[None][feat] Add Qwen-Image-Layered baseline support

docs/source/models/supported-models.md

@@ -165,6 +165,7 @@ For full documentation, see the [Visual Generation](./visual-generation.md) page
 | `Lightricks/LTX-2` | Text-to-Video (with Audio), Image-to-Video (with Audio) |
 | `Qwen/Qwen-Image` | Text-to-Image |
 | `Qwen/Qwen-Image-2512` | Text-to-Image |
+| `Qwen/Qwen-Image-Layered` | Image-to-Image |
 | `nvidia/Cosmos3-Nano` | Text-to-Image, Text-to-Video, Image-to-Video |
 | `nvidia/Cosmos3-Super` | Text-to-Image, Text-to-Video, Image-to-Video |
 
@@ -178,6 +179,8 @@ For full documentation, see the [Visual Generation](./visual-generation.md) page
 | **Wan 2.2** | Yes | Yes | No | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
 | **LTX-2** | Yes | Yes | No | Yes | Yes | No | No | Yes | Yes | Yes | Yes | No |
 | **Qwen-Image** [^2] | Yes | Yes | No | No | Yes | No | Yes | Yes | Yes | Yes | Yes | No |
+| **Qwen-Image-Layered** [^3] | No | No | No | No | No | No | Yes | Yes | No | No | No | No |
 | **Cosmos3** | Yes | Yes | No | Yes | Yes | Yes | Yes | Yes | Yes | No | No | Yes |
 
 [^vg1]: FLUX models use embedded guidance and do not have a separate negative prompt path, so CFG parallelism is not applicable.
+[^3]: Qwen-Image-Layered supports baseline BF16 image-conditioned layer decomposition. FP8 blockwise, NVFP4, `trtllm-serve` image-edit routing, and attention-parallel backends are not enabled yet.

docs/source/models/visual-generation.md

@@ -37,6 +37,7 @@ TensorRT-LLM **VisualGen** provides a unified inference stack for diffusion mode
 | `Lightricks/LTX-2` | Text-to-Video (with Audio), Image-to-Video (with Audio) |
 | `Qwen/Qwen-Image` | Text-to-Image |
 | `Qwen/Qwen-Image-2512` | Text-to-Image |
+| `Qwen/Qwen-Image-Layered` | Image-to-Image |
 | `nvidia/Cosmos3-Nano` | Text-to-Image, Text-to-Video, Image-to-Video |
 | `nvidia/Cosmos3-Super` | Text-to-Image, Text-to-Video, Image-to-Video |
 
@@ -53,6 +54,7 @@ Models are auto-detected from the checkpoint directory. Diffusers-format models
 | **Wan 2.2** | Yes | Yes | Yes [^3] | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | Yes | No |
 | **LTX-2** | Yes | Yes | Yes [^4] | Yes | Yes | Yes | No | No | Yes | Yes | Yes | Yes | No | No |
 | **Qwen-Image** [^5] | Yes | Yes | No | No | No | Yes | No | Yes | Yes | Yes | Yes | Yes | No | No |
+| **Qwen-Image-Layered** [^6] | No | No | No | No | No | No | No | Yes | Yes | No | No | No | No | No |
 | **Cosmos3** | Yes | Yes | No | No | Yes | Yes | Yes | Yes | Yes | Yes | No | No | Yes | No |
 
 [^1]: FLUX models use embedded guidance and do not have a separate negative prompt path, so CFG parallelism is not applicable.
@@ -65,6 +67,8 @@ Models are auto-detected from the checkpoint directory. Diffusers-format models
 
 [^5]: Qwen-Image ships a native BF16 implementation with per-module numerical parity against `diffusers.QwenImagePipeline` (cosine similarity >= 0.999 on the full 20B transformer) and supports `trtllm-serve` / `/v1/images/generations`. VisualGen supports FP8 blockwise and NVFP4 dynamic quantization from BF16 checkpoints, as well as direct loading of statically quantized FP8 and NVFP4 ModelOpt checkpoints.
 
+[^6]: Qwen-Image-Layered supports baseline BF16 image-conditioned layer decomposition and returns the generated RGBA layer stack as a saveable image grid. FP8 blockwise, NVFP4, cache acceleration, attention-parallel/Sage/VSA backends, Tensor Parallelism, and `trtllm-serve` image-edit routing are not enabled for this pipeline yet.
+
 ## Quick Start
 
 Here is a simple example to generate a video with Wan 2.1:

examples/visual_gen/README.md

@@ -23,6 +23,7 @@ python models/flux1.py
 python models/flux2.py
 python models/cosmos3_ti2v.py --prompt "A robot arm picks fruit in a grocery store"
 python models/qwen_image.py
+python models/qwen_image_layered.py --image /path/to/image.png
 
 # With engine config (quant, parallelism, etc.)
 python models/wan_t2v.py --visual_gen_args configs/wan2.2-t2v-fp4-1gpu.yaml
@@ -32,6 +33,7 @@ python models/flux1.py --visual_gen_args configs/flux1-dev-fp4-1gpu.yaml
 python models/flux2.py --visual_gen_args configs/flux2-dev-fp4-1gpu.yaml
 python models/cosmos3_ti2v.py --visual_gen_args configs/cosmos3-nano-1gpu.yaml --prompt "A robot arm picks fruit in a grocery store"
 python models/qwen_image.py --visual_gen_args configs/qwen-image-fp8-1gpu.yaml
+python models/qwen_image_layered.py --visual_gen_args configs/qwen-image-layered-1gpu.yaml --image /path/to/image.png
 ```
 
 Install deps from the repo root: `pip install -r requirements-dev.txt`.

examples/visual_gen/configs/qwen-image-layered-1gpu.yaml

@@ -0,0 +1,24 @@
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# 1-GPU Qwen-Image-Layered baseline.
+# Load with VisualGenArgs.from_yaml() or example scripts that accept --visual_gen_args.
+attention_config:
+  backend: VANILLA
+parallel_config:
+  cfg_size: 1
+  ulysses_size: 1
+cuda_graph_config:
+  enable: false

examples/visual_gen/models/qwen_image_layered.py

@@ -0,0 +1,73 @@
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Qwen-Image-Layered image decomposition.
+
+Usage:
+    python qwen_image_layered.py --image input.png
+    python qwen_image_layered.py --visual_gen_args ../configs/qwen-image-layered-1gpu.yaml \
+        --image input.png
+"""
+
+import argparse
+
+from tensorrt_llm import VisualGen, VisualGenArgs
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument(
+        "--model",
+        default="Qwen/Qwen-Image-Layered",
+        help="Hugging Face model id or local checkpoint path.",
+    )
+    parser.add_argument(
+        "--visual_gen_args",
+        "--extra_visual_gen_options",
+        dest="visual_gen_args",
+        help="Optional VisualGenArgs YAML file.",
+    )
+    parser.add_argument(
+        "--image",
+        required=True,
+        help="Input image path.",
+    )
+    parser.add_argument(
+        "--prompt",
+        default="",
+        help="Optional text prompt. Empty prompt enables image captioning.",
+    )
+    parser.add_argument(
+        "--output_path",
+        default="qwen_image_layered_output.png",
+        help="Path to save the layer grid image.",
+    )
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+    extra_args = VisualGenArgs.from_yaml(args.visual_gen_args) if args.visual_gen_args else None
+    visual_gen = VisualGen(model=args.model, args=extra_args)
+
+    params = visual_gen.default_params
+    params.image = args.image
+
+    output = visual_gen.generate(inputs=args.prompt, params=params)
+    saved = output.save(args.output_path)
+    print(f"Saved image to {saved}")
+
+
+if __name__ == "__main__":
+    main()

tensorrt_llm/_torch/visual_gen/models/__init__.py

@@ -37,13 +37,15 @@
 from .flux import Flux2Pipeline, FluxPipeline
 from .ltx2 import LTX2Pipeline  # noqa: F401
 from .qwen_image import QwenImagePipeline
+from .qwen_image_layered import QwenImageLayeredPipeline
 from .wan import WanImageToVideoPipeline, WanPipeline
 
 __all__ = [
     "AutoPipeline",
     "BasePipeline",
     "FluxPipeline",
     "Flux2Pipeline",
+    "QwenImageLayeredPipeline",
     "QwenImagePipeline",
     "WanPipeline",
     "WanImageToVideoPipeline",

tensorrt_llm/_torch/visual_gen/models/qwen_image_layered/__init__.py

@@ -0,0 +1,24 @@
+# SPDX-FileCopyrightText: Copyright (c) 2022-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Qwen-Image-Layered pipeline exports."""
+
+from .pipeline_qwen_image_layered import QwenImageLayeredPipeline
+from .transformer_qwen_image_layered import QwenEmbedLayer3DRope, QwenImageLayeredTransformer2DModel
+
+__all__ = [
+    "QwenEmbedLayer3DRope",
+    "QwenImageLayeredPipeline",
+    "QwenImageLayeredTransformer2DModel",
+]