The __wildcard__ syntax used by the sd-dynamic-prompts extension and A1111 swaps part of a prompt for a random pick from a list. Drop a token like __color__ or __pose__ into your template and it expands to a different word each generation. That's powerful, but until you actually render images and read the logs it's hard to know what a template really expands to every run — or whether a typo is pointing at a wildcard you never defined. This tool runs just that expansion in your browser ahead of time and previews a list of sample prompts. In the 'Wildcards' box, write a header line like `color:` and list one option per line below it (define as many wildcards as you like, stacked one after another). In 'Template', write a prompt containing tokens like __color__ and inline choices like {soft|harsh}; the right side fills with the requested number of randomly expanded samples. __name__ picks one option at random from that wildcard; {a|b|c} picks one on the spot — and a wildcard's option may itself contain __another_wildcard__ or {x|y}, which are expanded recursively (nested choices like {a|{b|c}} work too). Fix the seed and the exact same samples reproduce every time, so you can share results or isolate a bug; hit 'Reroll' to jump to a fresh seed and see different draws instantly. When you want every possible combination of {a|b|c} exhaustively, use prompt-matrix instead; this tool is for previewing a handful of random samples close to what a real generation produces. If you reference a wildcard you haven't defined, its token is left in place and its name is listed in a warning, so typos surface immediately. All processing happens on your device — your definitions, template, and generated samples are never uploaded, stored, or sent anywhere.