Add model-specific Description summary

.gitattributes

@@ -52,3 +52,4 @@ assets/edge_action_fd_umi_2chunk_output.mp4 filter=lfs diff=lfs merge=lfs -text
 assets/edge_action_id_av_0_output.png filter=lfs diff=lfs merge=lfs -text
 assets/edge_action_id_av_1_output.png filter=lfs diff=lfs merge=lfs -text
 assets/diffusers_outputs/edge_i2v_diffusers.mp4 filter=lfs diff=lfs merge=lfs -text
+assets/cosmos3_model_tree_edge.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -19,6 +19,17 @@ tags:
 
 ## Description
 
+Cosmos3-Edge is the most compact member of the Cosmos3 family — a 4B-parameter Omnimodal world model that generates coherent text, image, video, and action outputs from combinations of text, image, video, and action-trajectory inputs. It brings Cosmos3's multimodal understanding and world-generation capabilities to latency- and resource-constrained deployments.
+
+- **What's different from the larger models:** At 4B parameters, Cosmos3-Edge is the smallest model in the family (vs. Cosmos3-Nano 16B and Cosmos3-Super 64B), trading scale for efficiency under tighter compute, memory, and latency budgets while keeping the same MoT architecture and omnimodal interface.
+- **Benchmarks:** Evaluated across both reasoning (general, robotics, smart-infrastructure, driving) and generation, it delivers the family's highest image-to-video throughput at competitive quality on PAIBench, RBench, and PhysicsIQ. See the [Benchmarks](https://huggingface.co/nvidia/Cosmos3-Edge#benchmarks) section.
+- **Example usage and output:** See the [Usage](https://huggingface.co/nvidia/Cosmos3-Edge#usage) section.
+- **Hardware:** The compact 4B size is intended for efficient single-GPU deployment; see [Usage](https://huggingface.co/nvidia/Cosmos3-Edge#usage) and [PBR](https://huggingface.co/nvidia/Cosmos3-Edge#pbr-performance-benchmark-reporting) for latency estimates.
+
+![Cosmos3 model family tree](assets/cosmos3_model_tree_edge.png)
+
+Cosmos3-Edge was developed by NVIDIA as a part of Cosmos3.
+
 Cosmos3 is a collection of Omnimodal world models capable of generating dynamic, high-quality video, image, audio, and action commands from combinations of text, image, video, and action trajectory inputs. It serves as a foundational building block for a broad range of Physical AI applications and research spanning world understanding, world generation, simulation, and embodied policy learning.
 
 This model is ready for commercial and non-commercial use.