diff --git a/NAMESPACE b/NAMESPACE
index bc1c9d8c6..c3e82a55f 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -201,6 +201,8 @@ export(image_from_array)
 export(image_load)
 export(image_smart_resize)
 export(image_to_array)
+export(imagenet_decode_predictions)
+export(imagenet_preprocess_input)
 export(initializer_constant)
 export(initializer_glorot_normal)
 export(initializer_glorot_uniform)
diff --git a/NEWS.md b/NEWS.md
index 01873b8ce..fb4a15c91 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -60,6 +60,9 @@ New functions:
 
 - `layer_lstm()` and `layer_gru()` gain arg `use_cudnn`, default `'auto'`.
 
+- Fixed an issue where `application_preprocess_inputs()` would error if supplied
+  an R array as input.
+
 - Doc improvements.
 
 # keras3 0.1.0
diff --git a/R/applications.R b/R/applications.R
index 2b91c1da1..31e3fc5b3 100644
--- a/R/applications.R
+++ b/R/applications.R
@@ -3953,7 +3953,88 @@ list_model_names <- function() {
 }
 
 set_preprocessing_attributes <- function(object, module) {
-  attr(object, "preprocess_input") <- module$preprocess_input
+  .preprocess_input <- r_to_py(module)$preprocess_input
+
+  attr(object, "preprocess_input") <-
+    as.function.default(c(formals(.preprocess_input), bquote({
+      args <- capture_args(list(
+        x = function(x) {
+          if (!is_py_object(x))
+            x <- np_array(x)
+          if (inherits(x, "numpy.ndarray") &&
+              !py_bool(x$flags$writeable))
+            x <- x$copy()
+          x
+        }
+      ))
+      do.call(.(.preprocess_input), args)
+    })), envir = parent.env(environment()))
+
   attr(object, "decode_predictions") <- module$decode_predictions
   object
 }
+
+
+#' Decodes the prediction of an ImageNet model.
+#'
+#' @param preds Tensor encoding a batch of predictions.
+#' @param top integer, how many top-guesses to return.
+#'
+#' @return List of data frames with variables `class_name`, `class_description`,
+#'   and `score` (one data frame per sample in batch input).
+#'
+#' @export
+#' @keywords internal
+imagenet_decode_predictions <- function(preds, top = 5) {
+
+  # decode predictions
+  decoded <- keras$applications$imagenet_utils$decode_predictions(
+    preds = preds,
+    top = as.integer(top)
+  )
+
+  # convert to a list of data frames
+  lapply(decoded, function(x) {
+    m <- t(sapply(1:length(x), function(n) x[[n]]))
+    data.frame(class_name = as.character(m[,1]),
+               class_description = as.character(m[,2]),
+               score = as.numeric(m[,3]),
+               stringsAsFactors = FALSE)
+  })
+}
+
+
+#' Preprocesses a tensor or array encoding a batch of images.
+#'
+#' @param x Input Numpy or symbolic tensor, 3D or 4D.
+#' @param data_format Data format of the image tensor/array.
+#' @param mode One of "caffe", "tf", or "torch"
+#'   - caffe: will convert the images from RGB to BGR,
+#'     then will zero-center each color channel with
+#'     respect to the ImageNet dataset,
+#'     without scaling.
+#'   - tf: will scale pixels between -1 and 1, sample-wise.
+#'   - torch: will scale pixels between 0 and 1 and then
+#'     will normalize each channel with respect to the
+#'     ImageNet dataset.
+#'
+#' @return Preprocessed tensor or array.
+#'
+#' @export
+#' @keywords internal
+imagenet_preprocess_input <- function(x, data_format = NULL, mode = "caffe") {
+  args <- capture_args(list(
+    x = function(x) {
+      if (!is_py_object(x))
+        x <- np_array(x)
+      if (inherits(x, "numpy.ndarray") &&
+          !py_bool(x$flags$writeable))
+        x <- x$copy()
+      x
+    }
+  ))
+
+  preprocess_input <- r_to_py(keras$applications$imagenet_utils)$preprocess_input
+  do.call(preprocess_input, args)
+}
+
diff --git a/man/imagenet_decode_predictions.Rd b/man/imagenet_decode_predictions.Rd
new file mode 100644
index 000000000..793d417c3
--- /dev/null
+++ b/man/imagenet_decode_predictions.Rd
@@ -0,0 +1,21 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/applications.R
+\name{imagenet_decode_predictions}
+\alias{imagenet_decode_predictions}
+\title{Decodes the prediction of an ImageNet model.}
+\usage{
+imagenet_decode_predictions(preds, top = 5)
+}
+\arguments{
+\item{preds}{Tensor encoding a batch of predictions.}
+
+\item{top}{integer, how many top-guesses to return.}
+}
+\value{
+List of data frames with variables \code{class_name}, \code{class_description},
+and \code{score} (one data frame per sample in batch input).
+}
+\description{
+Decodes the prediction of an ImageNet model.
+}
+\keyword{internal}
diff --git a/man/imagenet_preprocess_input.Rd b/man/imagenet_preprocess_input.Rd
new file mode 100644
index 000000000..c9c163b70
--- /dev/null
+++ b/man/imagenet_preprocess_input.Rd
@@ -0,0 +1,32 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/applications.R
+\name{imagenet_preprocess_input}
+\alias{imagenet_preprocess_input}
+\title{Preprocesses a tensor or array encoding a batch of images.}
+\usage{
+imagenet_preprocess_input(x, data_format = NULL, mode = "caffe")
+}
+\arguments{
+\item{x}{Input Numpy or symbolic tensor, 3D or 4D.}
+
+\item{data_format}{Data format of the image tensor/array.}
+
+\item{mode}{One of "caffe", "tf", or "torch"
+\itemize{
+\item caffe: will convert the images from RGB to BGR,
+then will zero-center each color channel with
+respect to the ImageNet dataset,
+without scaling.
+\item tf: will scale pixels between -1 and 1, sample-wise.
+\item torch: will scale pixels between 0 and 1 and then
+will normalize each channel with respect to the
+ImageNet dataset.
+}}
+}
+\value{
+Preprocessed tensor or array.
+}
+\description{
+Preprocesses a tensor or array encoding a batch of images.
+}
+\keyword{internal}