fix seeding maybe

hydra_node/models.py

@@ -17,6 +17,7 @@ import time
 from PIL import Image
 import k_diffusion as K
 import contextlib
+import random
 
 def pil_upscale(image, scale=1):
     device = image.device
@@ -162,6 +163,11 @@ class StableInterface(nn.Module):
 
         return x_0
 
+def seed_everything(seed: int):
+    torch.manual_seed(seed)
+    np.random.seed(seed)
+    random.seed(seed)
+
 class StableDiffusionModel(nn.Module):
     def __init__(self, config):
         nn.Module.__init__(self)
@@ -282,8 +288,7 @@ class StableDiffusionModel(nn.Module):
                 CrossAttention.set_hypernetwork(module)
 
         if request.seed is not None:
-            torch.manual_seed(request.seed)
-            np.random.seed(request.seed)
+            seed_everything(request.seed)
 
         if request.image is not None:
             request.steps = 50
@@ -302,7 +307,7 @@ class StableDiffusionModel(nn.Module):
             start_noise = []
             for seed in range(request.seed, request.seed+request.n_samples):
                 main_noise.append(sample_start_noise(seed, request.latent_channels, request.height, request.width, request.downsampling_factor, self.device))
-                start_noise.append(sample_start_noise(None, request.latent_channels, request.height, request.width, request.downsampling_factor, self.device))
+                start_noise.append(sample_start_noise(seed, request.latent_channels, request.height, request.width, request.downsampling_factor, self.device))
 
             main_noise = torch.cat(main_noise, dim=0)
             start_noise = torch.cat(start_noise, dim=0)
@@ -424,8 +429,7 @@ class StableDiffusionModel(nn.Module):
     def sample_two_stages(self, request):
         request = DotMap(request)
         if request.seed is not None:
-            torch.manual_seed(request.seed)
-            np.random.seed(request.seed)
+            seed_everything(request.seed)
 
         if request.plms:
             sampler = self.plms