The problem of spreading secret data to embed into multiple cover images is called batch steganography and has been theoretically considered recently. Few works have been done in batch steganography, and in all of them, the payload is spread between cover images unwisely. We present the Adaptive batch steganography (ABS) approach and consider embedding capacity as a property of images. ABS is an approach to adaptively spread secret data among multiple cover images based on their embedding capacity. By splitting the payload based on image embedding capacity constraint, embedding can be done more secure than the state when the embedder does not know how much data can be hidden securely in an image. Furthermore, the number of required cover images to embed a piece of secret data in ABS is smaller than the number of required cover images in usual batch steganography. We apply an ensemble system that uses different steganalyzer units to determine the embedding capacity of a cover image. Each steganalyzer unit is formed by a combination of multiple steganalyzers from a same type, but each one trained to detect a certain payload. Experimental results showed the effectiveness of embedding in ABS and security enhancement of produced stego images.