Nymbo's Pastebin

New: Live Subtitles

Live Subtitles captions anything Windows is playing (a stream, a film, a call, a game) in a draggable overlay, with speech recognition running locally on your GPU or CPU. It also writes SRT and ASS subtitle files for videos.

View on GitHub

Hey, I'm Nymbo. This is the ground zero of my digital footprint, and my Next.js playground. I'm a web developer, indie game dev, AI researcher and I play vidya games and stuff. I run a few commercial platforms that aren't linked here, and I contribute a lot to Hugging Face. Feel free to reach out ✨

Network & Socials

The best way to contact me is via Twitter or Discord.

Discord:

Battle.net:

Nymbo Tools

An all-in-one MCP server: thirteen general-purpose tools any agent can call, from one URL.

Agent Terminal
Agent Skills
Web Fetch
Web Search
Code Interpreter
Shell Command
File System
Obsidian Vault
Memory Manager
Generate Speech
Generate Image
Generate Video
Deep Research

MCPhttps://nymbo-tools.hf.space/gradio_api/mcp/

Streamable HTTP. Paste it into any MCP client, Nymbot included.

Try it live Test with Nymbot59RunningHugging Face

On Hugging Face

Popular Spaces

The Spaces people keep coming back to, ranked by likes. Open one to use it right here.

385CPU

Serverless ImgGen Hub

FLUX, SD 3.5 and LoRAs in one hackable hub, on serverless inference. No GPU needed.

Running· Text to image· LoRAs

313ZeroGPU

Virtual Try-On

Put any garment on any photo of a person, with a high-fidelity diffusion model.

Running· Image editingbuilt on IDM-VTON

284ZeroGPU

Voice Clone Multilingual

Clone a voice from a short sample and have it speak sixteen languages.

Running· Voice cloning· 16 languagesbuilt on XTTS

66Static

Repo to Text

Turn a GitHub repository into one text file any LLM can read.

Running· LLM context

44CPU

Serverless TextGen Hub

Chat with serverless text models across inference providers, multimodal, and as an MCP server.

Running· Chat· MCP

Speech

Tiny speech models

Text to speech, voice cloning, and transcription from models of 4 to 600 million parameters. Every one runs on a free CPU Space, no GPU.

8MTTS
Paradee-TTS
Kokoro distilled to a tenth of its size, with one voice.
Running· English2
4M · 9MTTS
Inflect-v2 TTS
Two speech models of 4M and 9M parameters, side by side.
Running· English6
16.9 MBASR
Whistle-ASR
Speech to text in seven languages from a 16.9 MB model.
Running· 7 languages
44MTTS
Supra-TTS
A Glow-TTS voice with a HiFi-GAN vocoder.
Wakes on visit· English
66MTTS
Supertonic TTS
Fast, natural speech from a 66M model built for on-device use.
Running· English8
82MTTS
Kokoro TTS
The small-model favourite, with 54 voices.
Running· English
100MTTSVoice cloning
Pocket-TTS
Clone a voice and stream speech back, about 6x faster than real time on a CPU.
Running· English5
120MTTSVoice cloning
Sopro Voice Clone
Zero-shot voice cloning from a few seconds of audio, streamed.
Running· 4 languages2
600MASR
Parakeet-ASR
NVIDIA's Parakeet transcribing 25 languages, on a CPU.
Running· 25 languages1

Off the Hub

Plugins and apps

Two Obsidian plugins from the community directory, and two Windows apps: one captions whatever is playing, the other types what you say.

Obsidian plugin2,013 downloads

Markdown Minimap

A whole-note minimap for Obsidian that stays accurate on notes of any length, with code in colour, folds that follow yours, and click-and-drag to jump.

GitHub In Obsidian

Windows app4 stars

Live Subtitles

Live captions for anything Windows is playing, in a draggable overlay, from local speech models: Parakeet on the GPU or Nemotron on the CPU. It also writes SRT and ASS files for videos.

GitHub

Windows app

Cursor Dictation

Press a hotkey, talk, and what you said is typed wherever your cursor is. Parakeet runs locally on the GPU or CPU and punctuates for you, and your clipboard is put back afterwards.

GitHub

Obsidian plugin1,471 downloads

Copy Highlighter

Highlights text the moment you copy it in Obsidian, so you can see the copy worked. Styleable, and it never touches the network.