I made a modern and easy to use frontend for ComfyUI!

[searcc]

somni (GitHub Repository)

TL;DR: A modern frontend for ComfyUI. Gemini-style easy mode, IP-Adapter support, and built for both desktop and mobile.

---

✦ What is it

somni is a polished, opinionated frontend that runs alongside your existing ComfyUI install. It talks to ComfyUI over HTTP: your workflows, models, and outputs stay exactly where they are.

• Easy mode: a chat-style interface (think Gemini / ChatGPT) for one-prompt-and-go generation
• Pro mode: full sidebar with sampler, scheduler, seed, LoRAs, CFG, advanced options
• Reference image (IP-Adapter): General · Face · FaceID modes with a denoising slider
• Batch generation: generate N images, displayed in a scrollable preview
• Gallery with full-screen viewer, swipe-to-navigate on mobile, arrow buttons on desktop
• Favorites: star any option and its value persists across reloads
• Mobile-first design: phone-friendly bottom bar, swipe gestures, tap targets sized properly
• Smooth animations everywhere: toggles spring, popovers pop, gallery items stagger in
• No background services: runs as a single Python script when you want it, closes when you don't

---

✦ Using somni from your phone

The launch script binds to 0.0.0.0, so any device on your Wi-Fi can reach it.

1. Find your PC's local IP (ipconfig → look for IPv4 Address, usually 192.168.x.x)
2. On your phone, open http://<that-ip>:8080
3. Generate images from the couch

---

✦ Reference image (IP-Adapter)

Three modes, three workflows. Each needs specific model files in your ComfyUI install. somni's UI tells you which one is active, but the models are on you to download:

Mode	Needs
General	ip-adapter-plus_sdxl_vit-h.safetensors in ComfyUI/models/ipadapter/
Face	ip-adapter-plus-face_sdxl_vit-h.safetensors in ComfyUI/models/ipadapter/
FaceID	ip-adapter-faceid-plusv2_sdxl.bin in ipadapter/, matching LoRA in loras/, plus pip install insightface onnxruntime

All three modes also need:

• CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors in ComfyUI/models/clip_vision/
• The ComfyUI_IPAdapter_plus custom node (install via ComfyUI Manager)

Easiest path: open ComfyUI Manager → Install Models, search for "ipadapter". Pick what you want.

---

✦ How it works

server.py is a tiny Python proxy (~200 lines, stdlib only). It serves index.html and forwards everything else to ComfyUI, stripping Origin/Referer headers so ComfyUI's loopback host-check passes. It also adds two endpoints: /__list for gallery thumbnails and /__delete for delete buttons because vanilla ComfyUI doesn't expose them.

The entire UI is one HTML file. No build step. No npm. No bundler. Open the source and you can change anything.

---

✦ Roadmap

• Linux & macOS launch scripts (.sh)
• Multi-image reference (IP-Adapter combine mode)
• Workflow presets (save/load custom configurations)
• Inpainting

---

✦ License

MIT. Do whatever you want, just don't blame me.

---

Check it out!

[fernicar]

If you haven't seen it, here is a list with more:
https://github.com/light-and-ray/awesome-alternative-uis-for-comfyui

[gershu-ar]

somni (GitHub Repository)

TL;DR: A modern frontend for ComfyUI. Gemini-style easy mode, IP-Adapter support, and built for both desktop and mobile.

Nice work there, will try it out!