comfyui-generate

name	comfyui-generate
description	Generate images via ComfyUI API with support for Flux and LoRAs, plus image-to-video via Wan 2.2. Use when creating images, generating artwork, running ComfyUI workflows, or producing visual content.

ComfyUI Generate Skill

Generate images using ComfyUI API with Flux text-to-image and Wan 2.2 image-to-video.

For detailed workflow node maps, see references/workflows.md.

Usage

Generate an image (custom workflow)

python3 {baseDir}/generate_image.py generate \
    --prompt "a gothic fantasy queen in dark armor" \
    --base flux \
    [--lora creature_lora:0.8] \
    [--size 832x1216] \
    [--output /path/to/output.png]

Generate using a named workflow

python3 {baseDir}/generate_image.py generate \
    --workflow flux-lora \
    --prompt "a woman in a dark fantasy setting, dramatic lighting"

Named workflows are pre-built ComfyUI pipelines with model, LoRA, sampler, and all settings already configured. Only the positive prompt text is changed. The --base, --lora, and --size options are ignored when using a named workflow.

Image-to-video

python3 {baseDir}/generate_image.py generate \
    --workflow wan22-i2v \
    --prompt "She turns slowly, silk shifting with the movement" \
    --image /path/to/source.png \
    --output video.webm

Override workflow parameters with --set

python3 {baseDir}/generate_image.py generate \
    --workflow flux-lora \
    --prompt "crystal queen" \
    --set '96.inputs.lora_name=my-models/my-lora.safetensors' \
    --set '96.inputs.strength_model=0.8'

List available named workflows

python3 {baseDir}/generate_image.py generate --workflow help

Check queue / history

python3 {baseDir}/generate_image.py queue
python3 {baseDir}/generate_image.py history <prompt-id>

Named Workflows

Name	Model	LoRA Slots	Purpose
flux-lora	Flux-dev fp8	2 (nodes 96, 200)	Primary text-to-image, natural language prompt
wan22-i2v	Wan 2.2 I2V (MoE)	None	Image-to-video with auto-captioning and 2x upscale

For detailed node maps, parameter injection points, and data flow diagrams, see references/workflows.md.

Options

LoRA Support

--lora creature_lora:0.8
--lora style_lora:0.6

Multiple LoRAs can be chained. For named workflows, use --set to inject LoRA paths.

Sizes

• 832x1216 — Portrait 2:3 (default)
• 1024x1024 — Square
• 1216x832 — Landscape 3:2
• 896x1152 — Portrait 3:4
• 1152x896 — Landscape 4:3

LoRA Path Conventions

models/loras/
├── released/          # Published LoRAs
│   └── flux_1_d/      # Flux-trained LoRAs
├── testing/           # In-development LoRAs
│   └── flux_1_d/
└── third_party/       # Third-party style LoRAs

Environment Variables

Required:

• COMFYUI_HOST: ComfyUI server hostname (default: localhost)
• COMFYUI_PORT: ComfyUI server port (default: 8188)

Optional:

• COMFYUI_OUTPUT_DIR: Default output directory

Content Safety — Age & Maturity Requirements

All human characters must be unambiguously adult (18+). This is non-negotiable.

Before posting or scheduling ANY generated image containing a human figure:

1. Scan for chibi/young indicators — Reject images where:

   • Head-to-body ratio is exaggerated (head ≥ 1/3 of total height = chibi proportions)
   • Face has round, childlike features (button nose, puffy cheeks, no jawline definition)
   • Body proportions are childlike (short limbs, no waist definition, small frame)

2. Prompt for maturity — When generating characters, always include age-establishing language:

   • ✅ adult woman, mature face, defined cheekbones, tall figure, long legs
   • ❌ Do NOT use: cute girl, small, chibi, tiny, young, petite face

3. Stylization is not an excuse — Cartoon, anime, and 3D stylized renders can still read as clearly adult.

4. When in doubt, regenerate — If a generated image could plausibly be interpreted as depicting a minor, delete it and regenerate with stronger maturity prompts.

Examples

# Simple Flux generation
python3 {baseDir}/generate_image.py generate --prompt "a majestic dragon" --base flux

# Flux with LoRA (custom workflow)
python3 {baseDir}/generate_image.py generate \
    --prompt "fantasy creature, detailed" \
    --base flux \
    --lora creature_v2:0.7

# Named workflow: Flux with dual LoRA
python3 {baseDir}/generate_image.py generate \
    --workflow flux-lora \
    --prompt "a woman in crystalline armor, dramatic lighting"

# Image-to-video
python3 {baseDir}/generate_image.py generate \
    --workflow wan22-i2v \
    --prompt "gentle camera movement, atmospheric" \
    --image reference.png \
    --output video.webm

# Check queue
python3 {baseDir}/generate_image.py queue

Troubleshooting

ComfyUI connection refused

• Verify ComfyUI server is running: curl http://$COMFYUI_HOST:$COMFYUI_PORT/system_stats
• Check COMFYUI_HOST and COMFYUI_PORT environment variables

LoRA not found

• LoRA paths are relative to ComfyUI's models/loras/ directory
• Check that the file exists on the ComfyUI server

Output not downloaded / wrong file

• Workflows use _meta.output: true on the target SaveImage node
• PreviewImage nodes are skipped — only nodes tagged with output: true are downloaded

Queue stuck or timeout

• Check python3 {baseDir}/generate_image.py queue for stuck items
• Video workflows (wan22-i2v) take 5-15 minutes — increase timeout

Output

Generated images are saved to:

• Specified --output path, or
• $COMFYUI_OUTPUT_DIR/YYYYMMDD-HHMMSS.png, or
• Current directory with timestamp filename