comfyui

Generate images by submitting ComfyUI workflows through a localhost llm-compose proxy.

1|Updated Mar 18, 2024

Search Tags:#image generation #queue #text-to-image #sdxl #comfyui #prompt history #gpu mode swap

Ready to run

npx skills add https://github.com/erfianugrah/dotfiles --skill comfyui-erfianugrah

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: comfyui
Source: https://github.com/erfianugrah/dotfiles/tree/main/.pi/agent/skills/comfyui
Command: npx skills add https://github.com/erfianugrah/dotfiles --skill comfyui-erfianugrah

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It removes the friction of generating consistent images by routing ComfyUI requests through a local proxy and handling GPU mode switching automatically.

Core Features & Use Cases

Text-to-image generation via ComfyUI workflows: Use SDXL/Illustrious/Flux style workflows with configurable seed, steps, CFG, and sampler to produce images.
Queue inspection and generation history retrieval: Check /queue for pending jobs and fetch /history/{prompt_id} to find outputs and metadata.
Operational GPU mode swap: Seamlessly stops llama-server and starts ComfyUI when needed, with expected warm-up latency.

Quick Start

Ask the system to generate an SDXL image by sending your prompt through the local llm-compose proxy endpoint for ComfyUI (http://localhost:11434) and then retrieve the resulting PNG from the configured output directory.

Frequently Asked Questions about comfyui

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I generate images via ComfyUI workflows through a local proxy?▼

To generate images via ComfyUI workflows, you submit your prompt through a localhost llm-compose proxy endpoint and retrieve the resulting PNG outputs from the configured output directory using the prompt_id.

How does GPU mode swapping work between llama-server and ComfyUI?▼

GPU mode swapping works by automatically stopping the llama-server process and starting ComfyUI when image generation is requested, accommodating expected warm-up latency before processing the queue.

Can I use SDXL and Flux workflows for text-to-image generation with this queue management approach?▼

Yes, you can use SDXL, Illustrious, and Flux style workflows for text-to-image generation, applying configurable parameters like seed, steps, CFG, and sampler while managing asynchronous generation queues.

How do I check pending jobs and retrieve generation history in ComfyUI?▼

You check pending jobs by inspecting the /queue endpoint and retrieve generation history by fetching /history/{prompt_id} to discover recent outputs and metadata for your submitted workflows.

What configuration is needed to route ComfyUI requests through a local proxy?▼

You must configure the COMFYUI_PROXY_URL and COMFYUI_OUTPUT_DIR environment variables to properly route workflow requests to the proxy and locate the generated image files.

Why does text-to-image generation fail when the GPU is occupied by another process?▼

Text-to-image generation fails because the GPU is occupied by the llama-server, requiring a GPU mode swap to stop the server and start ComfyUI before submitting workflows to the queue.