Analysis, questions and prompts
Separate visible observations from confirmed facts and plan distinct images from one brief.
When analysis happens
After preparing assets, roles and requirements, enter canvas planning. One preliminary AI call jointly examines available product and reference images, proposes product observations, a shared visual direction, individual frame plans and up to three important questions.
Follow-up answers and conversation normally reuse this analysis instead of recognizing or translating every frame again. Review the plan after asset changes or mistaken observations. Analysis errors can be retried without starting image generation.
Visible does not mean confirmed
Appearance, visible texture, shape and color may be observations. Dimensions, capacity, material composition, performance, compatible models and package contents cannot become known facts from appearance alone.
“Looks metallic” does not confirm stainless steel. One vase in a photo does not establish the complete package contents. Correct mistaken observations and provide reliable facts.
Necessary questions
Up to three questions may request a choice, text, yes/no/unknown, or a replacement frame. If none are needed, go directly to plan review.
Unknown is not permission to invent. Supply missing dimensions or packaging facts, or replace that role with an evidence-based detail or use image while retaining the set count. Resolve critical questions before generation.
One brief, different images
The brief describes product, audience, style, priorities and exclusions. Planning combines it with observations, confirmed facts, template direction and selected image types. The set shares product identity, color direction and typography; each frame has its own objective, content, composition, copy and constraints.
Main-image rules depend on platform and category. Scenario images allow an environment; close-ups emphasize details. Repeated types also need distinct content. References contribute to joint analysis and written planning, but accurate automatic per-reference classification and matching is not guaranteed. Check asset purposes.
AI writing and language
AI writing refines the overall brief in the interface language. Image-copy language follows the target language and can differ from the interface.
Execution instructions default to English; this is not a guarantee of better results. Approved copy, brands, models and numbers remain literal. Punctuation in a creative brief does not automatically turn it into image captions.
For example:
Target home decor shoppers with a warm, clean style. Show visible texture and tabletop use. Use English copy. Add no accessories or unconfirmed capacity or material claims.