comfy-cli

Run ComfyUI workflows from the command line. Designed for agent automation — structured JSON output on stdout, human-readable text on stderr.

Quick Start

# Run a workflow (blocks until complete)
python3 run.py image "a cat sitting on a windowsill" -o ./output

# Get view URL without downloading
python3 run.py image "a cat" --no-download

Workflows

Workflow	Use Case	Default Model
image	General purpose, style-agnostic	z_image_turbo_bf16.safetensors
anime	Anime/manga style	NetaYumev35_pretrained_all_in_one.safetensors
sdxl	SDXL-specific workflows	See workflow JSON

Model Paths

Models must include the correct path prefix for ComfyUI's discovery:

• SDXL models: SDXL/ (e.g., SDXL/realDream_sdxlPony17_fp16.safetensors)
• Standard checkpoints: Just the filename (e.g., z_image_turbo_bf16.safetensors)
• UNet models: unet/ prefix

Output Convention

All generated images go to:

~/.openclaw/media/tool-image-generation/

Machine-Readable Output

The script outputs JSON lines to stdout at each stage:

status	meaning
submitting	Job submitted, waiting on queue
queued	In ComfyUI queue (includes prompt_id, queue_position)
done	Complete — includes paths or urls
failed	Error — includes error field with reason

Example parsing:

output=$(python3 run.py image "a cat" -o ./output 2>/dev/null)
last=$(echo "$output" | tail -1)
status=$(echo "$last" | jq -r '.status')
if [ "$status" = "done" ]; then
    echo "Files: $(echo "$last" | jq -r '.paths[]')"
fi

Recommended Models

• z_image_turbo_bf16.safetensors — Fast, versatile, good for most use cases
• NetaYumev35_pretrained_all_in_one.safetensors — Best anime/manga output
• realDream_sdxlPony17_fp16.safetensors — Photo-realistic SDXL

Requirements

• ComfyUI running locally (default: http://localhost:8188)
• Python 3.10+
• Models installed in your ComfyUI models directory