Combine multiple reference images with AI

A multi-image request is not simply a larger upload. Each source should answer a different visual question. One image can define the subject, another the material or wardrobe, and a third the location or composition. This multi-reference image AI workflow becomes harder to control when two references compete for the same role and the selected model has to infer which evidence should win.

Imagine2 caps the current workflow at three references. The smaller set keeps roles clear, uploads manageable, and the credit estimate easy to review before generation.

Refer to inputs by order and purpose: image one supplies the person, image two supplies the jacket, and image three supplies the environment. Then describe which traits from each source matter. For identity, name facial features and pose. For products, preserve shape and label placement. For style, specify whether color, texture, line quality, or lighting should transfer.

Do not ask the model to preserve every source completely. Combining images necessarily resolves differences in viewpoint, scale, light, and perspective. State the priority when evidence conflicts, such as preserving the product geometry while adapting reflections to the new environment.

Use image 1 for the person and preserve facial identity. Use image 2 for the dark green technical jacket, including its collar and pocket construction. Place the subject in the rainy station from image 3, matching the station light naturally while keeping the original waist-up camera crop.

Subject plus style works when identity comes from one source and the rendering approach comes from another. Product plus scene is useful when you combine images with AI for a campaign concept where the object must remain recognizable inside a new environment. Character plus environment plus prop gives each of the three available references a distinct responsibility.

Choose sources with compatible viewpoints when possible. A front-facing portrait, rear-view garment, and extreme overhead environment create avoidable ambiguity. Crop out interface chrome and unrelated subjects. Prefer clean, rights-cleared images that make the desired evidence easy to identify.

Each reference adds one credit to the request. One 1K output from three references costs eight credits; one 2K output costs ten. The estimate updates before generation, and failed outputs are refunded without repeatedly issuing credits when an error handler retries.

References and outputs are stored privately. Unsaved tasks expire after 24 hours, saved tasks after 30 days, and Delete removes associated objects immediately. Imagine2 never publishes a user task to Gallery by default. You remain responsible for rights, consent, trademarks, likenesses, and appropriate disclosure.