🎨 Image Generation (Creations — Optional)

With On My Chat, you can turn your node into a private generative art studio right inside the Creations workspace (Chats → Creations).

Optional Component: Image generation is completely optional — On My Chat runs chat, semantic search, Memory Cards, and MCP document tools without it. Whenever ComfyUI is configured on your GPU node (or reached remotely via Agentic P2P mode / the On My Chat Public Node), the Creations workspace activates automatically.

Creations — CyberCraft and Creative Studio

1. Two Engines Scaled to Your GPU (CyberCraft & Creative Studio)

When ComfyUI is connected, Creations provides two dedicated generation modes at the top of the studio so you can match model weight to available VRAM:

  • ✨ CyberCraft (SDXL Mode — Modest VRAM GPUs):
    • Built for fast, small-footprint generation on everyday GPUs with low-to-medium VRAM (e.g. 8–12 GB).
    • Includes one-click Painting Styles (Realistic, Fantasy, Perfect, Tooned), consistent Characters (With characters + character/style LoRA selectors), and Continue mode for iterative scene refinement.
  • âš¡ Creative Studio (Krea 2 Mode — High-VRAM GPUs):
    • Built for high-memory GPUs (16–24+ GB VRAM workstations or shared P2P GPU nodes).
    • Runs flagship high-parameter diffusion pipelines powered by CyberRealistic Krea 2 (alongside curated studio checkpoints), delivering cinematic lighting, lifelike skin/fabric detail, and high-resolution renders saved straight to your private storage.

2. Two Ways to Connect ComfyUI

Option A: Agentic P2P Mode (Zero Local GPU)

If your node runs in Agentic mode (AI_PROVIDER = agentic, e.g., installed on a Windows laptop with -WithAi -Agentic), no local ComfyUI installation is needed:

  • agentic_tunnel.py exposes a local listener (127.0.0.1:28188) that tunnels ComfyUI HTTP API calls (/prompt, /view, /object_info) and real-time WebSocket progress (/ws) over an encrypted WebRTC DataChannel to your remote GPU node or the Public AI Node (port 8188).

Option B: Local ComfyUI on Your Own GPU Host

To run CyberCraft and/or Creative Studio (Krea 2) locally on your own NVIDIA GPU machine:

  1. Install ComfyUI: 
    git clone https://github.com/comfyanonymous/ComfyUI /opt/ComfyUI
cd /opt/ComfyUI
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
  2. Place Models & LoRAs: Drop your SDXL (CyberCraft) and/or Krea 2 (Creative Studio) checkpoints and character .safetensors LoRA weights into /opt/ComfyUI/models/checkpoints/ and /opt/ComfyUI/models/loras/.
  3. Start ComfyUI on port 8188: 
    python3 main.py --listen 127.0.0.1 --port 8188
  4. Configure /opt/onmychat/config.ini & Restart: 
    COMFY_API_URL = http://localhost:8188
WORKFLOW_PATH = flow.json
COMFY_OUTPUT_DIR = /opt/ComfyUI/output
COMFY_INPUT_DIR = /opt/ComfyUI/input

    sudo systemctl restart onmychat

3. Generating Images in Creations & Chat

  1. Creations Workspace: Open Creations from the left sidebar in On My Chat, switch between ✨ CyberCraft and ⚡ Creative Studio, pick your style or character presets, and type your prompt in Describe an image….
  2. In-Chat /generate Command: You can also trigger image generation directly inside any conversation: 
    /generate Autumn park portrait in golden hour light, camel trench coat, 85mm lens

All generated artworks are saved directly into your configured On My Disk storage folder (Settings → AI Assistant) and can be browsed, downloaded, or shared via P2P links from any device.