Create and configure AI stack

We will configure the Artificial Intelligence Docker stack and bring the stack up through Arcane. The stack consists of the following containers:

• Whisper: Speech-to-text through AI service.
• Piper: Text-to-speech through AI service.
• Ollama: LLM chat engine.
• Open WebUI: AI platform for local AI execution
• ComfyUI: Node-based generative AI workflow engine

Steps

1. Run: ./scripts/create_ai_folder.sh to generate the containers' directories on the SSD.
2. Edit the stack file: nano ./files/ai-stack.yml.
3. Replace TZ=America/New_York with your system time zone. You can use this list as a reference: https://en.wikipedia.org/wiki/List_of_tz_database_time_zones.
4. If you have an Nvidia GPU, skip this step. Delete or comment out the uncommented whisper and piper sections and then uncomment the commented whisper and piper sections. Under the ollama and comfyui sections, delete or comment out the entire runtime and deploy properties.
5. If you don't have an Nvidia GPU, under the comfyui section. change the property image to sinfallas/comfyui:0.9.2-intel if you have an Intel GPU or to sinfallas/comfyui:0.9.2-amd if you have AMD GPU.
6. If you want, you can change the whisper container's model (medium-int8) to a smaller or bigger model depending on your hardware. Available options are: tiny, base, small, medium, large & turbo.
7. Copy all contents of the file to the clipboard. Save and exit with Ctrl + X, Y, Enter.
8. Add stack in Arcane from the browser.
   1. Access Arcane through https://arcane.myhome.duckdns.org.
   2. Click "Projects" on the left menu and create a new project.
   3. Name it "ai" and paste the content of the ai-stack.yml that you copied to the clipboard and create the stack. From now on, modifications to the stack must be made through Arcane and not in the file.
9. Configure ComfyUI.
   1. Access ComfyUI through https://comfyui.myhome.duckdns.org.
   2. Click "Manager".
   3. Click "Model Manager".
   4. Search the model you would like to use for image generation and click "Install".
   5. Close the dialogs and in the Load Checkpoint node, change the model to the one you just installed.
   6. Click the file menu and select Export (API) and save the file on your PC. If you don't see this option, click Settings and under Comfy tab, enable Dev Mode and try again.
10. Configure Open WebUI.
    1. Access Open WebUI through https://openwebui.myhome.duckdns.org.
    2. Use the wizard to create a user account and password. It is again recommended to use Bitwarden for the same.
    3. Click on your user's icon and select Admin Panel.
    4. Click the Settings tab.
    5. If you want to allow more people to create accounts, enable Enable New Sign Ups. By default, an admin needs to approve it. If you don't want to need to approve, change Default User Role to user and click Save.
    6. Click the Connections section.
    7. Disable OpenAI API and enable Ollama API.
    8. Set http://ollama:11434 as the url.
    9. Click Save.
    10. Click the Models section.
    11. Click the Manage Models icon in the top right corner. Here you can download and delete models to your ollama instance. For Home Assistant, it is recommended to download qwen2.5:7b-instruct. For regular chatting, you can also use gemma3:12b, mistral-nemo:12b or qwen3:14b.
    12. Click the Web Search section.
    13. If you want to allow your model to search the internet, enable Web Search. You can use DDGS search engine with DuckDuckGo as the backend or whichever engine you prefer. Click Save.
    14. Click the Images section.
    15. Enable Image Generation.
    16. Set Model to the filename of the model you downloaded in ComfyUI including the extension.
    17. Set Image Generation Engine to ComfyUI.
    18. Set ComfyUI Base URL to http://comfyui:8188.
    19. Under ComfyUI Workflow click Upload and upload the file you saved from ComfyUI.
    20. Under ComfyUI Workflow Nodes you will need to map the properties to the node's ids. For a detailed guide see: https://docs.openwebui.com/features/image-generation-and-editing/comfyui/#identifying-node-ids-and-input-keys-in-comfyui.
    21. Click Save.