actually originally used OpenAI’s o1-preview for the recaps because I found its writing to be the best at the time, but it didn’t have the option for structured JSON outputs and it was quite expensive. I’ve since been using Gemini 2.5 Pro.
In short, here is the pipeline for how the site works:
Transcribe the uploaded session recording to text using OpenAI’s Whisper-V3 model (distil-large-v3 for English and whisper-large-v3 for multilingual)
Condense the transcription into a recap and extract notes using an LLM (currently using Gemini 2.5 Pro)
Narrate the text using ElevenLabs
Combine the narration and your selected background music track into an immersive, narrated video recap
I’ve been keeping up on the bleeding edge of transcription tech and LLMs and often update my site to whatever tops the leaderboards.
In terms of cost, transcribing ~3-5 hours on average of a TTRPG session is expensive when using an API. This alone accounted for nearly 60% of the cost of generating a recap, with another 25ish% for the narration generation and 15% for the LLM costs. Swapping to Whisper distil-large-v3 brought the transcription costs down to maybe 20% or so which was a huge win without any significant transcription accuracy decrease.
Interesting enough, Nvidia just released parakeet-tdt-0.6b-v2, which supposedly is more accurate than whisper-large-v3 and can transcribe 1hr of audio in 1 SECOND. Anybody have any experience with it who can confirm this?? I might have to switch over.
I’m not into RPGs or any games for that matter, but I can 100% see the appeal of this. If followed through with, I could see this becoming a very popular project. Maybe use it to create a log of your journey developing the app/site.
Are you aware of free tokens from OpenAI… up to 11 million in some cases? Not everyone is eligible but might be worth a look?
Very useful information! Thanks for sharing.
Sounds like you’re the type of dev we want around here. Again, welcome. We’ve got a few gaming threads, but hopefully we’ll have even more in the future. It’s a great time to be a gamedev!
I have a lot of features I’m super excited to add for sure.
Organize recaps into folders/campaigns
Phonetic pronunciation dictionary
Auto language detection
Discord bot that handles recording
Mobile recording/companion app
Cheaper text-only mode
More Sci-fi music
Recap progress indicator/ estiamted time remaining for recap generation
Referral sign ups that grant free credits when user signs up using referall
Shop where you can unlock new music and narrators
Recap page sharing via link like Google Drive
Campaign wiki that auto updates with new information as you generate more recaps
RAG magic search bar that lets you ask questions about your campaign/sessions
That last one is actually pretty interesting. I have a little bit of experience with RAG (Retrieval-Augmented Generation) and I know it’s possible and would be super handy, but making sure it doesn’t cost big time $$$ is tough. Anybody else experimenting with RAG?
As soon as my site gets enough traction it’ll make sense to swap off of using APIs for a lot of the tech and to instead reserve a GPU instance through one of the cloud providers. APIs make sense now because it’s cheaper to use the more on-demand pricing than paying to reserving a GPU all day.
I’ve been having a lot of fun working on it. I recently graduated with a software engineering degree so I know my way around the tech, but I def need to hone my marketing skills. Everybody who I’ve introduced to my site says they absolutely love it, but finding people to show it to without coming off like a walking ad is tough!
We’re more interested in the dev side of things here, but like I said, it sounded like you have a cool project, and we’re trying to attract more devs. Gaming is my thing, and I just wish we had more of it here sometimes.
My marketing advice would be to keep the narrow focus (do one thing great instead of 33 things mediocre! lol)… try to do Freemium if you can.
There’s a Discord I’m on with a bunch of tabletop game devs, including the man who built this…
It’s not AI, but I’m building something similar with AI … (Again, beware feature-creep! lol)…
Sounds like you have a solid idea, though, and hopefully this will be a good place for you to ask questions, get answers, and share your dev journey with us.
I’ve been thinking about smaller (non-API) models for a while now for games… I think if you could somehow fit a model into an executable, you’d have a winner with a one-time price?
This is amazing! Love it. We were talking about the need for this last night at our game. I’m a dev who’s spent the past 2 years building a 5e platform by myself so I understand where you’re coming from. I’d love to connect. I haven’t opened my site (squire5) up yet but let me know if you’re interested in checking it out. Either way, I’m excited about trying your service at tonight’s game. Cheers!
