Generate and Edit Images with AI

This skill is an intelligent image-generation studio that automatically routes a request between two models: ai-studio-image for humanized, influencer-style human photos, and stability-ai for art, illustration, and image editing. It detects the type of image requested and picks the right model automatically, covering generation, editing, upscaling, background removal, inpainting, and realistic photos of people in one workflow. The source frames the model itself as a "visual creative director" choosing the right brush for each piece: humanized photos with Gemini, art and editing with Stability.

Its decision matrix asks, in order: is this a realistic photo of a person or influencer (route to ai-studio-image); if not, is it an illustration, art, or drawing (route to stability-ai in generate, ultra, or core mode); if not, is it an edit of an existing image (route to stability-ai img2img, inpaint, search-replace, or erase); if not, is it an upscale or background removal (route to stability-ai upscale or remove-bg); otherwise ask for more detail.

ai-studio-image runs on Gemini 2.0 Flash and specializes in hyper-realistic photos of people with a human touch, for requests like an Instagram-style lifestyle photo, a professional headshot, or a humanized educational photo. Its advantages are that it is free (gemini-2.0-flash-exp), applies five layers of narrative humanization (device, lighting, imperfection, authenticity, environment), and ships 20 preconfigured templates (10 influencer, 10 educational) with subtle imperfections that make the photo feel credible. Its limitations are one image at a time in about 9 seconds, roughly 1K resolution, no custom aspect ratio, and a 50-image-per-day free tier.

stability-ai runs on SD3.5 Large under a Community license and specializes in art, illustration, and image editing and manipulation, with modes generate (art/illustration), ultra (maximum quality), core (fast iteration), img2img (transform an image), upscale and upscale-creative, remove-bg, inpaint (edit an area), search-replace (swap an object), and erase (remove an object). It offers 15 styles: photorealistic, anime, digital-art, oil-painting, watercolor, pixel-art, 3d-render, concept-art, comic, minimalist, fantasy, sci-fi, sketch, pop-art, and noir. Its limitations are that it consumes credits under the Community license and is not specialized for realistic photos of people.

The generation workflow is: analyze the requested image type and goal, select the right model from the decision matrix, build a prompt optimized for that model, generate with the correct parameters, present the result with metadata, and offer variations or adjustments. Example commands from the source include:

python generate.py remove-bg --image produto.jpg

For ai-studio-image prompts, the recommended structure is subject, action or pose, environment, lighting, and a human detail, avoiding art terms, artist names, or non-photographic styles. For stability-ai prompts, the recommended structure is subject, action, artistic style, cinematic lighting, quality terms, a reference artist, and colors, with useful negative prompts such as "blurry, low quality, watermark, text, ugly, deformed, extra fingers, bad anatomy, worst quality."

If ai-studio-image fails (daily limit or API error), the skill falls back to stability-ai in ultra mode with an adapted prompt and informs the user of the model switch; if stability-ai fails on insufficient credits, it falls back to ai-studio-image with an adapted prompt, or, if that image type isn't supported, points the user to recharging; if both fail, it generates a detailed prompt the user can run manually and suggests DALL-E, Midjourney, or Leonardo AI as alternatives.

Use it when you need specialized help generating, editing, upscaling, or background-removing an image, and want the right underlying model picked for you automatically rather than choosing between a photo-realism model and an art model yourself.

Do not use it when the task is unrelated to image generation, when a simpler and more specific tool already covers the request, or when the user needs general-purpose assistance without image-generation expertise.

Inputs are the image request in natural language, and for editing modes an existing image file and, where needed, a mask. Outputs are a generated or edited image saved to a path, with its dimensions and file size, the optimized prompt actually used, and suggested variations such as an art-style version, a humanized version, or a style/lighting adjustment.

It is for anyone generating marketing, social, or illustrative images who wants automatic routing between a realistic-people model and an art/editing model instead of manually picking and prompting each one, and who is comfortable with the Windows-path scripts referenced for both underlying skills.