diff --git a/verl/workers/rollout/vllm_rollout/vllm_omni_async_server.py b/verl/workers/rollout/vllm_rollout/vllm_omni_async_server.py
index 6d21f66e710..df493e42d7b 100644
--- a/verl/workers/rollout/vllm_rollout/vllm_omni_async_server.py
+++ b/verl/workers/rollout/vllm_rollout/vllm_omni_async_server.py
@@ -171,7 +171,20 @@ async def generate(
                 extra_args[k] = v
         sampling_kwargs["extra_args"] = extra_args
         if lora_request is not None:
-            sampling_kwargs["lora_request"] = lora_request
+            # vllm-omni >=0.19 renames lora_request/lora_scale to the plural
+            # lora_requests/lora_scales (list form) for multi-LoRA composition.
+            # Detect which field the installed version exposes so this adapter
+            # works on both pre- and post-rename releases.
+            if "lora_requests" in OmniDiffusionSamplingParams.__dataclass_fields__:
+                sampling_kwargs["lora_requests"] = [lora_request]
+                # On the post-rename API the singular "lora_scale" key from
+                # sampling_params would have fallen into extra_args above (no
+                # such field on the dataclass); lift it back out into the
+                # plural lora_scales so the scale actually takes effect.
+                if "lora_scale" in extra_args:
+                    sampling_kwargs["lora_scales"] = [extra_args.pop("lora_scale")]
+            else:
+                sampling_kwargs["lora_request"] = lora_request
         diffusion_sampling_params = OmniDiffusionSamplingParams(**sampling_kwargs)
 
         # Call AsyncOmni.generate() with the correct API