Wan2.7 image generation and editing: references, sequences and output limits

Confirm the model and task

Alibaba Cloud's Chinese documentation lists wan2.7-image and wan2.7-image-pro for text-to-image, sequences, editing and multiple references. It distinguishes 4K single-image text generation on pro from the 2K maximum for editing and sequences. Do not apply one task's maximum to every operation.

HUUUIAI's Wan2.7 label is a product entry. Check its live controls for the actual integration. Regional documentation can describe different versions, so verify the title, model and region together.

Reference order and selection geometry

The API documentation uses ordered reference images and original-image pixel coordinates for bounding boxes. Developers should not substitute coordinates measured on a zoomed canvas. Users should express selections through the controls actually provided by the product.

Assign references to identity, setting, material or style. Inspect perspective and contact shadows after combining them.

Product-scene prompt

Place the ceramic vase from image 1 on the window-side table in image 2. Preserve its silhouette, glaze and decoration without copying image 1's background. Match image 2's natural side light and perspective, add a contact shadow and no text.

Review sequences and preserve files

The documented sequence count is a maximum, not a guaranteed number of results. A fixed seed can improve stability without promising identical pixels. Returned resource links expire, so save the final files promptly rather than treating temporary URLs as permanent storage.