Fix Flux2 LoRA guidance conversion

Author: yadferhad

src/diffusers/loaders/lora_conversion_utils.py

@@ -2421,6 +2421,8 @@ def _convert_non_diffusers_flux2_lora_to_diffusers(state_dict):
         "txt_in": "context_embedder",
         "time_in.in_layer": "time_guidance_embed.timestep_embedder.linear_1",
         "time_in.out_layer": "time_guidance_embed.timestep_embedder.linear_2",
+        "guidance_in.in_layer": "time_guidance_embed.guidance_embedder.linear_1",
+        "guidance_in.out_layer": "time_guidance_embed.guidance_embedder.linear_2",
         "final_layer.linear": "proj_out",
         "final_layer.adaLN_modulation.1": "norm_out.linear",
         "single_stream_modulation.lin": "single_stream_modulation.linear",

src/diffusers/loaders/lora_pipeline.py

@@ -5703,6 +5703,13 @@ def load_lora_weights(
         """
         See [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for more details.
         """
+        transformer = getattr(self, self.transformer_name, None)
+        if transformer is None:
+            raise ValueError(
+                "Flux2 LoRA weights can only be loaded into a pipeline that defines a `transformer` component. "
+                "This modular sub-pipeline only exposes a subset of components."
+            )
+
         if not USE_PEFT_BACKEND:
             raise ValueError("PEFT backend is required for this method.")
 
@@ -5726,7 +5733,7 @@ def load_lora_weights(
 
         self.load_lora_into_transformer(
             state_dict,
-            transformer=getattr(self, self.transformer_name) if not hasattr(self, "transformer") else self.transformer,
+            transformer=transformer,
             adapter_name=adapter_name,
             metadata=metadata,
             _pipeline=self,

tests/lora/test_lora_conversion_utils_flux2.py

@@ -0,0 +1,64 @@
+# coding=utf-8
+# Copyright 2025 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import unittest
+
+import torch
+
+from diffusers.loaders.lora_conversion_utils import _convert_non_diffusers_flux2_lora_to_diffusers
+from diffusers.modular_pipelines.flux2.encoders import Flux2KleinTextEncoderStep
+
+
+class Flux2LoraConversionTests(unittest.TestCase):
+    def test_convert_non_diffusers_flux2_lora_maps_guidance_embedder(self):
+        state_dict = {
+            "diffusion_model.img_in.lora_A.weight": torch.randn(2, 2),
+            "diffusion_model.img_in.lora_B.weight": torch.randn(2, 2),
+            "diffusion_model.txt_in.lora_A.weight": torch.randn(2, 2),
+            "diffusion_model.txt_in.lora_B.weight": torch.randn(2, 2),
+            "diffusion_model.time_in.in_layer.lora_A.weight": torch.randn(2, 2),
+            "diffusion_model.time_in.in_layer.lora_B.weight": torch.randn(2, 2),
+            "diffusion_model.time_in.out_layer.lora_A.weight": torch.randn(2, 2),
+            "diffusion_model.time_in.out_layer.lora_B.weight": torch.randn(2, 2),
+            "diffusion_model.guidance_in.in_layer.lora_A.weight": torch.randn(2, 2),
+            "diffusion_model.guidance_in.in_layer.lora_B.weight": torch.randn(2, 2),
+            "diffusion_model.guidance_in.out_layer.lora_A.weight": torch.randn(2, 2),
+            "diffusion_model.guidance_in.out_layer.lora_B.weight": torch.randn(2, 2),
+        }
+
+        converted_state_dict = _convert_non_diffusers_flux2_lora_to_diffusers(state_dict)
+
+        expected_keys = {
+            "transformer.x_embedder.lora_A.weight",
+            "transformer.x_embedder.lora_B.weight",
+            "transformer.context_embedder.lora_A.weight",
+            "transformer.context_embedder.lora_B.weight",
+            "transformer.time_guidance_embed.timestep_embedder.linear_1.lora_A.weight",
+            "transformer.time_guidance_embed.timestep_embedder.linear_1.lora_B.weight",
+            "transformer.time_guidance_embed.timestep_embedder.linear_2.lora_A.weight",
+            "transformer.time_guidance_embed.timestep_embedder.linear_2.lora_B.weight",
+            "transformer.time_guidance_embed.guidance_embedder.linear_1.lora_A.weight",
+            "transformer.time_guidance_embed.guidance_embedder.linear_1.lora_B.weight",
+            "transformer.time_guidance_embed.guidance_embedder.linear_2.lora_A.weight",
+            "transformer.time_guidance_embed.guidance_embedder.linear_2.lora_B.weight",
+        }
+
+        self.assertEqual(set(converted_state_dict.keys()), expected_keys)
+
+    def test_flux2_text_subpipeline_rejects_transformer_lora_loading(self):
+        text_pipe = Flux2KleinTextEncoderStep().init_pipeline()
+
+        with self.assertRaisesRegex(ValueError, "defines a `transformer` component"):
+            text_pipe.load_lora_weights({})