diff --git a/nodes.py b/nodes.py
index 689e8c0..b5089fb 100644
--- a/nodes.py
+++ b/nodes.py
@@ -1806,6 +1806,9 @@ class WanVideoSampler:
                         image_cond = torch.cat([mask_latents, masked_video_latents_input], dim=0).to(device)
                     clip_fea = None
                     fun_ref_image = control_embeds.get("fun_ref_image", None)
+                    if fun_ref_image is not None:
+                        if transformer.ref_conv.weight.dtype in [torch.float8_e4m3fn, torch.float8_e5m2]:
+                            raise ValueError("Fun-Control reference image won't work with this specific fp8_scaled model, it's been fixed in latest version of the model")
                 control_start_percent = control_embeds.get("start_percent", 0.0)
                 control_end_percent = control_embeds.get("end_percent", 1.0)
             else:
diff --git a/nodes_model_loading.py b/nodes_model_loading.py
index b84f79b..aec9f5d 100644
--- a/nodes_model_loading.py
+++ b/nodes_model_loading.py
@@ -1048,7 +1048,7 @@ class WanVideoModelLoader:
                 dtype = torch.float8_e5m2
             else:
                 dtype = base_dtype
-            params_to_keep = {"norm", "bias", "time_in", "patch_embedding", "time_", "img_emb", "modulation", "text_embedding", "adapter", "add"}
+            params_to_keep = {"norm", "bias", "time_in", "patch_embedding", "time_", "img_emb", "modulation", "text_embedding", "adapter", "add", "ref_conv"}
             if not lora_low_mem_load:
                 log.info("Using accelerate to load and assign model weights to device...")
                 param_count = sum(1 for _ in transformer.named_parameters())