Run the AI fully offline

Use the editor's AI ad generation with a model that runs entirely on your own computer - private, free, and it even works with no internet at all.

JUMI Ad Animator's AI features are bring-your-own: the editor talks directly to whichever AI you configure, and nothing passes through our servers. Most people paste an API key from a cloud provider - but you don't have to. With LM Studio or Ollama, the "AI" is a model running on your own machine. Your briefs, your copy, your ideas: none of it leaves your computer, which makes this the most private way to use AI anywhere on the web.

What you need

A reasonably recent computer - Apple Silicon Macs and Windows machines with 16 GB of RAM handle small models comfortably.
LM Studio (the easiest option, with a friendly app) or Ollama (a lightweight command-line tool). Both are free.
A small instruction-tuned model. Ad copy is a light task: models in the 3-8 billion parameter range (for example Llama 3.2 3B or an 8B model) write perfectly good headlines.

Option 1 - LM Studio (recommended)

Download LM Studio from lmstudio.ai and install it like any app.
In the app, search for and download a small model (try a Llama 3.2 or Qwen instruct model).
Open the Developer tab and click Start server - make sure the CORS option is enabled (or run lms server start --cors from a terminal). LM Studio now listens at http://localhost:1234.
In the Ad Animator editor, click AI → Provider settings, choose LM Studio (local - fully private) and hit Save & test. No API key needed.

Option 2 - Ollama

Install Ollama from ollama.com, then pull a model, for example: ollama pull llama3.2.
Start it so the browser is allowed to talk to it: OLLAMA_ORIGINS='*' ollama serve (on Windows, set the OLLAMA_ORIGINS environment variable to * before starting Ollama).
In the editor, click AI → Provider settings, choose Ollama (local - fully private), set the model name you pulled, and hit Save & test.

Then just generate

Click AI in the editor's top bar, describe what you're advertising, pick a tone, choose how many frames and the total run time (networks differ - many cap the loop at 6 or 30 seconds, some cap frame count instead), and hit Generate ad set. With an image provider configured you can also have it paint the background - one image for the whole ad, or one per frame. The model writes the copy and picks one of the editor's hand-tuned colour palettes; the editor itself builds the layout at every size in your project - the same engine that powers the templates - so the result is a real, editable, animated ad set, not an AI collage. Headlines stay in sync across sizes, and your brand kit's colours and font apply automatically.

Cloud keys work too

If you'd rather not run a local model, the same settings panel accepts your own API key for Google Gemini (free tier available), OpenAI, Anthropic, OpenRouter, Groq or Mistral. Your key is stored only in your browser - never in project files, never on our servers - and your briefs go straight from your browser to the provider you chose. The privacy trade-off is yours to make, which is rather the point. One caveat: background image generation is gated by some providers behind paid or verified accounts - if images fail while copy works, that's usually why; OpenAI Images is the most reliable choice.

Why we built it this way: the editor works entirely on your device, and we weren't going to break that promise for AI. No accounts, no server in the middle, no usage meter - bring your own model, local or cloud, and it's simply a feature of a free tool. Read more in our privacy notice.