diff --git a/CLAUDE.md b/CLAUDE.md
index 38e3d04..e9cd12d 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -25,7 +25,7 @@ Root `compose.yaml` uses Docker Compose's `include` directive to orchestrate mul
 - **kit**: Unified toolkit with Vert file converter and miniPaint image editor (path-routed)
 - **jelly**: Jellyfin media server with hardware transcoding
 - **drop**: PairDrop peer-to-peer file sharing
-- **ai**: AI infrastructure with Open WebUI, Crawl4AI, and pgvector (PostgreSQL)
+- **ai**: AI infrastructure with Open WebUI, ComfyUI proxy, Crawl4AI, and pgvector (PostgreSQL)
 - **asciinema**: Terminal recording and sharing platform (PostgreSQL)
 - **restic**: Backrest backup system with restic backend
 - **netdata**: Real-time infrastructure monitoring
@@ -457,6 +457,14 @@ AI infrastructure with Open WebUI, Crawl4AI, and dedicated PostgreSQL with pgvec
   - Designed for integration with Open WebUI and n8n workflows
   - Data persisted in `ai_crawl4ai_data` volume
 
+- **comfyui**: ComfyUI reverse proxy exposed at `comfy.ai.pivoine.art:80`
+  - Nginx-based proxy to ComfyUI running on RunPod GPU server
+  - Node-based UI for Flux.1 Schnell image generation workflows
+  - Proxies to RunPod via Tailscale VPN (100.121.199.88:8188)
+  - Protected by Authelia SSO authentication
+  - WebSocket support for real-time updates
+  - Stateless architecture (no data persistence on VPS)
+
 **Configuration**:
 - **Claude Integration**: Uses Anthropic API with OpenAI-compatible endpoint
 - **API Base URL**: `https://api.anthropic.com/v1`
@@ -492,12 +500,55 @@ Open WebUI function for generating images via Flux.1 Schnell on RunPod GPU:
 
 See `ai/FLUX_SETUP.md` for detailed setup instructions and troubleshooting.
 
+**ComfyUI Image Generation**:
+ComfyUI provides a professional node-based interface for creating Flux image generation workflows:
+
+**Architecture**:
+```
+User → Traefik (VPS) → Authelia SSO → ComfyUI Proxy (nginx) → Tailscale → ComfyUI (RunPod:8188) → Flux Model (GPU)
+```
+
+**Access**:
+1. Navigate to https://comfy.ai.pivoine.art
+2. Authenticate via Authelia SSO
+3. Create node-based workflows in ComfyUI interface
+4. Use Flux.1 Schnell model from HuggingFace cache at `/workspace/ComfyUI/models/huggingface_cache`
+
+**RunPod Setup** (via Ansible):
+ComfyUI is installed on RunPod using the Ansible playbook at `/home/valknar/Projects/runpod/playbook.yml`:
+- Clone ComfyUI from https://github.com/comfyanonymous/ComfyUI
+- Install dependencies from `models/comfyui/requirements.txt`
+- Create model directory structure (checkpoints, unet, vae, loras, clip, controlnet)
+- Symlink Flux model from HuggingFace cache
+- Start service via `models/comfyui/start.sh` on port 8188
+
+**To deploy ComfyUI on RunPod**:
+```bash
+# Run Ansible playbook with comfyui tag
+ssh -p 16186 root@213.173.110.150
+cd /workspace/ai
+ansible-playbook playbook.yml --tags comfyui --skip-tags always
+
+# Start ComfyUI service
+bash models/comfyui/start.sh &
+```
+
+**Proxy Configuration**:
+The VPS runs an nginx proxy (`ai/comfyui-nginx.conf`) that:
+- Listens on port 80 inside container
+- Forwards to RunPod via Tailscale (100.121.199.88:8188)
+- Supports WebSocket upgrades for real-time updates
+- Handles large file uploads (100M limit)
+- Uses extended timeouts for long-running generations (300s)
+
+**Note**: ComfyUI runs directly on RunPod GPU server, not in a container. All data is stored on RunPod's `/workspace` volume.
+
 **Integration Points**:
 - **n8n**: Workflow automation with AI tasks (scraping, RAG ingestion, webhooks)
 - **Mattermost**: Can send AI-generated notifications via webhooks
 - **Crawl4AI**: Internal API for advanced web scraping
 - **Claude API**: Primary LLM provider via Anthropic
-- **Flux via RunPod**: Image generation through orchestrator (GPU server)
+- **Flux via RunPod**: Image generation through orchestrator (GPU server) or ComfyUI
 
 **Future Enhancements**:
 - GPU server integration (IONOS A10 planned)