Choose Your AI Provider
Loki Studio supports multiple AI providers for generating titles, descriptions, tags, and thumbnail text. Choose the option that best fits your needs:
🧠
Grimnir (Recommended)
Pro Feature
Built-in LLM that runs entirely on your machine. No internet required after download. Best quality/privacy balance.
- One-time download, use forever
- No API costs
- 100% offline capable
- Multiple model choices
🦙
Ollama / LM Studio
Free for All Users
Use external local LLM servers. Great if you already run Ollama or prefer managing models separately.
- Free to use
- Choose any model
- Local processing
- Requires separate install
☁️
OpenAI / Claude
Pro Feature
Cloud AI for highest quality. Use GPT-4 or Claude with your own API keys. Best for premium content.
- Highest quality output
- Always up-to-date models
- Requires API key
- Pay-per-use costs
Grimnir Setup (Built-in LLM)
Grimnir is Loki Studio's integrated AI engine, powered by the Skuld module. It runs state-of-the-art language models directly on your GPU.
Step 1: Select Grimnir Mode
- Go to Application Settings (Yggdrasil tab)
- Click the LLM Providers sub-tab
- In the LLM Mode section, select "Grimnir (Built-in LLM)"
Step 2: Download a Model
Grimnir supports multiple models optimized for different use cases:
| Model |
Speed |
Quality |
Best For |
| qwen3:8b |
Fast |
6.8/10 |
BEST overall - All languages |
| nemotron:30b |
Medium |
6.1/10 |
Excellent descriptions (20GB+ VRAM) |
| gemma2:9b |
Fast |
6.1/10 |
Consistent multilingual |
| gpt-oss:20b |
Slow |
5.5/10 |
Good vocabulary variety |
| qwen2.5-coder:7b |
Fast |
5.5/10 |
Tech/coding content only |
| deepseek-r1:7b |
Slow |
4.0/10 |
Not recommended |
| llama3.2:3b |
Very Fast |
3.5/10 |
Not recommended |
- Click a model button to start downloading
- Wait for the download to complete (size varies by model)
- The status will change to "Ready: [model-name]" in green
Step 3: Configure Max Tokens
The Max Tokens setting controls how much "thinking time" the AI gets. For reasoning models:
- Lower values (4K-8K): Faster responses, shorter content
- Higher values (16K-32K): More detailed, better quality
- Maximum (200K): Full reasoning capability for complex content
Recommendation: Start with qwen3:8b at 16K tokens. Adjust based on your results and generation speed.
Ollama Setup
Ollama is a free, open-source tool for running LLMs locally. It's available to all Loki Studio users.
Step 1: Install Ollama
- Download Ollama from ollama.ai
- Run the installer
- Ollama runs as a background service on port 11434
Step 2: Download a Model
Open a terminal/command prompt and run:
ollama pull qwen3:8b
Or install directly from Loki Studio using the quick-install buttons.
Step 3: Configure in Loki Studio
- Go to Application Settings > LLM Providers
- Select "Bring Your Own (Third-Party)" mode
- Choose Ollama (Local) from the provider dropdown
- Base URL should be
http://localhost:11434
- Select your model from the dropdown (click Refresh if needed)
LM Studio Setup
LM Studio provides a graphical interface for managing local LLMs with an OpenAI-compatible API.
Step 1: Install LM Studio
- Download from lmstudio.ai
- Install and launch the application
- Download your preferred model through the LM Studio interface
Step 2: Start the Local Server
- In LM Studio, go to the Local Server tab
- Select your model
- Click Start Server
- Note the URL (usually
http://localhost:1234)
Step 3: Configure in Loki Studio
- Go to Application Settings > LLM Providers
- Select "Bring Your Own (Third-Party)" mode
- Choose LM Studio (Local) from the provider dropdown
- Enter the LM Studio server URL
OpenAI / Claude Setup
For the highest quality AI output, you can use cloud-based models from OpenAI or Anthropic. This is a Pro feature.
OpenAI (GPT-4) Setup
- Create an account at platform.openai.com
- Navigate to API Keys in your account settings
- Create a new API key and copy it
- In Loki Studio, go to Application Settings > LLM Providers
- Select "Bring Your Own (Third-Party)" mode
- Choose OpenAI (GPT-4, GPT-3.5)
- Paste your API key
- Select your preferred model (GPT-4 recommended for quality)
Claude (Anthropic) Setup
- Create an account at console.anthropic.com
- Navigate to API Keys
- Create a new API key and copy it
- In Loki Studio, go to Application Settings > LLM Providers
- Select "Bring Your Own (Third-Party)" mode
- Choose Claude (Anthropic)
- Paste your API key
Cost Note: Cloud AI providers charge per token. Monitor your usage to avoid unexpected bills. A typical video's metadata generation costs $0.01-0.05 with GPT-4.
Language Recommendations
Different models perform better with different languages:
- All Languages: qwen3:8b (recommended for all content)
- English Only: Any model works well
- Japanese/Korean/Chinese: qwen3:8b or gemma2:9b
- Russian/Hindi: qwen3:8b
- Tech/Coding Content: qwen2.5-coder:7b (English only)
- High VRAM (20GB+): nemotron:30b for premium quality
Troubleshooting
Model download fails or is stuck
Check your internet connection and available disk space. Models can be 4-20GB. Try clicking the download button again to resume.
Ollama not connecting
Ensure Ollama is running (check system tray on Windows). Verify the URL is correct: http://localhost:11434. Try restarting Ollama.
OpenAI/Claude API errors
Verify your API key is correct and has available credits. Check that you've selected a valid model. Ensure your account isn't rate-limited.
Generation is very slow
Try a smaller model or reduce Max Tokens. Ensure your GPU drivers are up to date. For Grimnir, verify CUDA is being used (check device setting).
NordIQ / Loki Studio