From 2b582c2084266358cfe1b91c249faf739460ed0d Mon Sep 17 00:00:00 2001
From: "Jha, Gopi Krishna" <gopi.krishna.jha@intel.com>
Date: Wed, 30 Sep 2026 10:28:56 +0530
Subject: [PATCH] Enable Intel XPU support for Cosmos3-Edge

---
 cookbooks/cosmos3/README.md                   | 25 ++++++++++++++++++-
 .../cosmos3/generator/audiovisual/README.md   |  7 ++++--
 2 files changed, 29 insertions(+), 3 deletions(-)

diff --git a/cookbooks/cosmos3/README.md b/cookbooks/cosmos3/README.md
index 9dd21aa5..e631d311 100644
--- a/cookbooks/cosmos3/README.md
+++ b/cookbooks/cosmos3/README.md
@@ -19,7 +19,8 @@ backend you want to run and follow that one section.
 
 ## Prerequisites
 
-- Linux with NVIDIA GPU access.
+- Linux with NVIDIA GPU access, or an Intel GPU supported by PyTorch XPU for
+  the Diffusers backend.
 - [`uv`](https://docs.astral.sh/uv/getting-started/installation/), `git`, and `git-lfs` installed.
 - Hugging Face access to the gated Cosmos3 model repos. Generator also requires
   access to the gated
@@ -165,6 +166,28 @@ uv pip install --torch-backend=cu130 \
   transformers
 ```
 
+For Intel GPUs supported by PyTorch XPU, use the `xpu` Torch backend instead:
+
+```bash
+uv venv --python 3.13 --seed --managed-python
+source .venv/bin/activate
+
+uv pip install --torch-backend=xpu \
+  "diffusers @ git+https://github.com/huggingface/diffusers.git" \
+  accelerate \
+  av \
+  cosmos_guardrail \
+  huggingface_hub \
+  imageio \
+  imageio-ffmpeg \
+  torch \
+  torchvision \
+  transformers
+```
+
+The Intel XPU path is supported by the Diffusers Cosmos3-Edge generator. The
+other backends in this guide retain their documented NVIDIA GPU requirements.
+
 ## TensorRT-LLM Generator
 
 OpenAI-compatible **VisualGen** server for Generator audiovisual text-to-image,
diff --git a/cookbooks/cosmos3/generator/audiovisual/README.md b/cookbooks/cosmos3/generator/audiovisual/README.md
index 627275a1..5c1778cf 100644
--- a/cookbooks/cosmos3/generator/audiovisual/README.md
+++ b/cookbooks/cosmos3/generator/audiovisual/README.md
@@ -118,10 +118,12 @@ from diffusers.utils import export_to_video
 
 prompt = json.load(open("assets/prompts/text2video/robot_kitchen.json"))
 negative = json.load(open("assets/negative_prompts/text2video/neg_prompt.json"))
+device = "xpu" if torch.xpu.is_available() else "cuda"
 
 pipe = Cosmos3OmniPipeline.from_pretrained(
-    "nvidia/Cosmos3-Nano", torch_dtype=torch.bfloat16, device_map="cuda"
+    "nvidia/Cosmos3-Nano", torch_dtype=torch.bfloat16
 )
+pipe.to(device)
 pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=10.0)
 
 result = pipe(
@@ -137,7 +139,7 @@ result = pipe(
     enable_sound=False,
     add_resolution_template=False,
     add_duration_template=False,
-    generator=torch.Generator(device="cuda").manual_seed(1234),
+    generator=torch.Generator(device=device).manual_seed(1234),
 )
 export_to_video(result.video, "/tmp/cosmos3_t2v_diffusers.mp4", fps=24)
 ```
@@ -148,6 +150,7 @@ To run **Cosmos3-Super** instead, load the larger checkpoint:
 To run **Cosmos3-Edge** instead, load `nvidia/Cosmos3-Edge` and use its
 single-GPU, no-audio settings: `height=480`, `width=832`, `num_frames=121`,
 `num_inference_steps=20`, `guidance_scale=5.0`, and `flow_shift=8.0`.
+The Diffusers device selection above supports both CUDA and Intel XPU devices.
 
 ### Notebook walkthrough