Anyone drafting a batch of campaign visuals with an object-reference workflow eventually hits the same wall: the tenth variation looks nothing like the first, and nobody can say why. The product is technically still the product, but the lighting shifted, the angle drifted, the background changed personality, and now the creative director is asking which version is actually "on brief." The real problem isn't the tool generating the images. It's that nobody decided in advance what was allowed to move and what wasn't.

Why Drafts Drift Without a Fixed Anchor

Object-reference generation works by holding onto some visual information from a source image while reinterpreting the rest. That's the entire appeal — you can keep a product's shape or a mascot's silhouette while exploring new settings, moods, or compositions. But "reinterpreting the rest" is a wide net. Without an explicit list of locked attributes, small inconsistencies compound. A bottle's label wraps differently in one frame. A logo's color temperature shifts under a warmer light. A character's proportions stretch slightly wider in a horizontal crop than a vertical one.

None of these drifts happen because a workflow is broken. They happen because the brief was implicit rather than written down. Treating "stay on brand" as a single instruction is like telling a photographer to "make it look good" — technically true, practically useless.

A Reasoning Framework: Sort Elements Into Three Buckets

Before generating anything, it helps to sort the visual elements of a campaign asset into three categories:

Fixed — attributes that must not change across any draft: exact product shape, logo placement, brand color values, mandatory text, packaging proportions.

Bounded — attributes allowed to vary within a defined range: lighting direction, background texture, camera angle within a stated arc, prop placement.

Open — attributes free to explore without constraint: backdrop concept, mood, secondary color accents, supporting characters or scenery.

This sorting step takes minutes but saves entire review cycles. Once a team agrees which bucket an element belongs to, every subsequent draft can be judged against that agreement instead of against a vague sense of "does this feel right."

Hypothetical Worked Example

Consider a hypothetical case: a small team is drafting six social ad variations for a reusable water bottle. Before generating anything, they might sort attributes like this:

Fixed: bottle silhouette, cap color, embossed logo position

Bounded: background setting (outdoor only, daylight only), camera angle (front-facing to 30-degree turn)

Open: seasonal props, background color palette, supporting text style

With that grid in hand, a designer generating object-reference drafts can quickly flag a variation where the cap color shifted slightly warmer, or where the bottle appears at a 60-degree angle outside the agreed bound — without needing a full brand review to catch it. This is illustrative reasoning only, not a documented outcome from any specific project.

A Reusable Decision Grid

Before running a batch of drafts, fill in a version of this table for the asset in question: Element Category Allowed Range Who Approves Changes Core product shape Fixed None Brand lead Logo and wordmark Fixed None Brand lead Primary color values Fixed None Brand lead Camera angle Bounded Define degrees or descriptors Creative director Lighting mood Bounded Define time-of-day range Creative director Background setting Open Any, within category Designer Supporting props Open Any, within category Designer

Keeping this grid attached to the project brief — rather than in someone's head — means new team members or freelancers can evaluate drafts consistently, even without sitting in the original planning conversation.

Where a Lightweight Tool Fits Into This Process

Once the fixed-versus-variable decisions are made, the actual generation step becomes faster if the tool supports quick object-reference drafts without heavy setup. Nano Banana 2 Lite is one such option: a third-party, independently operated site (not an official Google or DeepMind product) built for rapid prompt-based image creation and object-reference workflows, useful for early advertising concepts, product mockups, storyboards, and social graphics. The decision grid above is what should be checked against the output — the tool itself doesn't know which attributes matter to a brand.

Where This Approach Breaks Down

The fixed/bounded/open framework has real limits. It assumes someone can articulate the brand constraints clearly enough to write them into a table, which is harder for younger brands still discovering their visual identity. It also assumes a human reviewer checks each draft against the grid — no current object-reference tool can reliably self-verify that a logo's exact pixel color matched across ten variations. Reviewers should still zoom in on fixed elements manually rather than trusting a quick visual scan, since drift in small details like embossing or trim color is easy to miss on a thumbnail. Finally, for genuinely experimental creative work, imposing three categories too early can flatten interesting directions before they've had a chance to surface.

A Practical Starting Point

The underlying discipline — deciding what must hold steady before asking a tool to generate variations — applies regardless of which platform does the generating. A five-minute sorting exercise, documented in a shared grid, turns a chaotic batch of drafts into a reviewable set, and turns "does this look right" into a question with an actual answer.