diff --git a/daft.py b/daft.py
index dd27d07..bf71692 100644
--- a/daft.py
+++ b/daft.py
@@ -237,7 +237,6 @@ def render(self, ctx):
         l["ha"] = _pop_multiple(l, "center", "ha", "horizontalalignment")
 
         # Deal with ``fixed`` nodes.
-        scale = self.scale
         if self.fixed:
             # MAGIC: These magic numbers should depend on the grid/node units.
             self.offset[1] += 6
@@ -249,7 +248,7 @@ def render(self, ctx):
             if p["fc"] == "none":
                 p["fc"] = "k"
 
-        diameter = ctx.node_unit * scale
+        diameter = ctx.node_unit * self.scale
         if self.aspect is not None:
             aspect = self.aspect
         else:
diff --git a/examples/brewer.py b/examples/brewer.py
new file mode 100644
index 0000000..9e3a0d3
--- /dev/null
+++ b/examples/brewer.py
@@ -0,0 +1,30 @@
+"""
+Brewer's project, sort of
+=========================
+
+This uses a HACK of `Plate` to put a label on the plot.
+"""
+
+from matplotlib import rc
+rc("font", family="serif", size=12)
+rc("text", usetex=True)
+
+import daft
+
+pgm = daft.PGM([5., 2.5], origin=[2.0, 0.0], aspect=2.1)
+pgm.add_node(daft.Node("counts", r"$dN/dm$", 6.0, 2.0))
+pgm.add_node(daft.Node("stars", r"stars", 4.5, 2.0,))
+pgm.add_node(daft.Node("pixels", r"pixels", 3.0, 2.0, observed=True))
+pgm.add_node(daft.Node("psf", r"psf", 4.5, 1.0))
+pgm.add_node(daft.Node("noise", r"noise", 3.0, 1.0))
+pgm.add_edge("stars", "pixels")
+pgm.add_edge("psf", "pixels")
+pgm.add_edge("noise", "pixels")
+pgm.add_edge("counts", "stars")
+pgm.add_plate(daft.Plate([2.0, 0.0, 7.0, 2.5],
+                         label=r"$p(\mbox{pixels}\,|\,\mbox{stars})\,p(\mbox{stars}\,|\,dN/dm)\,p(dN/dm)$",
+                         label_offset=[30,10],
+                         rect_params={"ec": "none",}))
+pgm.render()
+pgm.figure.savefig("brewer.pdf")
+pgm.figure.savefig("brewer.png", dpi=300)
diff --git a/examples/hayabusa.py b/examples/hayabusa.py
new file mode 100644
index 0000000..b0c8261
--- /dev/null
+++ b/examples/hayabusa.py
@@ -0,0 +1,48 @@
+"""
+Hayabusa
+========
+
+No comment.
+
+"""
+
+from matplotlib import rc
+rc("font", family="serif", size=12)
+rc("text", usetex=True)
+import daft
+
+pgm = daft.PGM([4.8, 3.2], origin=[-0.4, -0.7])
+
+# Hierarchical parameters.
+pgm.add_node(daft.Node("theta", r"$\theta$", 0, 1))
+pgm.add_node(daft.Node("Omega", r"$\Omega$", 4, 1))
+pgm.add_node(daft.Node("Sigma", r"$\Sigma$", 0, 0))
+
+# spectral measurements
+pgm.add_node(daft.Node("a", r"$a_n$", 1, 2, observed=True))
+pgm.add_node(daft.Node("z", r"$z_n$", 2, 2, observed=True))
+
+# Latent variables.
+pgm.add_node(daft.Node("M", r"$M_n$", 1, 1))
+pgm.add_edge("a", "M")
+pgm.add_edge("theta", "M")
+pgm.add_node(daft.Node("K", r"$K_n$", 2, 1))
+pgm.add_edge("z", "K")
+pgm.add_node(daft.Node("DM", r"$DM_n$", 3, 1))
+pgm.add_edge("z", "DM")
+pgm.add_edge("Omega", "DM")
+
+# Data.
+pgm.add_node(daft.Node("m", r"$m_n$", 2, 0, observed=True))
+pgm.add_edge("M", "m")
+pgm.add_edge("K", "m")
+pgm.add_edge("DM", "m")
+pgm.add_edge("Sigma", "m")
+
+# And a plate.
+pgm.add_plate(daft.Plate([0.5, -0.6, 3, 3], label=r"quasars $n$"))
+
+# Render and save.
+pgm.render()
+pgm.figure.savefig("hayabusa.pdf")
+pgm.figure.savefig("hayabusa.png", dpi=150)
diff --git a/examples/stochastic.py b/examples/stochastic.py
new file mode 100644
index 0000000..4306f61
--- /dev/null
+++ b/examples/stochastic.py
@@ -0,0 +1,42 @@
+"""
+The Quintessential PGM
+======================
+
+This is a demonstration of a very common structure found in graphical models.
+It has been rendered using Daft's default settings for all the parameters
+and it shows off how much beauty is baked in by default.
+
+"""
+
+from matplotlib import rc
+rc("font", family="serif", size=12)
+rc("text", usetex=True)
+
+import daft
+
+# Instantiate the PGM.
+pgm = daft.PGM([5., 3.1], origin=[-0.5, 0.3])
+
+# Hierarchical parameters.
+pgm.add_node(daft.Node("alpha", r"$\alpha$", 0.5, 3., fixed=True))
+pgm.add_node(daft.Node("weather", r"stochastic", 0.5, 2., aspect=2.5))
+pgm.add_node(daft.Node("pop", r"population", 2., 3., aspect=2.5))
+pgm.add_node(daft.Node("science", r"target", 2., 2., aspect=2.5))
+pgm.add_node(daft.Node("data", r"data set", 1.5, 1., aspect=2.5))
+pgm.add_node(daft.Node(r"noise", r"\noindent noise\\ model", 3.7, 1., aspect=2.5))
+
+# Add in the edges.
+pgm.add_edge("alpha", "weather")
+pgm.add_edge("weather", "data")
+pgm.add_edge("pop", "science")
+pgm.add_edge("science", "data")
+pgm.add_edge("noise", "data")
+
+# And a plate.
+pgm.add_plate(daft.Plate([-0.3, 0.5, 3.1, 2.], label=r"data sets",
+    shift=-0.1))
+
+# Render and save.
+pgm.render()
+pgm.figure.savefig("stochastic.pdf")
+pgm.figure.savefig("stochastic.png", dpi=150)
diff --git a/examples/xdqso.py b/examples/xdqso.py
new file mode 100644
index 0000000..7fd6b94
--- /dev/null
+++ b/examples/xdqso.py
@@ -0,0 +1,54 @@
+"""
+Extreme deconvolution of stars
+==============================
+
+The (very simple) model that transformed SDSS-III.
+
+"""
+
+from matplotlib import rc
+rc("font", family="serif", size=12)
+rc("text", usetex=True)
+
+import daft
+
+# Instantiate the PGM.
+pgm = daft.PGM([4.00, 3.55], origin=[-0.6, 0.3])
+
+# Hierarchical parameters.
+pgm.add_node(daft.Node("alpha", r"$\alpha$", 1, 3.5, fixed=True))
+pgm.add_node(daft.Node("theta", r"$\theta$", 1, 2.5))
+pgm.add_node(daft.Node("Sigma", r"$\Sigma$", 1, 1))
+pgm.add_node(daft.Node("sigma", r"$\sigma$", 3, 1))
+
+# Latent variables.
+pgm.add_node(daft.Node("q", r"$q_m$", 2, 3))
+pgm.add_node(daft.Node("X", r"$X_m$", 2, 2))
+pgm.add_node(daft.Node("Xt", r"$X_n$", 0, 2))
+
+# Data.
+pgm.add_node(daft.Node("qt", r"$q_n$", 0, 3, observed=True))
+pgm.add_node(daft.Node("x", r"$x_m$", 2, 1, observed=True))
+pgm.add_node(daft.Node("xt", r"$x_n$", 0, 1, observed=True))
+
+# Add in the edges.
+pgm.add_edge("alpha", "theta")
+pgm.add_edge("theta", "q")
+pgm.add_edge("theta", "qt")
+pgm.add_edge("theta", "X")
+pgm.add_edge("theta", "Xt")
+pgm.add_edge("sigma", "x")
+pgm.add_edge("Sigma", "xt")
+pgm.add_edge("X", "x")
+pgm.add_edge("Xt", "xt")
+pgm.add_edge("q", "X")
+pgm.add_edge("qt", "Xt")
+
+# And plates.
+pgm.add_plate(daft.Plate([-0.5, 0.5, 1, 3], label=r"train", shift=-0.1))
+pgm.add_plate(daft.Plate([1.5, 0.5, 1, 3], label=r"test", shift=-0.1))
+
+# Render and save.
+pgm.render()
+pgm.figure.savefig("xdqso.pdf")
+pgm.figure.savefig("xdqso.png", dpi=150)
diff --git a/examples/xdstars.py b/examples/xdstars.py
new file mode 100644
index 0000000..108aadd
--- /dev/null
+++ b/examples/xdstars.py
@@ -0,0 +1,41 @@
+"""
+Extreme deconvolution of stars
+==============================
+
+The (very simple) model that transformed SDSS-III.
+
+"""
+
+from matplotlib import rc
+rc("font", family="serif", size=12)
+rc("text", usetex=True)
+
+import daft
+
+# Instantiate the PGM.
+pgm = daft.PGM([2.2, 3.55], origin=[-0.6, 0.3])
+
+# Hierarchical parameters.
+pgm.add_node(daft.Node("alpha", r"$\alpha$", 1, 3.5, fixed=True))
+pgm.add_node(daft.Node("theta", r"$\theta$", 1, 2.5))
+pgm.add_node(daft.Node("sigma", r"$\Sigma$", 1, 1))
+
+# Latent variable.
+pgm.add_node(daft.Node("X", r"$X_n$", 0, 2))
+
+# Data.
+pgm.add_node(daft.Node("x", r"$x_n$", 0, 1, observed=True))
+
+# Add in the edges.
+pgm.add_edge("alpha", "theta")
+pgm.add_edge("theta", "X")
+pgm.add_edge("sigma", "x")
+pgm.add_edge("X", "x")
+
+# And a plate.
+pgm.add_plate(daft.Plate([-0.5, 0.5, 1, 2], label=r"stars $n$", shift=-0.1))
+
+# Render and save.
+pgm.render()
+pgm.figure.savefig("xdstars.pdf")
+pgm.figure.savefig("xdstars.png", dpi=150)