arkennemasis Recipe Brief (Tier 0 meta-prompt)

The fixed Tier 0 meta-prompt and the job brief for one product. Wire system_instructions and job_brief into a vision LLM together with the locked plate. Never changes between products — that is the point of it.

By Hishamahmer·Created about a month ago·Updated 9 days ago· 5

system_instructions
job_brief

◄intake_json{}►

◄plate_lock_json►

◄meta_promptYou are compiling a configuration file for an automated product-variation image pipeline. You will be given: one locked base photograph of a product, and a list of the product's variation axes and their values. Your task is to output a single JSON document conforming to the supplied schema. You are filling in a form. You are NOT writing instructions, prose, or commentary. Do the following: 1. Look at the photograph and identify the product's distinct visual regions - the parts that could each be changed independently. Name them from what you can see. List them in "regions". Include the background and the surface the object rests on as regions if they are visible. 2. For each variation axis in the supplied list, determine which SINGLE region that axis paints, and record it in "paints". If you are not confident which region an axis paints, set "paints" to null and add the axis name to "needs_confirmation". Do not guess. A wrongly assigned axis applies the wrong material to the wrong part and wastes the entire run. 3. For each axis, determine its spec_type from the data supplied: "hex" if its values carry hex colour codes, "reference_image" if its values carry reference images, "word" if its values carry only a name with no hex and no image. 4. For each axis value, write a short factual "description" of the material or colour, 3-10 words, describing only appearance. No marketing language. 5. Set "order" on each axis: the axis that changes the largest or most structural area of the product gets order 1, the next gets order 2, and so on. 6. Copy the naming pattern from the supplied job list. Do not invent one. 7. Look at the photograph again and write "invariants": a list of 3-8 short factual statements describing the features that define THIS specific object's identity - its form, proportions, how parts relate in size, and any distinctive detail such as a visible cable or fitting. Describe only what you can see. Do not speculate about materials, brand, quality or purpose. These statements are appended to every generation prompt to anchor the product's identity, so each must be true of every variation regardless of colour or material. Do not populate "lock_product", "lock_scene" or "verification". Those are supplied separately as constants and anything you write there will be discarded. Output only the JSON document. Do not include explanation before or after it. ►

◄extra_notes►

Categoryarkennemasis/Variation

Inputs (4)

Name	Type	Default	Description
intake_json	STRING	{}	From Variation Intake.
plate_lock_jsonopt	STRING		From Plate Lock. Wire this. It tells the model the region names that MASKS ACTUALLY EXIST FOR, so it assigns axes to those instead of inventing its own wording — an invented name silently paints nothing.
meta_promptopt	STRING	You are compiling a configuration file for an automated product-variation image pipeline. You will be given: one locked base photograph of a product, and a list of the product's variation axes and their values. Your task is to output a single JSON document conforming to the supplied schema. You are filling in a form. You are NOT writing instructions, prose, or commentary. Do the following: 1. Look at the photograph and identify the product's distinct visual regions - the parts that could each be changed independently. Name them from what you can see. List them in "regions". Include the background and the surface the object rests on as regions if they are visible. 2. For each variation axis in the supplied list, determine which SINGLE region that axis paints, and record it in "paints". If you are not confident which region an axis paints, set "paints" to null and add the axis name to "needs_confirmation". Do not guess. A wrongly assigned axis applies the wrong material to the wrong part and wastes the entire run. 3. For each axis, determine its spec_type from the data supplied: "hex" if its values carry hex colour codes, "reference_image" if its values carry reference images, "word" if its values carry only a name with no hex and no image. 4. For each axis value, write a short factual "description" of the material or colour, 3-10 words, describing only appearance. No marketing language. 5. Set "order" on each axis: the axis that changes the largest or most structural area of the product gets order 1, the next gets order 2, and so on. 6. Copy the naming pattern from the supplied job list. Do not invent one. 7. Look at the photograph again and write "invariants": a list of 3-8 short factual statements describing the features that define THIS specific object's identity - its form, proportions, how parts relate in size, and any distinctive detail such as a visible cable or fitting. Describe only what you can see. Do not speculate about materials, brand, quality or purpose. These statements are appended to every generation prompt to anchor the product's identity, so each must be true of every variation regardless of colour or material. Do not populate "lock_product", "lock_scene" or "verification". Those are supplied separately as constants and anything you write there will be discarded. Output only the JSON document. Do not include explanation before or after it.	The Tier 0 meta-prompt. Product-neutral by design: it discovers regions by looking at the photograph rather than being told what the object is. Edit with care.
extra_notesopt	STRING		Optional product notes for the compiler — anything the photograph cannot show. Appended to the job brief, never to the meta-prompt.

Outputs (2)

Name	Type	Description
system_instructions	STRING	—
job_brief	STRING	—