Generate reusable art style references from a set of input images. Produces a visual reference sheet and prose description that capture the shared style, enabling consistent visual identity across future image generation.
Generates a reusable art style reference from a set of input images. Produces both a visual reference sheet (via Gemini) and a prose style description, along with a manifest tracking the source material.
Before proceeding, invoke /loreduck:shell-environment to learn about the shell
commands available in this skill and the environment they run in.
loreduck/styles/ folder.loreduck/ folder.Interpret $ARGUMENTS as a description of the desired styleguide. Extract two things:
character-portrait, "watercolor landscapes" becomes watercolor-landscape).**/*Lark*, **/*Cawthorn*). Confirm each match with the user if ambiguous.Create the output directory if it doesn't exist:
loreduck/styles/<style-name>/
Use compose_image with all input images to produce a style reference sheet:
compose_image "<prompt>" "loreduck/styles/<style-name>/visual-guide.webp" <input-image-1> <input-image-2> ...
The prompt should ask Gemini to create a style reference sheet — not a collage or composite, but an analytical breakdown. The prompt must instruct the model to extract and present:
Example prompt structure:
Create an art style reference sheet analyzing these images. Extract and present:
the shared color palette as labeled swatches, the rendering technique and
brushwork style, the lighting approach, and any recurring compositional motifs.
Lay this out as a clean reference sheet an artist could use to reproduce the style.
Read each input image directly (Claude is multimodal) and write a prose style description to:
loreduck/styles/<style-name>/textual-guide.md
The prose guide should describe the shared art style across these dimensions:
Write it as a concise, actionable reference — something that could be dropped into a future image generation prompt to reproduce the style.
If any input image does not already exist somewhere within the CWD, copy it into loreduck/styles/<style-name>/references/ so the styleguide is self-contained.
Use standard file operations (Bash cp) to bring external files in.
Compute a SHA-256 hash for each input image and write a manifest to:
loreduck/styles/<style-name>/manifest.json
Schema:
{
"style": "<style-name>",
"inputs": [
{
"source": "<path-relative-to-manifest>",
"sha256": "<hex-digest>"
}
],
"outputs": {
"visual": "visual-guide.webp",
"prose": "textual-guide.md"
}
}
Compute hashes using sha256sum:
sha256sum path/to/image.webp
The source path for each input should be relative to the manifest file itself (i.e., relative to loreduck/styles/<style-name>/). If an image was copied in during Step 5, use its new path (within the styleguide folder) as the source.
The default styleguide is tracked as a relative symlink at loreduck/styles/default pointing to the directory of a named styleguide (e.g., default -> character-portrait).
Check whether loreduck/styles/default already exists:
If it does not exist (this is the first styleguide), automatically set the new styleguide as the default:
ln -s <style-name> loreduck/styles/default
If it already exists (other styleguides are present), ask the user whether they'd like to make the new styleguide the default. If they agree, update the symlink:
ln -snf <style-name> loreduck/styles/default
If the user asks to change the default styleguide (via /loreduck:styleguide or otherwise):
List the available styleguides by scanning directories in loreduck/styles/ (excluding the default symlink itself).
If the user specified a style name, resolve it directly. Otherwise, present the available options and let them choose.
Update the symlink:
ln -snf <style-name> loreduck/styles/default
Confirm the change to the user.
When the user invokes /loreduck:styleguide, interpret $ARGUMENTS as a description of the style to capture or a request to manage styleguides. If the request is to create a new styleguide, execute Steps 1–7 in order: parse the request, set up the output folder, generate the visual reference sheet, write the prose guide, handle any external images, create the manifest, and manage the default symlink. If the request is to change the default styleguide, follow the "Changing the default on request" procedure in Step 7. Present the results to the user with paths to the generated files.