Generation, costs and failures
Review each frame, approve the quote and generate one image at a time.
Before generation
Check role, specific content, product asset, copy, model and size for every frame. Select corresponding product assets for details or packaging when appropriate. The primary product image is the default.
Each frame is quoted using its actual model and size, then summed. A failed quote prevents paid generation. The current server quote determines the cost; package credits are not a fixed constant.
Entering planning, applying a template or sending planning messages does not directly charge image-generation credits. Start only with explicit generation confirmation.
One image per request
Combinations submit frames sequentially, one image per request. Each has shared visual direction plus its own content, copy and restrictions. Grouping a set does not create hidden first-result reference dependencies.
The first result does not automatically replace the product reference. A failed product-reference input does not silently become reference-free generation.
Progress, pause and refresh
Overview and nodes display pending, running, completed or failed states. Pause stops later submissions; an already submitted image still finishes. Check remaining plans and pricing before resuming.
Refresh continues waiting for existing submitted jobs. Repeated clicks should not resubmit them. For network uncertainty, inspect the original job before creating new paid attempts.
Failed frames and insufficient balance
A failed frame preserves other results and is not automatically retried with payment. Other frames may continue after an ordinary failure. Insufficient balance stops further submissions.
Explicit retry or regeneration creates a new attempt and may incur a new cost. Review the error and quote first. For persistent errors, send feedback with task details, never API keys.
Review the output
Check product identity, structure, colors, copy and confirmed specifications. Actual output pixel dimensions can differ from the requested size. Check downloaded files against your platform's requirements before publishing.
Changing a plan does not automatically overwrite completed images. Regenerate only the selected frames that need changes. See Canvas.
Saved requirements and prompt preparation
Expand Requirements and prompt preparation in the task overview. Choose the current product source, operation, edit target, preserved details, composition, background and shadows. These requirements are saved with the project and survive conversation proposals. Changing identity clears old per-frame product selections. Existing successful results are not automatically regenerated.
For an edit, upload the selected previous result as a project reference, then select it as the base. The product source establishes identity, the base supplies the prior composition, and style/layout references supply only their assigned guidance. Inputs are sent in that order. Models without multi-image support are rejected before quoting.
Prepare and preview prompt (free) shows the execution instructions and an editable frame prompt. Missing edit targets/bases, identical editable/preserved objects and explicit shadow conflicts block quoting. Checks cannot detect every semantic contradiction; review the instructions yourself. Changes invalidate the previous quote.
Current channels redraw the full image. Preserved pixels are not guaranteed. Masked editing, automatic output inspection and automatic paid repairs are not provided.