Separate brand settings from product evidence
A brand palette cannot establish a bottle's volume, a garment's seam or which campaign claim was approved. When a new conversation produces an attractive but wrong variation, check what information was actually provided in that conversation. Do not assume that a previous chat's reference images or corrections are available elsewhere.
Kittl documents current-chat memory and says each new Creative Agent chat begins fresh. Applying a Brand supports consistency across chats. Earlier chats can be revisited through History. These are documented context boundaries, not evidence that a generated image will always preserve every supplied detail.
Prepare a compact continuation brief
For a fictional reusable-bottle campaign, carry forward the approved reference photograph, the exact variant name, the desired placement and a list of changes already rejected. State which photograph establishes the actual cap, logo and proportions. A mood image should be labeled as style inspiration rather than the identity reference.
Use a short block with four fields: product identity source, approved creative direction, permitted changes and forbidden changes. Add the chosen prior output only when it is useful and you have the right to use it. This is a human-authored handoff practice, not a special import format or guaranteed memory mechanism.
Choose continuation or a deliberate new direction
If the next request is a small refinement to the same campaign, first locate the relevant existing conversation. If the new job is for a different product or a deliberately separate direction, create a fresh brief instead of relying on vague instructions such as make it like before. Keep a clear campaign identifier outside the assistant conversation.
Before sending, inspect which Brand is active and which references are attached. Brand direction and SKU identity serve different purposes. A correctly applied visual style can coexist with a product error, so review both independently after generation.
Test whether the handoff is sufficient
Ask a colleague to read the continuation brief without the original chat. Could they identify the right physical variant, explain the permitted edit and reject a known wrong output? If not, add the missing evidence before authorizing more generation. This check evaluates the brief; it is not a completed model test.
Review the resulting asset against the source product, not just against the preceding generated image. A sequence of plausible copies can carry an earlier error forward. Stop and repair the reference chain when the real product cannot be identified unambiguously.
This guide concerns cross-chat context and creative handoff. It does not claim that CartVisual imports Kittl conversations, that a Brand stores all campaign decisions or that any memory feature replaces product verification.
Related production decisions
Continue with the adjacent checks when they apply: Is a photo saved from a messaging app good enough for AI product editing?. Why does my duplicated product image stay unchanged after Kittl finishes its AI area edit?.