Living Culinary Codex · publiczny skill Open Skill
Living Culinary Codex
Living Culinary Codex
Health
0 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4150474/2026-10-15/n4vxfytx.jpg
# Living Culinary Codex Turn a dish into an oversized antique culinary codex, photographed the way a museum specimen is: the left page is a glowing miniature origin kitchen, the right page is a readable illustrated recipe, and the dish itself shares the frame with the open codex. This skill has no structured state: one conversation, one finished image, driven by the dialogue and the canvas assets. ## When to use - The user gives (or merely names) a dish and wants to see this "codex" image. - The user brings a photo of a dish and wants to keep its recognizable traits (color, texture, vessel, plating) while getting this style. - The user wants to revise an existing render: swap the dish, change the origin scene, adjust the lighting, or make the recipe page more legible. ## Input and output - Input: text (dish name, plus origin / recipe material when available) + optional reference image (a photo of the dish). - Output: a single vertical image, 3:4, both pages in frame. **One call generates one image**; for several dishes or several versions, call once per image. ## Steps ### 1. Pin down the dish Extract the dish name from the user's input. If no dish is specified (they only said "make me a codex"), ask which dish first — do not pick one for them. ### 2. Origin and recipe material - If the user already supplied origin / recipe material: use it directly and note the source. - If historical accuracy matters and the user gave nothing: use `search_web` to check the dish's place of origin, the rough era, key ingredients and technique. Put only **verifiable** facts into the image. - If it is disputed or unfindable: do not invent anything (see the first item under "Notes"). ### 3. Assemble the English prompt Build it from the template under "Prompt conventions" below. **Keep the fixed passages verbatim and replace only the curly-brace slots.** The prompt is in English; keep the dish name as the user gave it (an English or romanized name may be appended). The right-page recipe body should be 3–6 short English lines or numbered fragments (short lines render more reliably). ### 4. Generate Text-to-image only: ``` generate_image( prompt = <English prompt assembled from the template>, model = 't2i', orientation = 'portrait', # vertical aspect_ratio = '3:4', # fixed frame spec for this skill name = '<dish name> · Living Culinary Codex', entity_type = 'codex_images', ) ``` When the user supplies a photo of the dish: ``` generate_image( prompt = <the same prompt>, input_urls = [<the user's dish photo>], # reference for dish form / color / vessel / plating only model = 'i2i', orientation = 'portrait', aspect_ratio = '3:4', name = '<dish name> · Living Culinary Codex', entity_type = 'codex_images', ) ``` - Use `i2i` when a reference image exists, `t2i` when it does not; if you need finer paper fiber, vessel and handwriting detail, step up to `i2i.hd` / `t2i.hd` (`.hd*` does not support NSFW content, which this skill never needs). - If the two pages' lettering must be sharper, pass `resolution='1080P'`; otherwise leave it out (default tier). - Do **not** paste the whole reference photo onto the codex pages: it only makes the dish itself look more like real food; both pages still follow the template. ### 5. Reply Summarize in one or two natural-language sentences: which dish, what origin scene the left page shows, which ingredients / steps the right page lists. **Do not paste long URLs.** When useful, call `set_guidance` in the **same turn** with 2–3 follow-up actions tied to the current render (for example "do one for a different dish", "change the left page to a night-market stall scene", "redo it with a dish photo attached"), and do not call any other tool in that turn. (No need for `write_todos` when making a single image; if the user names several dishes and wants a run of codices, first `write_todos` to list the dish