Merge pull request #79 from noskill/flux1

fix attention kwargs for flux
singnet · Aug 27, 2024 · 53af130 · 53af130
2 parents 01e3b03 + 8c86566
commit 53af130
Showing 1 changed file with 7 additions and 3 deletions.
diff --git a/multigen/pipes.py b/multigen/pipes.py
@@ -197,9 +197,13 @@ def try_set_scheduler(self, inputs):
 
     def load_lora(self, path, multiplier=1.0):
         self.pipe.load_lora_weights(path)
-        if 'cross_attention_kwargs' not in self.pipe_params:
-            self.pipe_params['cross_attention_kwargs'] = {}
-        self.pipe_params['cross_attention_kwargs']["scale"] = multiplier
+        if self.model_type == ModelType.FLUX:
+            attention_kw = 'joint_attention_kwargs'
+        else:
+            attention_kw = 'cross_attention_kwargs'
+        if attention_kw not in self.pipe_params:
+            self.pipe_params[attention_kw] = {}
+        self.pipe_params[attention_kw]["scale"] = multiplier
         self._loras.append(path)
 
     def add_hypernet(self, path, multiplier=None):