Generate and iterate on images locally with Stable Diffusion models.
npx skills add https://github.com/amd/gaia --skill image-gen-amd
Please help me install this Agent Skill. Skill: image-gen Source: https://github.com/amd/gaia/tree/main/hub/skills/image-gen Command: npx skills add https://github.com/amd/gaia --skill image-gen-amd
list_sd_models() and respects per-model defaults for steps, CFG scale, and size (e.g., SDXL-Turbo vs SDXL-Base-1.0). - Seed-Based Iteration: Uses get_generation_history() to reuse prompts and seeds so requests like "same but at sunset" modify the existing image rather than producing an unrelated one. - Use Case: A user asks for a logo concept, then says "make it warmer" — the Skill retrieves the prior generation's parameters, adjusts only the lighting clause, and regenerates with the same seed. ## Quick Start Ask the assistant to draw a picture of your subject, for example: "Generate an image of a red bicycle leaning against a brick wall in morning sunlight."High-intent search queries and answers about installing and using this skill.
Describe the subject, setting, style, and lighting you want, and the generate_image tool renders it with a local Stable Diffusion model. The Skill expands short prompts automatically and reports the saved file path and the exact prompt used.
Ask for a change like "same but at sunset" and the Skill reads the previous generation from get_generation_history, modifies only the requested part of the prompt, and reuses the same seed so the new image stays recognizably similar.
SDXL-Turbo is the default and converges in about 4 steps with CFG around 1.0, producing images in seconds. SDXL-Base-1.0 at 1024x1024 takes minutes but suits photorealistic output when the user accepts the wait.
The first use of a model downloads several gigabytes of weights, which can exceed the request timeout. Pre-fetch the model with "lemonade-server pull <model>" and retry rather than restarting the server.
No. All generation calls go through Lemonade Server for local inference, so it must be started first. If the tool reports it cannot reach the server, start Lemonade Server and retry.
No. SDXL-Turbo is distilled for roughly 4 steps and guidance-free sampling, so raising steps only multiplies wait time and raising CFG scale degrades the image. Leave steps, cfg_scale, and size unset unless you have a specific reason.
LocalKinAIGenerate images from prompts via a local OllamaDiffuser API server.
prasanna591Generates multi-format banner designs for social media, ads, web heroes, and print with AI-generated visuals.
inbharataiGenerate alt text, captions, and metadata for images.
taracodlabsGenerate images from text prompts using Stable Diffusion via HuggingFace Diffusers or Inference API.