Defining Required vs Optional Elements

When building a multi-asset campaign, you often start with one perfect 'hero shot' and then use AI to create variants for different sizes and platforms. Reference drift happens when the product's appearance shifts gradually across these assets. Because generative AI doesn't have a memory of the physical product, it makes a new statistical guess every time you prompt it. Small, unnoticed changes in one image compound in the next.

Consider a campaign for a premium olive oil. In asset one, the oil is a rich, deep gold. By asset six, it has shifted to a bright yellow. By asset twelve, the oil looks pale and the bottle's cap has subtly changed shape. Because these changes happened incrementally, you don't notice them until you look at the entire set side-by-side. At that point, the product in the final assets no longer matches the physical bottle on the shelf. Fixing this requires redoing half the set, wasting hours of production time. To prevent this, you must establish a 'ground truth'—a locked set of references that every new asset is checked against before it is approved.

In professional product imagery, the primary goal isn't just a beautiful picture; it is accuracy. The ground truth standard means the product shown in an advertisement must match the actual physical product on the shelf. If a customer buys a product based on a visual in an ad, and that product looks or behaves differently in reality, it is considered misleading. In many regions, advertising regulators hold the advertiser responsible for all AI output, regardless of whether the error was a human choice or a machine hallucination.

Take the example of a robot-puppy toy. The company used AI to generate footage of the toy in various settings. However, the AI-generated version began to behave like a living, breathing dog rather than a mechanical toy. Because the footage suggested capabilities the actual product did not possess, regulators blocked the ad. The images were visually polished, but they failed the ground truth standard.

As a producer, you apply this standard by treating the original photograph as the only source of truth. You use that original shot to define what is 'untouchable' before you ever open a generative tool.

To maintain the ground truth standard, you must identify required elements. These are the specific, untouchable product details that must remain identical to the original photograph across every single asset in the campaign. If a detail is marked as required, it cannot be altered, interpreted, or 'improved' by generative AI. Any shift in these pixels is considered a production error, not a creative choice.

Consider the olive-oil campaign. You start by auditing the hero shot and listing the required elements in your visual brief. First, the label text is required; the AI cannot hallucinate a new font or misspells a word. Second, the cap geometry—the exact curve and height of the plastic—is required to ensure the bottle's silhouette remains consistent. Third, the exact oil color is required; if the physical product is a deep amber, the assets cannot shift toward yellow.

By explicitly marking these details as required, you create a hard boundary. When you review a generated asset, you don't ask if it looks 'good'; you check these specific elements against the original photograph to ensure they haven't drifted.

While required elements are locked to preserve the product's identity, optional elements are the environmental settings and props that can be safely generated or altered by AI. These elements provide the mood, context, and storytelling of the campaign without compromising the ground truth. Because these parts of the image do not define the physical product, you can use generative tools to experiment with them to fit different platforms or aesthetic needs.

In our olive-oil example, you might decide that the original photo was shot in a sterile studio, but the campaign needs an 'outdoor Mediterranean' feel. You categorize the stone table texture, the late-afternoon sunlight, and the addition of a decorative olive branch as optional. These elements can be entirely hallucinated by the AI.

As a designer, you use this categorization to decide where to spend your review time. You don't need to stress if the AI generates a slightly different stone grain or a different leaf on the olive branch; these are creative variations. By clearly labeling these as optional in the brief, you give your team permission to use the speed of AI for the setting while keeping the focus on protecting the required product pixels.

The generative boundary is the precise line you draw in your brief that tells the operator—and the AI—where it is permitted to invent pixels and where it must be restricted. While the previous steps categorized what is required and optional, the boundary defines where those rules apply spatially within the image frame. Without a clear boundary, a tool like Adobe Firefly may 'bleed' into required areas, subtly altering a product's edge or color while trying to blend it into a new background.

In the olive-oil campaign, your boundary specifies that Firefly is authorized to generate everything outside the bottle's silhouette, including the stone table and the distant Mediterranean landscape. However, the pixels within the silhouette of the bottle are strictly off-limits. You treat the bottle as a protected zone. If the AI attempts to 'enhance' the lighting on the glass by changing the oil's color or softening the label's edge, it has crossed the boundary. By establishing this spatial limit, you ensure that the generative power of the AI is used only for the environment, leaving the ground truth of the product untouched.

To prevent reference drift from entering your pipeline, you must move these definitions out of your head and into a structured visual brief. A professional brief acts as a contract between the producer and the designer; it explicitly separates required product data from optional creative choices so there is no ambiguity during the generation process. Without this structure, a designer might mistakenly 'fix' a product color to match a generated background, inadvertently breaking the ground truth standard.

In the olive-oil campaign, the producer organizes these requirements into a clear table. Under the Required column, they list the Label, the Cap, and the Oil Color. Under the Optional column, they list the Lighting, the Table Texture, and the Background Foliage.

When the production begins, the designer uses this table as a pre-flight checklist for every asset. Before exporting a variant, they look at the 'Required' list and verify that each item remains identical to the hero shot. If the lighting in the 'Optional' column has shifted from golden hour to midday, it is a creative choice. If the oil color in the 'Required' column has shifted from deep amber to pale yellow, it is a production error that must be fixed before the asset is approved.

The final step in establishing your ground truth is validation. Even with a structured brief, errors occur during the planning phase—often because a producer or designer assumes a detail is 'obvious' and forgets to list it as required. Validation is the process of reviewing the completed brief against the actual physical product to ensure no essential identity marker has been accidentally categorized as optional. If a product detail is left off the 'Required' list, the AI operator will treat it as a variable that can be 'improved' or altered to fit the scene.

Imagine the olive-oil campaign again. During the final brief review, the designer notices that while the label and oil color are marked as required, the unique, tapered shape of the bottle's neck is missing from the list. By default, this makes the neck shape 'optional.' If the designer sends this brief to production, Adobe Firefly might subtly thicken the neck or change its angle to better balance the composition of a vertical social media asset. To the AI, this is a helpful adjustment; to the brand, it is a failure of the ground truth standard. The designer catches this, moves 'Bottle Neck Geometry' to the Required column, and locks the boundary before a single pixel is generated.