keras-team/keras · error · ValueError

Inputs have incompatible shapes. Received shapes {shape1} an

Error message

Inputs have incompatible shapes. Received shapes {shape1} and {shape2}

What it means

Merge layers (Add, Multiply, Average, Maximum, etc.) broadcast their inputs and require non-batch dimensions to be either equal or 1. This error fires during build/compute_output_shape when two inputs disagree on a dimension and neither is 1, so broadcasting cannot resolve them.

Source

Thrown at keras/src/layers/merging/base_merge.py:93

        """

        if None in [shape1, shape2]:
            return None
        elif len(shape1) < len(shape2):
            return self._compute_elemwise_op_output_shape(shape2, shape1)
        elif not shape2:
            return shape1
        output_shape = list(shape1[: -len(shape2)])
        for i, j in zip(shape1[-len(shape2) :], shape2):
            if i is None or j is None:
                output_shape.append(None)
            elif i == 1:
                output_shape.append(j)
            elif j == 1:
                output_shape.append(i)
            else:
                if i != j:
                    raise ValueError(
                        "Inputs have incompatible shapes. "
                        f"Received shapes {shape1} and {shape2}"
                    )
                output_shape.append(i)
        return tuple(output_shape)

    def build(self, input_shape):
        # Used purely for shape validation.
        if not isinstance(input_shape[0], (tuple, list)):
            raise ValueError(
                "A merge layer should be called on a list of inputs. "
                f"Received: input_shape={input_shape} (not a list of shapes)"
            )
        if len(input_shape) < 1:
            raise ValueError(
                "A merge layer should be called "
                "on a list of at least 1 input. "
                f"Received {len(input_shape)} inputs. "

View on GitHub (pinned to 7a34a03db6)

Solutions

1. Fix the upstream layers so both branches produce the same non-batch shape
2. Insert a Dense/Conv projection or Reshape on one branch to align dimensions before the merge
3. Use 1 for a dimension to exploit broadcasting intentionally
4. If shapes are dynamic (None), confirm the runtime shapes actually match

Example fix

# before
out = layers.Add()([enc, dec])  # (None,64) + (None,128) -> ValueError

# after
dec = layers.Dense(64)(dec)
out = layers.Add()([enc, dec])

def broadcastable(s1, s2):
    return len(s1) == len(s2) and all(a == b or a == 1 or b == 1 or a is None or b is None for a, b in zip(s1, s2))
assert broadcastable(tuple(x.shape), tuple(y.shape))

def shapes_broadcastable(s1, s2) -> bool:
    return len(s1) == len(s2) and all(a == b or a == 1 or b == 1 or a is None or b is None for a, b in zip(s1, s2))

When it happens

Trigger: Calling Add()([x, y]) where x has shape (None, 32) and y has shape (None, 16); Multiply on feature maps with different channel counts; shape inference on inputs with conflicting dims.

Common situations: Feeding embeddings of different dimensions into an Add; forgetting a projection/reshape layer before merging encoder and decoder branches; off-by-one pooling that changes a spatial dim on one branch only.

Related errors

• Cannot merge tensors with different batch sizes. Received te
• Architecture configuration does not match {weights_name} var
• Model name "{name}" does not match weights variant "{weights
• DenseNet does not support the `channels_first` image data fo
• The last dimension of `query_shape` and `value_shape` must b

AI-assisted analysis of keras-team/keras@7a34a03db6 (2026-08-25). Data as JSON: /api/errors/30dfa63592a6fda8. Report an issue: GitHub.