LibraryLibrary

Anthropic's assistant. Strong at long-form writing, analysis and agentic coding; Claude Code brings the same models into the terminal.

OpenAI's assistant, with image generation, browsing and code execution bundled into one interface.

Google's multimodal assistant, tightly wired into Workspace and with very large context windows.

Moonshot's assistant, built around very long context. Useful when a model has to hold a whole book or codebase at once.

Open-weight reasoning models at a fraction of the usual API price. The cheapest way to put a capable model behind your own product.

Answers only from the documents you give it, and cites the page it took each claim from. The one assistant that will not invent a source.

The most opinionated image model on the market. Unmatched for art direction and consistent character work via style and character references.

Black Forest Labs' image family. Open-weight variants make it the default when you need to fine-tune or run a model locally.

Adobe's generative suite, trained on licensed material and embedded directly in Photoshop and Express — the safest option for commercial work.

A real-time canvas that generates while you draw. The fastest way to explore a visual direction before committing to a render.

Upscaling that invents plausible detail rather than interpolating it — skin, fabric, foliage. Powerful and opinionated, so it will happily change what it enlarges.

Google Labs pointed at marketing production: campaign assets generated to sit inside a brand you already have. An experiment, with the lifespan that implies.

Generative video with a filmmaker's toolkit around it: camera control, inpainting, and frame-accurate editing.

OpenAI's text-to-video model, strong on physical coherence and long shots.

Fast video generation plus photogrammetry and 3D capture from the same company.

The lab behind Hailuo. Its video model reads motion and physical weight better than most, and the text and voice models sit in the same account.

A filmmaking front end for Veo: shot lists, camera moves and continuity between scenes, instead of one prompt at a time.

ByteDance's enterprise arm, and the way to reach its Seedance video and Seedream image models from outside China. Built for teams wiring models into a product.

Voice synthesis, cloning and dubbing. The reference point for narration and character voice.

Full songs from a prompt — vocals, arrangement and mix. Useful for scratch tracks and mood pieces.

Full tracks from a prompt, in the same territory as Suno. Worth keeping both, for the days one model's read on a genre is simply wrong.

A node graph for image and video pipelines. Steep, local, and the only sane way to make a generative process repeatable.

The public commons for open models, datasets and demos. Start here when you want to know what actually exists.

An editor built around model-assisted editing — useful for designers who write enough code to break things.

Where you prototype against Gemini and leave with an API key. The shortest path from an idea to a call you can ship.

Describe an app and get a working full-stack one, then keep editing it in prose. Good for a prototype that runs, less so for a codebase you have to inherit.

Upscayl uses AI models to enhance your images by guessing what the details could be. It uses Real-ESRGAN and Vulkan architecture to achieve this.

One command to pull a model, one endpoint to talk to it. The plainest way to run models on your own machine.

A desktop app for local models with the browser and the tuning knobs visible. Ollama with a face on it.

Local models on Apple Silicon with chat, live performance readouts and editor hooks. MIT-licensed, for people who would rather nothing left the laptop.

Fine-tuning that fits in far less VRAM and finishes far sooner. The reason one consumer GPU is now enough to train on.

LM Studio's job, narrowed to Apple Silicon. The one to try if the Mac is the machine you actually run models on.

Run thousands of open models behind one API, without touching a GPU yourself.

Generative media models behind a fast API, tuned for image and video latency rather than for breadth of catalogue.

Serverless GPUs for Python. Write a function, decorate it, and it runs on hardware you never provisioned.

GPUs by the hour or by the request, with templates for the usual model stacks. The middle ground between a laptop and a datacentre.

A marketplace for other people's idle GPUs. The cheapest compute you will find, with the reliability that implies.