Shima Styler (Gallery)

"Watercolor" is an easy prompt word. Knowing whether a specific artist's watercolor look is what you want is the hard part, and a text dropdown doesn't help you decide. Shima StyleGallery renders the style library as a visual grid inside the node - thumbnails you can click to select - so you're picking a look you can see rather than a name you're guessing at. It's the Gallery member of the pack's Styler family, and it's the one you open when discovery is the job.

The frontend does the heavy lifting: an HTML widget renders the grid, filters by category, and tracks your selected styles, writing the picks back into a hidden selected_styles field. The backend takes that selection, resolves each entry against the library, and builds the styled conditioning.

The inputs

• base_prompt - your prompt, merged with the selected style(s).
• mode - Single or Stack. Single for one style; Stack for mixing several, joined with the connector.
• connector - ", " (default), " and ", " mixed with ", " + ".
• style_strength - −10 to 10, default 1. It applies surgical weighting: only the style portion gets (…:strength), leaving your base prompt at 1.0 - so dialing the style down doesn't drag your subject down with it.
• show_missing - Show All vs Hide Missing Images; toggles whether entries without a thumbnail still appear (they do as text-only rows).
• clip, conditioning_positive, conditioning_negative, base_string/negative_string**, selected_styles - the conditioning inputs plus the hidden selection field.

Outputs: pos_clip_out, neg_clip_out (conditioning), positive, negative (strings), and style_name (the name of the last-applied style, handy for filenames).

How it merges styles

Selected entries become "style of {name}" templates (user styles use their stored positive), optional negatives are collected from user styles, and prompt injection works through {prompt} macros where the template defines one. Weighting is applied only to the style tokens, then the base and style are combined.

Installing it

Part of KDB-USJP/shima_wf:

cd ComfyUI/custom_nodes
git clone https://github.com/KDB-USJP/shima_wf

Restart ComfyUI. Style data and thumbnails ship in the repo (assets/data/shim_a_sheets.xlsx plus the frontend data file); openpyxl is auto-installed on first run. If the grid shows no images, the pack's js assets didn't load - hard-refresh your browser.

The honest take

Gallery is the most pleasant way into the style library and the least automatable: it's a human-picking node, so it belongs in an interactive workflow, not a batch job. Two things to know. First, "show missing" - the data file has hundreds of artists and only some ship with thumbnails, so expect text-only rows unless you add your own images. Second, like the whole Styler family, the artist list is positional: reorder the XLSX and the same thumbnail may point at a different artist. For "find the look by eye, then lock it," nothing else in the pack competes.