# Models, LoRAs and APIs — Tool Pantry

> Where to find and run open image models: Hugging Face, Civitai, Replicate, fal.ai and Tensor.art — weights, LoRAs and pay-per-use APIs.

Source: https://toolpantry.app/ai/model-hosting

## About

Two different jobs share this page. One is finding a model: weights, fine-tunes and LoRAs, usually with sample images so you can judge before downloading. The other is running one without owning a GPU, which is what the API providers sell.

## Tools

- **[Hugging Face](https://huggingface.co/)** (Open weights, Hub) — Where most open models and weights are published, with model cards that state the licence. Read the card before you ship anything.
- **[Civitai](https://civitai.com/)** (Community, LoRAs) — Community host for fine-tunes and LoRAs, with sample images and the prompts that produced them.
- **[Replicate](https://replicate.com/)** (API, Pay per use) — Runs open models behind a simple API. Useful for prototyping a pipeline before you commit to hardware.
- **[fal.ai](https://fal.ai/)** (API, Fast inference) — Inference endpoints tuned for latency, including video models. The usual choice when a workflow needs to feel interactive.
- **[Tensor.art](https://tensor.art/)** (Hosting, Community) — Model hosting and generation in one place, with a browsing experience closer to Civitai than to a cloud console.


## FAQ

**Is a LoRA the same as a model?**

No. A LoRA is a small adapter applied on top of a base model. It only works with the base it was trained against, which is the most common reason a downloaded LoRA does nothing.

**What licence do open models use?**

It varies per model and some restrict commercial use or require attribution. The model card states it; the brand name does not.

**API or local?**

Local for volume, privacy and iteration. API when you need a specific model your hardware cannot run, or you are still deciding.