Alert creation overrides user-defined summarization prompts in VSS UI

Accelerated Computing Intelligent Video Analytics Visual AI Agent
1

Setup:

Steps to Reproduce:

  1. Open the VSS UI application.

  2. Upload a video.

  3. Set the following prompts:

    • Caption Summarization Prompt

    • Summary Aggregation Prompt
      (Ensure the prompts do not mention warehouse, operational inefficiency, etc.)

  4. Navigate to the Create Alerts tab.

  5. Create a new alert.

  6. Click Summarize and wait for the summary to be generated.

Actual Outcome:
The generated summary includes default prompt content related to warehouse operations, ignoring the user-defined prompts.

Expected Outcome:
The summary should reflect the user-defined caption summarization and summary aggregation prompts.

Additional Notes:

3

Thanks for your feedback.

Could you share the Caption Summarization Prompt and Summary Aggregation Prompt that you modified?

I tried to reproduce the problem following your steps, but I didn’t find any difference.

I even modified the code here.

--- a/src/vss-engine/src/via_stream_handler.py
+++ b/src/vss-engine/src/via_stream_handler.py
@@ -1835,14 +1835,15 @@ class ViaStreamHandler:
         req_info.nvtx_summarization_start = nvtx.start_range(
             message="Summarization-" + str(req_info.request_id), color="blue"
         )
+        print(f"***************** summary_aggregation_prompt {req_info.summary_aggregation_prompt} caption_summarization_prompt {req_info.caption_summarization_prompt} vlm_prompt {req_info.vlm_request_params.vlm_prompt}")
         if not self._args.disable_ca_rag and not skip_ca_rag:
             with self._lock:
                 self._create_ctx_mgr_pool(self._ca_rag_config)
                 req_info._ctx_mgr = self.get_ctx_mgr(req_info.assets)
             try:
-                config = deepcopy(self._ca_rag_config)
-                config["context_manager"]["uuid"] = req_info.stream_id
-                req_info._ctx_mgr.configure(config=config)
+                if req_info._ctx_mgr:
+                    ca_rag_config = self.update_ca_rag_config(req_info)
+                    req_info._ctx_mgr.configure(ca_rag_config)
4

Please find attached a recording of me reproducing the defect. I’ve also included the relevant logs.

Kindly note the line starting with:

2025-10-28 13:21:40,384 … Configuring init for 7 with config: … 'prompts': {'caption': 'Write a concise and clear dense caption for the provided warehouse video, …

It appears that the tools are using an incorrect configuration.

Let me know if you need any further details.

logs_01.txt (12.8 MB)

5

Thanks for your feedback. Yes, you are right.

I now know how to reproduce the problem. This issue is not fixed in the nvcr.io/nvidia/blueprint/vss-engine:2.4.0 image, so brevlab cloud also include this issue. But it has been fixed in the latest code (That is why I couldn’t reproduce it in my previous reply).

Please refer to this commit.

This patch that you provided is also necessary.

6

Thank you, when are you planning a release containing this fix?

7

It should be included in the next release; please wait for the official announcement for the exact date.

As a workaround, you can add the following configuration to your compose.yml file to override the code in the container with your local code for debugging.

volumes:
- xxxx/video-search-and-summarization/src/vss-engine/src:/opt/nvidia/via/via-engine
8

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks.