Skip to content

ComfyUI Studio

A web management platform for AI image and video generation built on ComfyUI. Runs as a Docker container on RunPod (or any NVIDIA GPU machine) with a dedicated interface for managing models, LoRAs, workflows, presets, custom nodes, an LLM assistant, and running generation jobs -- without touching the ComfyUI graph editor.

What You Get

  • Model Manager -- download, organize, and manage 126 AI models across 14 categories (checkpoints, diffusion models, LoRAs, VAE, text encoders, ControlNet, IP-Adapter, upscalers, and more)
  • Workflow Runner -- execute 12 predefined workflows through simple forms instead of building node graphs, including inpainting
  • Presets -- save and reuse workflow configurations, create presets from CivitAI images, run them from the web UI, Telegram, or API
  • Job Queue -- real-time progress tracking with WebSocket, live preview frames, ETA estimation
  • Job History -- browse past generations, retry failed jobs, export results
  • Asset Manager -- browse input/output files, upload references, download results
  • CivitAI Integration -- check model status, fetch generation data, detect dependencies, add models to catalog from CivitAI
  • Chrome Extension -- RunPod pod management and CivitAI browser integration from the browser toolbar
  • Telegram Bot -- run presets from Telegram by sending photos and answering questions
  • LLM Assistant -- run large language models locally on the GPU for prompt generation and chat
  • Gallery System -- download and browse CivitAI images for LoRA model galleries
  • Live Updates -- application code updates from Git without rebuilding the Docker image
  • 38 Custom Nodes -- pre-installed for image generation, video generation, face swap, upscaling, segmentation

Supported Features

Category What's Included
Video Generation WAN 2.1/2.2 (I2V, T2V, SVI Pro multi-scene), HunyuanVideo, CogVideoX, AnimateDiff, LTX Video
Image Generation Flux, SDXL, SD 1.5 with ControlNet, IP-Adapter, FaceID, inpainting
Processing Face swap (ReActor), upscaling (4x), segmentation (SAM2), frame interpolation (RIFE)
LLM Local llama-server with Qwen, Llama models via GGUF
Integrations CivitAI (model lookup, generation data, presets), Telegram bot, Chrome extension

Supported GPUs

Any NVIDIA GPU from V100 to B200. The Docker image is configurable via build arguments for CUDA version, attention optimizations (SageAttention, FlashAttention), and optional components (llama-server).

Generation GPUs Attention
Blackwell / Hopper B200, B100, H200, H100 SageAttention v2 (FP8), FlashAttention v3
Ampere / Ada Lovelace A100, A6000, L40S, RTX 4090/4080/4070/4060, RTX 3090/3080/3070/3060 SageAttention v1, FlashAttention v2
Turing RTX 2080/2070, T4 xformers only
Volta V100 xformers only