image-generation

name	Image Generation
description	Generate AI images via ComfyUI on junkpile using the tsr CLI. Supports text-to-image with model/LoRA selection, and displays results on the MARAUDER visor viewport. <example> Context: User wants to generate an image user: "generate a mech in a city" </example> <example> Context: User wants a specific style user: "generate a portrait photo, use dreamshaper" </example> <example> Context: User wants multiple images user: "generate 4 images of cyberpunk scenes" </example> <example> Context: User wants to see available models user: "what models do we have?" </example> <example> Context: User wants to show an image on the HUD user: "show that on the hud" </example> <example> Context: User wants a grid of images user: "generate a grid of robot designs" </example>
version	1.0.0

Image Generation Skill

Generate AI images using ComfyUI on junkpile via the tsr CLI and display results on the MARAUDER visor.

Quick Reference

SKILL="${CLAUDE_PLUGIN_ROOT}/skills/tsr/tsr.sh"

# Generate an image (default: SDXL, 1024x1024, 20 steps)
bash $SKILL generate "a giant mech robot in fog"

# Generate with options
bash $SKILL generate "portrait photo" --model dreamshaper --steps 30

# Generate and display on HUD
bash $SKILL generate "cyberpunk city at night" --hud

# Generate multiple images
bash $SKILL batch "robot designs" --count 4

# Generate and display as grid on HUD
bash $SKILL batch "landscape scenes" --count 9 --hud --columns 3

# Display an existing image on HUD
bash $SKILL hud /tmp/image.png
bash $SKILL hud /tmp/image.png --title "RECON" --caption "Target acquired"

# Display multiple images as grid on HUD
bash $SKILL grid /tmp/img1.png /tmp/img2.png /tmp/img3.png

# List available models
bash $SKILL models

# Check ComfyUI status
bash $SKILL status

Commands

Command	Description
generate <prompt> [opts]	Generate a single image
batch <prompt> [opts]	Generate multiple images
hud <file> [opts]	Display image on HUD viewport
grid <files...> [opts]	Display image grid on HUD viewport
models	List available checkpoint models
status	Check ComfyUI/tensors API status

Generate Options

Flag	Default	Description
--model, -m	(SDXL default)	Checkpoint model name
--steps	20	Sampling steps
--width, -W	1024	Image width
--height, -H	1024	Image height
--cfg	7.0	CFG scale
--seed, -s	-1 (random)	Random seed
--negative, -n	(auto)	Negative prompt
--lora, -l	(none)	LoRA model name
--output, -o	/tmp/tsr-*.png	Output path
--hud	false	Display result on HUD viewport
--title	"GENERATED"	HUD display title
--caption	(prompt text)	HUD display caption

Batch Options

All generate options plus:

Flag	Default	Description
--count, -c	4	Number of images to generate
--columns	auto	Grid columns when --hud is used

HUD Display Options

Flag	Default	Description
--title	"IMAGE"	Header title text
--caption	(filename)	Caption below image
--classification	"[ GENERATED ]"	Right-aligned tag
--no-tint	false	Disable green tint

Infrastructure

• ComfyUI: Running on junkpile (10.0.0.2), RTX 2000 Ada GPU
• tensors API: Port 5003 on junkpile
• tsr CLI: /Users/chi/.local/bin/tsr
• Remote flag: --remote junkpile (always used, generation runs on GPU)
• HUD bridge: http://127.0.0.1:9876 (MARAUDER visor eval endpoint)

Default Negative Prompt

When no --negative is provided, the skill appends: blurry, low quality, text, watermark, deformed, ugly

Models (as of 2026-04-10)

Run bash $SKILL models for current list. Known checkpoints:

• SDXL base (default)
• DreamShaper 8 (SD 1.5)
• Super Robot Diffusion Rise (mecha)
• MechaDream (mecha)

Prerequisites

• tsr CLI installed (/Users/chi/.local/bin/tsr)
• ComfyUI running on junkpile
• MARAUDER visor running (for --hud display)
• junkpile reachable at 10.0.0.2