diff --git a/tensorflow_probability/python/internal/backend/numpy/random_generators.py b/tensorflow_probability/python/internal/backend/numpy/random_generators.py
index b14238caa1..004e2bf6f9 100644
--- a/tensorflow_probability/python/internal/backend/numpy/random_generators.py
+++ b/tensorflow_probability/python/internal/backend/numpy/random_generators.py
@@ -66,9 +66,18 @@ def _bcast_shape(base_shape, args):
   return bcast_shape
 
 
+def _rng_from_seed(seed):
+  if seed is None:
+    return np.random
+  elif isinstance(seed, int):
+    return np.random.RandomState(seed & 0xFFFFFFFF)
+  else:
+    return np.random.RandomState(np.array(seed, dtype=np.uint32))
+
+
 def _binomial(shape, seed, counts, probs, output_dtype=np.int32, name=None):  # pylint: disable=unused-argument
   """Massaging dtype and nan handling of np.random.binomial."""
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   invalid_count = (np.int64(counts) < 0) != (counts < 0)
   if np.any(invalid_count):
     raise ValueError('int64 overflow: {} -> {}'.format(
@@ -82,7 +91,7 @@ def _binomial(shape, seed, counts, probs, output_dtype=np.int32, name=None):  #
 
 
 def _categorical(logits, num_samples, dtype=None, seed=None, name=None):  # pylint: disable=unused-argument
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   dtype = utils.numpy_dtype(dtype or np.int64)
   if not hasattr(logits, 'shape'):
     logits = np.array(logits, np.float32)
@@ -107,7 +116,7 @@ def _categorical_jax(logits, num_samples, dtype=None, seed=None, name=None):  #
 
 def _gamma(shape, alpha, beta=None, dtype=np.float32, seed=None,
            name=None):  # pylint: disable=unused-argument
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   scale = 1. if beta is None else (1. / beta)
   shape = _ensure_shape_tuple(shape)
   return rng.gamma(shape=alpha, scale=scale, size=shape).astype(dtype)
@@ -133,7 +142,7 @@ def _gamma_jax(shape, alpha, beta=None, dtype=np.float32, seed=None, name=None):
 
 def _normal(shape, mean=0.0, stddev=1.0, dtype=np.float32, seed=None,
             name=None):  # pylint: disable=unused-argument
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   dtype = utils.common_dtype([mean, stddev], dtype_hint=dtype)
   shape = _bcast_shape(shape, [mean, stddev])
   return rng.normal(loc=mean, scale=stddev, size=shape).astype(dtype)
@@ -151,7 +160,7 @@ def _normal_jax(shape, mean=0.0, stddev=1.0, dtype=np.float32, seed=None,
 
 def _poisson(shape, lam, dtype=np.float32, seed=None,
              name=None):  # pylint: disable=unused-argument
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   dtype = utils.common_dtype([lam], dtype_hint=dtype)
   shape = _ensure_shape_tuple(shape)
   return rng.poisson(lam=lam, size=shape).astype(dtype)
@@ -209,7 +218,7 @@ def _poisson_jax(shape, lam, dtype=np.float32, seed=None,
 
 
 def _shuffle(value, seed=None, name=None):  # pylint: disable=unused-argument
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   ret = np.array(value)
   rng.shuffle(ret)
   return ret
@@ -225,7 +234,7 @@ def _shuffle_jax(value, seed=None, name=None):  # pylint: disable=unused-argumen
 def _truncated_normal(
     shape, seed, means=0.0, stddevs=1.0, minvals=-2.0, maxvals=2.0, name=None):  # pylint: disable=unused-argument
   from scipy import stats  # pylint: disable=g-import-not-at-top
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   std_low = (minvals - means) / stddevs
   std_high = (maxvals - means) / stddevs
   std_samps = stats.truncnorm.rvs(
@@ -248,7 +257,7 @@ def _truncated_normal_jax(
 def _uniform(shape, minval=0, maxval=None, dtype=np.float32, seed=None,
              name=None):  # pylint: disable=unused-argument
   """Numpy uniform random sampler."""
-  rng = np.random if seed is None else np.random.RandomState(seed & 0xffffffff)
+  rng = _rng_from_seed(seed)
   if minval is not None:
     minval = ops.convert_to_tensor(minval, dtype=dtype)
   if maxval is not None: