Generate and Edit Images with AI

Image-Studio is a routing skill that sits in front of two backend image generators - ai-studio-image (Gemini 2.0 Flash) and stability-ai (Stable Diffusion 3.5 Large) - and picks between them automatically based on a decision matrix: a realistic photo of a person or influencer routes to ai-studio-image, while illustration, art, image edits, upscaling, and background removal route to stability-ai. Ai-studio-image is used for hyper-realistic people photos, is free, applies "5 layers of narrative humanization" (device, lighting, imperfection, authenticity, environment) for credibility, ships 20 pre-configured templates (10 influencer, 10 educational), but is limited to one ~1K-resolution image at a time and a 50-image/day free tier, and doesn't support a custom aspect ratio. Stability-ai covers ten generation modes - generate, ultra, core, img2img, upscale, upscale-creative, remove-bg, inpaint, search-replace, and erase - across 15 named styles (photorealistic, anime, digital-art, oil-painting, watercolor, pixel-art, 3d-render, concept-art, comic, minimalist, fantasy, sci-fi, sketch, pop-art, noir), and runs on a community-license credit system rather than being free. Both backends are driven through a shared generate.py script, for example:

python generate.py --template "instagram-lifestyle" --customization "cafe, morning, smile"

The skill also defines a fallback chain: if ai-studio-image fails (daily limit, API error), it retries via stability-ai's ultra mode with an adapted prompt; if stability-ai fails (insufficient credits), it retries via ai-studio-image; if both fail, it hands the user a ready-to-use prompt and suggests external alternatives like DALL-E, Midjourney, or Leonardo AI.

Use it for any image generation, editing, upscaling, background-removal, or inpainting request where the right backend isn't already obvious - the skill's job is exactly that routing decision. Do NOT rely on it for realistic person photography where photorealistic accuracy above roughly 1K resolution or a custom aspect ratio matters, since ai-studio-image is capped on both; stability-ai is explicitly noted as not specialized for realistic people photos, so a request combining both needs (a high-res, custom-ratio realistic portrait) falls outside what either backend does well.

Input is a natural-language image request; the skill classifies it via its decision tree (photo of a person, illustration/art, edit of an existing image, upscale/background removal, or ask for more detail) and builds an optimized prompt for whichever backend it selects, following separate prompt structures for humanized photos versus stylized art, including a list of negative-prompt terms for stability-ai (blurry, watermark, extra fingers, bad anatomy, and similar). Output is the generated or edited image file plus a response summary naming the model used, the mode, generation time, file path, dimensions, and file size, along with offered variations such as an alternate-model version or a style/lighting adjustment.

Built to run as a skill inside Claude Code, Antigravity, Cursor, Gemini CLI, and Codex CLI, and depends on two separate underlying skills/scripts - ai-studio-image (Gemini 2.0 Flash) and stability-ai (Stable Diffusion 3.5 Large) - each exposing its own generate.py entry point.

Anyone generating or editing images through an AI coding or agent tool who wants one command to automatically pick between a free realistic-photo model and a paid-credit art/editing model, rather than manually choosing and re-prompting for each backend.