diff --git a/ai_diffusion/__init__.py b/ai_diffusion/__init__.py
index 363287a52..99d4a7cd0 100644
--- a/ai_diffusion/__init__.py
+++ b/ai_diffusion/__init__.py
@@ -1,6 +1,6 @@
 """Generative AI plugin for Krita"""
 
-__version__ = "1.26.0"
+__version__ = "1.27.0"
 
 import importlib.util
 
diff --git a/ai_diffusion/client.py b/ai_diffusion/client.py
index 8f4588b0d..395ceee2d 100644
--- a/ai_diffusion/client.py
+++ b/ai_diffusion/client.py
@@ -154,7 +154,7 @@ def __getitem__(self, key: ControlMode | UpscalerName | str):
     def find(self, key: ControlMode | UpscalerName | str, allow_universal=False) -> str | None:
         if key in [ControlMode.style, ControlMode.composition]:
             key = ControlMode.reference  # Same model with different weight types
-        result = self._models.resources.get(resource_id(self.kind, self.arch, key))
+        result = self._models.find(ResourceId(self.kind, self.arch, key))
         if result is None and allow_universal and isinstance(key, ControlMode):
             result = self.find(ControlMode.universal)
         return result
diff --git a/ai_diffusion/comfy_workflow.py b/ai_diffusion/comfy_workflow.py
index 6c8fe5a4a..1514ed49f 100644
--- a/ai_diffusion/comfy_workflow.py
+++ b/ai_diffusion/comfy_workflow.py
@@ -417,6 +417,14 @@ def load_dual_clip(self, clip_name1: str, clip_name2: str, type: str):
             node = "DualCLIPLoaderGGUF"
         return self.add_cached(node, 1, clip_name1=clip_name1, clip_name2=clip_name2, type=type)
 
+    def load_triple_clip(self, clip_name1: str, clip_name2: str, clip_name3: str):
+        node = "TripleCLIPLoader"
+        if any(f.endswith(".gguf") for f in (clip_name1, clip_name2, clip_name3)):
+            node = "TripleCLIPLoaderGGUF"
+        return self.add_cached(
+            node, 1, clip_name1=clip_name1, clip_name2=clip_name2, clip_name3=clip_name3
+        )
+
     def load_vae(self, vae_name: str):
         return self.add_cached("VAELoader", 1, vae_name=vae_name)
 
diff --git a/ai_diffusion/resources.py b/ai_diffusion/resources.py
index 18511d4c3..91266f119 100644
--- a/ai_diffusion/resources.py
+++ b/ai_diffusion/resources.py
@@ -6,10 +6,10 @@
 
 # Version identifier for all the resources defined here. This is used as the server version.
 # It usually follows the plugin version, but not all new plugin versions also require a server update.
-version = "1.26.0"
+version = "1.27.0"
 
 comfy_url = "https://github.com/comfyanonymous/ComfyUI"
-comfy_version = "1b8089528502a881d0ed2918b2abd54441743dd0"
+comfy_version = "8ce2a1052ca03183768da0aaa483024e58b8008c"
 
 
 class CustomNode(NamedTuple):
@@ -56,7 +56,7 @@ class CustomNode(NamedTuple):
         "GGUF",
         "ComfyUI-GGUF",
         "https://github.com/city96/ComfyUI-GGUF",
-        "454955ead3336322215a206edbd7191eb130bba0",
+        "68ad5fb2149ead7e4978f83f14e045ecd812a394",
         ["UnetLoaderGGUF", "DualCLIPLoaderGGUF"],
     )
 ]
diff --git a/ai_diffusion/workflow.py b/ai_diffusion/workflow.py
index 4cc771ff4..c58507b78 100644
--- a/ai_diffusion/workflow.py
+++ b/ai_diffusion/workflow.py
@@ -112,7 +112,10 @@ def load_checkpoint_with_lora(w: ComfyWorkflow, checkpoint: CheckpointInput, mod
             case Arch.sdxl:
                 clip = w.load_dual_clip(te_models["clip_g"], te_models["clip_l"], type="sdxl")
             case Arch.sd3:
-                clip = w.load_dual_clip(te_models["clip_g"], te_models["clip_l"], type="sd3")
+                if te_models.find("t5"):
+                    clip = w.load_triple_clip(te_models["clip_l"], te_models["clip_g"], te_models["t5"])
+                else:
+                    clip = w.load_dual_clip(te_models["clip_g"], te_models["clip_l"], type="sd3")
             case Arch.flux:
                 clip = w.load_dual_clip(te_models["clip_l"], te_models["t5"], type="flux")
             case _: