Nymbo/Model_Repo_Template
Text-to-Image • Updated • 3
Nymbo PRO
Nymbo
AI & ML interests
I like unrestricted, free, utilitarian stuff. I tend to archive good spaces because I don't trust y'all to keep them good :) Most spaces with runtime errors just need a restart. Paused spaces should work but require a GPU, duplicate to use.
Recent Activity
posted an update about 9 hours ago
I rewrote a bunch of my MCP tools as three agent skills: `web-search`, `speech-kokoro`, and `hf-inference`. https://huggingface.co/datasets/Nymbo/Skills
# web-search
`web-search` is the real star here. It's basically my `Web_Search` + `Web_Fetch` MCP tools turned into a portable skill and requires no authentication.
For normal web search it can use DuckDuckGo, Bing, Brave, Yahoo, or Wikipedia, and in `auto` mode it walks through multiple backends with fallback + deduping. News search does the same thing across DuckDuckGo, Bing, and Yahoo, with a smart fallback to general web search if news results come up empty.
It also supports images, videos, and book search, max result limits, day/week/month/year filters, backend selection, page/offset pagination, and output as plain text, Markdown, or JSON.
Search finds the right page, then fetch can pull that URL back as clean Markdown, raw HTML, or extracted links, with `--max-chars` and `--offset` so long pages can be read in chunks instead of dumping the whole internet into context.
# speech-kokoro
`speech-kokoro` is local text-to-speech using Kokoro-82M.
It can generate from direct text or a text file, list available Kokoro voices, use a different voice id, set voice/speed/output path, and handle longer text by segmenting and stitching the audio back together. It uses CUDA if PyTorch sees CUDA, otherwise it falls back to CPU. No OpenAI Audio API, no hosted TTS service, just local speech generation.
# hf-inference
`hf-inference` wraps Hugging Face serverless inference for image and video generation.
For images, the default model is `Tongyi-MAI/Z-Image-Turbo`. For videos, the default is `Wan-AI/Wan2.2-T2V-A14B`. You can swap in any compatible model id, set negative prompts, width/height, steps, CFG scale, seeds, provider order, and pass extra JSON for weird model/provider-specific options. It checks `HF_READ_TOKEN` / `HF_TOKEN` from the environment.
---
Download them here:
https://github.com/Nymbo/Public-Skills updated a dataset about 9 hours ago
Nymbo/Skills published a dataset about 9 hours ago
Nymbo/SkillsOrganizations
