Local Image Generator

Generate images locally using Stable Diffusion. Auto-detects your hardware and picks the optimal model, device, and resolution.

Phase 0: Detect Compute Environment

Run this at the start of every invocation. It determines everything downstream.

python3 -c "
import platform, shutil, subprocess, json

info = {'os': platform.system(), 'arch': platform.machine(), 'ram_gb': 0, 'gpu': 'none', 'vram_gb': 0, 'device': 'cpu', 'dtype': 'float32'}

# RAM
try:
    if platform.system() == 'Darwin':
        import os; info['ram_gb'] = round(os.sysconf('SC_PAGE_SIZE') * os.sysconf('SC_PHYS_PAGES') / (1024**3))
    elif platform.system() == 'Linux':
        with open('/proc/meminfo') as f:
            for line in f:
                if line.startswith('MemTotal'):
                    info['ram_gb'] = round(int(line.split()[1]) / (1024**2))
                    break
    else:
        import ctypes
        mem = ctypes.c_ulonglong(0)
        ctypes.windll.kernel32.GetPhysicallyInstalledMemory(ctypes.byref(mem))
        info['ram_gb'] = round(mem.value / (1024**2))
except: pass

# GPU detection
try:
    import torch
    if torch.cuda.is_available():
        info['gpu'] = torch.cuda.get_device_name(0)
        info['vram_gb'] = round(torch.cuda.get_device_properties(0).total_mem / (1024**3))
        info['device'] = 'cuda'
        info['dtype'] = 'float16'
    elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
        info['gpu'] = 'Apple Silicon (MPS)'
        info['vram_gb'] = info['ram_gb']  # unified memory
        info['device'] = 'mps'
        info['dtype'] = 'float16'
    elif hasattr(torch, 'hip') or 'AMD' in str(getattr(torch, '_C', '')):
        info['gpu'] = 'AMD (ROCm)'
        info['device'] = 'cuda'  # ROCm uses cuda API
        info['dtype'] = 'float16'
except ImportError:
    pass

print(json.dumps(info))
"

Parse the JSON output and store it internally as COMPUTE. Present the results to the user:

Detected hardware:

• OS: {os} ({arch})
• RAM: {ram_gb} GB
• GPU: {gpu} ({vram_gb} GB VRAM)
• Compute device: {device}

Model Selection Matrix

Based on the detected hardware, recommend a model from this table:

Present the recommendation and let the user choose:

AskUserQuestion: "Which image generation model should I use?"
Options:
- [Recommended model] (Recommended) — [reason based on their hardware]
- SDXL-Turbo — Best quality, needs 8+ GB VRAM, ~6 GB download
- SD-Turbo — Good quality, needs 4+ GB VRAM, ~3 GB download
- Tiny-SD — Lower quality, runs on any hardware, ~1 GB download

Store the chosen model as MODEL.

Resolution Selection

Based on model and VRAM:

Steps Selection

Phase 1: Determine What to Generate

If the user provided a slug and prompt (argument after the skill name), parse them and skip to Phase 3.

Expected argument format: {slug} {prompt} (e.g., beginners-guide-to-rag abstract knowledge retrieval system with floating documents)

If only a slug was provided, read the blog post to generate an appropriate prompt:

# Find the blog post
cat content/blog/{SLUG}.mdx 2>/dev/null | head -50

Extract the title, description, and key themes. Generate a prompt using one of the SAI style templates below that best fits the post's topic. Each image MUST use a different style to avoid visual repetition across blog posts.

SAI Style Templates (pick ONE per image)

Each template wraps your subject description in a distinct visual style. Replace {subject} with a short, vivid description of the post's core concept as a visual metaphor.

Subject Description Guidelines

Write the {subject} as a vivid visual metaphor, not a literal description. Never include hands, fingers, faces, or human figures.

• Good: "a crystalline data pipeline splitting light into rainbow streams"
• Bad: "data pipeline architecture diagram"
• Good: "mechanical clockwork gears meshing with glowing circuit traces"
• Bad: "AI system with nodes and connections"

Vary across posts: color palette, physical metaphor (clockwork, rivers, crystals, bridges, constellations), and composition.

If no input was provided, use AskUserQuestion to ask for a slug and description.

Phase 2: Confirm with User

Present the generation plan:

I'll generate a hero image for {slug}: Model: {MODEL} on {device} Resolution: {width}x{height} ({steps} steps) Prompt: "{prompt}" Seed: {seed or "random"} Estimated time: {estimate based on device and model}

Want me to adjust anything before generating?

Time estimates:

Phase 3: Install Dependencies

Check and install what's needed based on platform:

# Check Python + torch
python3 -c "import torch; print(torch.__version__)" 2>&1

If torch is missing, install based on platform:

Then install diffusers: