diff --git a/installer.py b/installer.py
index d59908bf0..c0fd2faf9 100644
--- a/installer.py
+++ b/installer.py
@@ -603,7 +603,7 @@ def check_diffusers():
     if args.skip_git:
         install('diffusers')
         return
-    sha = '9b721db205729d5a6e97a72312c3a0f4534064f1' # diffusers commit hash
+    sha = '764b62473ac10afd9c52e6b3f3f528f719bc7a34' # diffusers commit hash
     pkg = pkg_resources.working_set.by_key.get('diffusers', None)
     minor = int(pkg.version.split('.')[1] if pkg is not None else -1)
     cur = opts.get('diffusers_version', '') if minor > -1 else ''
diff --git a/modules/model_quant.py b/modules/model_quant.py
index 3275a6e91..45f940c59 100644
--- a/modules/model_quant.py
+++ b/modules/model_quant.py
@@ -12,6 +12,7 @@ from installer import installed, install, log, setup_logging
 ao = None
 bnb = None
 optimum_quanto = None
+trt = None
 quant_last_model_name = None
 quant_last_model_device = None
 debug = os.environ.get('SD_QUANT_DEBUG', None) is not None
@@ -103,6 +104,33 @@ def create_quanto_config(kwargs = None, allow: bool = True, module: str = 'Model
     return kwargs
 
 
+def create_trt_config(kwargs = None, allow: bool = True, module: str = 'Model', modules_to_not_convert: list = []):
+    from modules import shared
+    if allow and (module == 'any' or module in shared.opts.trt_quantization):
+        load_trt()
+        if trt is None:
+            return kwargs
+        trt_config_data = {
+            "fp8": {"quant_type": "FP8", "quant_method": "modelopt", "modules_to_not_convert": []},
+            "int8": {"quant_type": "INT8", "quant_method": "modelopt", "modules_to_not_convert": []},
+            "int4": {"quant_type": "INT4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "modules_to_not_convert": ["conv", "patch_embed"]},
+            "nf4": {"quant_type": "NF4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "scale_block_quantize": 8, "scale_channel_quantize": -1, "modules_to_not_convert": ["conv"]},
+            "nvfp4": {"quant_type": "NVFP4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "modules_to_not_convert": ["conv"]},
+        }
+        trt_quant_config = trt_config_data[shared.opts.trt_quantization_type].copy()
+        for m in modules_to_not_convert:
+            if m not in trt_quant_config['modules_to_not_convert']:
+                trt_quant_config['modules_to_not_convert'].append(m)
+        trt_config = diffusers.quantizers.quantization_config.NVIDIAModelOptConfig(**trt_quant_config)
+        log.debug(f'Quantization: module={module} type=tensorrt dtype={shared.opts.trt_quantization_type}')
+        if kwargs is None:
+            return trt_config
+        else:
+            kwargs['quantization_config'] = trt_config
+            return kwargs
+    return kwargs
+
+
 def get_sdnq_devices():
     from modules import devices, shared
     if shared.opts.device_map == "gpu":
@@ -122,6 +150,7 @@ def get_sdnq_devices():
         return_device = None
     return quantization_device, return_device
 
+
 def create_sdnq_config(kwargs = None, allow: bool = True, module: str = 'Model', weights_dtype: str = None, modules_to_not_convert: list = [], modules_dtype_dict: dict = {}):
     from modules import shared
     if allow and (shared.opts.sdnq_quantize_mode in {'pre', 'auto'}) and (module == 'any' or module in shared.opts.sdnq_quantize_weights):
@@ -224,6 +253,11 @@ def create_config(kwargs = None, allow: bool = True, module: str = 'Model', modu
         if debug:
             log.trace(f'Quantization: type=torchao config={kwargs.get("quantization_config", None)}')
         return kwargs
+    kwargs = create_trt_config(kwargs, allow=allow, module=module, modules_to_not_convert=modules_to_not_convert)
+    if kwargs is not None and 'quantization_config' in kwargs:
+        if debug:
+            log.trace(f'Quantization: type=tensorrt config={kwargs.get("quantization_config", None)}')
+        return kwargs
     return kwargs
 
 
@@ -309,6 +343,30 @@ def load_quanto(msg='', silent=False):
     return None
 
 
+def load_trt(msg='', silent=False):
+    global trt # pylint: disable=global-statement
+    if trt is not None:
+        return trt
+    try:
+        install('nvidia-modelopt')
+        import pydantic
+        if pydantic.__version__.startswith('1'):
+            log.error('Quantization: type=tensorrt pydantic==2 required')
+            return None
+        import modelopt
+        trt = modelopt
+        fn = f'{sys._getframe(3).f_code.co_name}:{sys._getframe(2).f_code.co_name}:{sys._getframe(1).f_code.co_name}' # pylint: disable=protected-access
+        log.debug(f'Quantization: type=tensorrt version={trt.__version__} fn={fn}') # pylint: disable=protected-access
+        return trt
+    except Exception as e:
+        if len(msg) > 0:
+            log.error(f"{msg} failed to import tensorrt: {e}")
+        trt = None
+        if not silent:
+            raise
+    return None
+
+
 def upcast_non_layerwise_modules(model, dtype): # pylint: disable=unused-argument
     from diffusers.hooks.layerwise_casting import _GO_LC_SUPPORTED_PYTORCH_LAYERS
     model_children = list(model.children())
diff --git a/modules/shared.py b/modules/shared.py
index 0e67b5fd1..f5774461a 100644
--- a/modules/shared.py
+++ b/modules/shared.py
@@ -193,6 +193,11 @@ options_templates.update(options_section(("quantization", "Model Quantization"),
     "sdnq_dequantize_fp32": OptionInfo(False, "Dequantize using full precision", gr.Checkbox),
     "sdnq_quantize_shuffle_weights": OptionInfo(False, "Shuffle weights in post mode", gr.Checkbox),
 
+    "nunchaku_sep": OptionInfo("<h2>Nunchaku Engine</h2>", "", gr.HTML),
+    "nunchaku_quantization": OptionInfo([], "SVDQuant enabled", gr.CheckboxGroup, {"choices": ["Model", "TE"]}),
+    "nunchaku_attention": OptionInfo(False, "Nunchaku attention", gr.Checkbox),
+    "nunchaku_offload": OptionInfo(False, "Nunchaku offloading", gr.Checkbox),
+
     "bnb_quantization_sep": OptionInfo("<h2>BitsAndBytes</h2>", "", gr.HTML),
     "bnb_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "TE", "LLM", "VAE"]}),
     "bnb_quantization_type": OptionInfo("nf4", "Quantization type", gr.Dropdown, {"choices": ["nf4", "fp8", "fp4"]}),
@@ -218,10 +223,9 @@ options_templates.update(options_section(("quantization", "Model Quantization"),
     "layerwise_quantization_storage": OptionInfo("float8_e4m3fn", "Layerwise casting storage", gr.Dropdown, {"choices": ["float8_e4m3fn", "float8_e5m2"]}),
     "layerwise_quantization_nonblocking": OptionInfo(False, "Layerwise non-blocking operations", gr.Checkbox),
 
-    "nunchaku_sep": OptionInfo("<h2>Nunchaku Engine</h2>", "", gr.HTML),
-    "nunchaku_quantization": OptionInfo([], "SVDQuant enabled", gr.CheckboxGroup, {"choices": ["Model", "TE"]}),
-    "nunchaku_attention": OptionInfo(False, "Nunchaku attention", gr.Checkbox),
-    "nunchaku_offload": OptionInfo(False, "Nunchaku offloading", gr.Checkbox),
+    "trt_quantization_sep": OptionInfo("<h2>TensorRT</h2>", "", gr.HTML),
+    "trt_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "TE", "LLM", "VAE"]}),
+    "trt_quantization_type": OptionInfo("nvfp4", "Quantization type", gr.Dropdown, {"choices": ["int8", "int4", "fp8", "nf4", "nvfp4"]}),
 
     "nncf_compress_sep": OptionInfo("<h2>NNCF: Neural Network Compression Framework</h2>", "", gr.HTML, {"visible": cmd_opts.use_openvino}),
     "nncf_compress_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "TE", "VAE"], "visible": cmd_opts.use_openvino}),