update version to 2.0.7

* add FireRed-Image-Edit-1.1

* flux.2-dev-edit

* flux.2-dev-edit

* flux.2-dev-edit

README.md

@@ -32,7 +32,7 @@ We believe that a well-developed open-source code framework can lower the thresh
 > DiffSynth-Studio has undergone major version updates, and some old features are no longer maintained. If you need to use old features, please switch to the [last historical version](https://github.com/modelscope/DiffSynth-Studio/tree/afd101f3452c9ecae0c87b79adfa2e22d65ffdc3) before the major version update.
 
 > Currently, the development personnel of this project are limited, with most of the work handled by [Artiprocher](https://github.com/Artiprocher) and [mi804](https://github.com/mi804). Therefore, the progress of new feature development will be relatively slow, and the speed of responding to and resolving issues is limited. We apologize for this and ask developers to understand.
-- **January 19, 2026**: Added support for [openmoss/MOVA-720p](https://modelscope.cn/models/openmoss/MOVA-720p) and [openmoss/MOVA-360p](https://modelscope.cn/models/openmoss/MOVA-360p) models, including training and inference capabilities. [Documentation](/docs/en/Model_Details/Wan.md) and [example code](/examples/mova/) are now available.
+- **March 19, 2026**: Added support for [openmoss/MOVA-720p](https://modelscope.cn/models/openmoss/MOVA-720p) and [openmoss/MOVA-360p](https://modelscope.cn/models/openmoss/MOVA-360p) models, including training and inference capabilities. [Documentation](/docs/en/Model_Details/Wan.md) and [example code](/examples/mova/) are now available.
 
 - **March 12, 2026**: We have added support for the [LTX-2.3](https://modelscope.cn/models/Lightricks/LTX-2.3) audio-video generation model. The features includes text-to-audio/video, image-to-audio/video, IC-LoRA control, audio-to-video, and audio-video inpainting. We have supported the complete inference and training functionalities. For details, please refer to the [documentation](/docs/en/Model_Details/LTX-2.md) and [code](/examples/ltx2/).
 

README_zh.md

@@ -33,7 +33,7 @@ DiffSynth 目前包括两个开源项目：
 
 > 目前本项目的开发人员有限，大部分工作由 [Artiprocher](https://github.com/Artiprocher) 和 [mi804](https://github.com/mi804) 负责，因此新功能的开发进展会比较缓慢，issue 的回复和解决速度有限，我们对此感到非常抱歉，请各位开发者理解。
 
-- **2026年1月19日** 新增对 [openmoss/MOVA-720p](https://modelscope.cn/models/openmoss/MOVA-720p) 和 [openmoss/MOVA-360p](https://modelscope.cn/models/openmoss/MOVA-360p) 模型的支持，包括完整的训练和推理功能。[文档](/docs/zh/Model_Details/Wan.md)和[示例代码](/examples/mova/)现已可用。
+- **2026年3月19日** 新增对 [openmoss/MOVA-720p](https://modelscope.cn/models/openmoss/MOVA-720p) 和 [openmoss/MOVA-360p](https://modelscope.cn/models/openmoss/MOVA-360p) 模型的支持，包括完整的训练和推理功能。[文档](/docs/zh/Model_Details/Wan.md)和[示例代码](/examples/mova/)现已可用。
 
 - **2026年3月12日** 我们新增了 [LTX-2.3](https://modelscope.cn/models/Lightricks/LTX-2.3) 音视频生成模型的支持，模型支持的功能包括文生音视频、图生音视频、IC-LoRA控制、音频生视频、音视频局部Inpainting，框架支持完整的推理和训练功能。详细信息请参考 [文档](/docs/zh/Model_Details/LTX-2.md) 和 [示例代码](/examples/ltx2/)。
 

diffsynth/diffusion/base_pipeline.py

@@ -339,6 +339,38 @@ class BasePipeline(torch.nn.Module):
             noise_pred = noise_pred_posi
         return noise_pred
 
+    def compile_pipeline(self, mode: str = "default", dynamic: bool = True, fullgraph: bool = False, compile_models: list = None, **kwargs):
+        """
+        compile the pipeline with torch.compile. The models that will be compiled are determined by the `compilable_models` attribute of the pipeline.
+        If a model has `_repeated_blocks` attribute, we will compile these blocks with regional compilation. Otherwise, we will compile the whole model.
+        See https://docs.pytorch.org/docs/stable/generated/torch.compile.html#torch.compile for details about compilation arguments.
+        Args:
+            mode: The compilation mode, which will be passed to `torch.compile`, options are "default", "reduce-overhead", "max-autotune" and "max-autotune-no-cudagraphs. Default to "default".
+            dynamic: Whether to enable dynamic graph compilation to support dynamic input shapes, which will be passed to `torch.compile`. Default to True (recommended).
+            fullgraph: Whether to use full graph compilation, which will be passed to `torch.compile`. Default to False (recommended).
+            compile_models: The list of model names to be compiled. If None, we will compile the models in `pipeline.compilable_models`. Default to None.
+            **kwargs: Other arguments for `torch.compile`.
+        """
+        compile_models = compile_models or getattr(self, "compilable_models", [])
+        if len(compile_models) == 0:
+            print("No compilable models in the pipeline. Skip compilation.")
+            return
+        for name in compile_models:
+            model = getattr(self, name, None)
+            if model is None:
+                print(f"Model '{name}' not found in the pipeline.")
+                continue
+            repeated_blocks = getattr(model, "_repeated_blocks", None)
+            # regional compilation for repeated blocks.
+            if repeated_blocks is not None:
+                for submod in model.modules():
+                    if submod.__class__.__name__ in repeated_blocks:
+                        submod.compile(mode=mode, dynamic=dynamic, fullgraph=fullgraph, **kwargs)
+            # compile the whole model.
+            else:
+                model.compile(mode=mode, dynamic=dynamic, fullgraph=fullgraph, **kwargs)
+            print(f"{name} is compiled with mode={mode}, dynamic={dynamic}, fullgraph={fullgraph}.")
+
 
 class PipelineUnitGraph:
     def __init__(self):

diffsynth/models/anima_dit.py

@@ -1270,6 +1270,9 @@ class LLMAdapter(nn.Module):
 
 
 class AnimaDiT(MiniTrainDIT):
+
+    _repeated_blocks = ["Block"]
+
     def __init__(self):
         kwargs = {'image_model': 'anima', 'max_img_h': 240, 'max_img_w': 240, 'max_frames': 128, 'in_channels': 16, 'out_channels': 16, 'patch_spatial': 2, 'patch_temporal': 1, 'model_channels': 2048, 'concat_padding_mask': True, 'crossattn_emb_channels': 1024, 'pos_emb_cls': 'rope3d', 'pos_emb_learnable': True, 'pos_emb_interpolation': 'crop', 'min_fps': 1, 'max_fps': 30, 'use_adaln_lora': True, 'adaln_lora_dim': 256, 'num_blocks': 28, 'num_heads': 16, 'extra_per_block_abs_pos_emb': False, 'rope_h_extrapolation_ratio': 4.0, 'rope_w_extrapolation_ratio': 4.0, 'rope_t_extrapolation_ratio': 1.0, 'extra_h_extrapolation_ratio': 1.0, 'extra_w_extrapolation_ratio': 1.0, 'extra_t_extrapolation_ratio': 1.0, 'rope_enable_fps_modulation': False, 'dtype': torch.bfloat16, 'device': None, 'operations': torch.nn}
         super().__init__(**kwargs)

diffsynth/models/flux2_dit.py

@@ -879,6 +879,9 @@ class Flux2Modulation(nn.Module):
 
 
 class Flux2DiT(torch.nn.Module):
+
+    _repeated_blocks = ["Flux2TransformerBlock", "Flux2SingleTransformerBlock"]
+
     def __init__(
         self,
         patch_size: int = 1,

diffsynth/models/flux_dit.py

@@ -275,6 +275,9 @@ class AdaLayerNormContinuous(torch.nn.Module):
 
 
 class FluxDiT(torch.nn.Module):
+
+    _repeated_blocks = ["FluxJointTransformerBlock", "FluxSingleTransformerBlock"]
+
     def __init__(self, disable_guidance_embedder=False, input_dim=64, num_blocks=19):
         super().__init__()
         self.pos_embedder = RoPEEmbedding(3072, 10000, [16, 56, 56])

diffsynth/models/ltx2_dit.py

@@ -1280,6 +1280,7 @@ class LTXModel(torch.nn.Module):
     LTX model transformer implementation.
     This class implements the transformer blocks for the LTX model.
     """
+    _repeated_blocks = ["BasicAVTransformerBlock"]
 
     def __init__(  # noqa: PLR0913
         self,

diffsynth/models/qwen_image_dit.py

@@ -549,6 +549,9 @@ class QwenImageTransformerBlock(nn.Module):
 
 
 class QwenImageDiT(torch.nn.Module):
+
+    _repeated_blocks = ["QwenImageTransformerBlock"]
+
     def __init__(
         self,
         num_layers: int = 60,

diffsynth/models/wan_video_dit.py

@@ -336,6 +336,9 @@ class WanToDanceInjector(nn.Module):
 
 
 class WanModel(torch.nn.Module):
+
+    _repeated_blocks = ["DiTBlock"]
+
     def __init__(
         self,
         dim: int,

diffsynth/models/z_image_dit.py

@@ -326,6 +326,7 @@ class RopeEmbedder:
 class ZImageDiT(nn.Module):
     _supports_gradient_checkpointing = True
     _no_split_modules = ["ZImageTransformerBlock"]
+    _repeated_blocks = ["ZImageTransformerBlock"]
 
     def __init__(
         self,

diffsynth/pipelines/anima_image.py

@@ -39,6 +39,7 @@ class AnimaImagePipeline(BasePipeline):
             AnimaUnit_PromptEmbedder(),
         ]
         self.model_fn = model_fn_anima
+        self.compilable_models = ["dit"]
     
     
     @staticmethod

diffsynth/pipelines/flux2_image.py

@@ -42,6 +42,7 @@ class Flux2ImagePipeline(BasePipeline):
             Flux2Unit_ImageIDs(),
         ]
         self.model_fn = model_fn_flux2
+        self.compilable_models = ["dit"]
     
     
     @staticmethod

diffsynth/pipelines/flux_image.py

@@ -103,6 +103,7 @@ class FluxImagePipeline(BasePipeline):
             FluxImageUnit_LoRAEncode(),
         ]
         self.model_fn = model_fn_flux_image
+        self.compilable_models = ["dit"]
         self.lora_loader = FluxLoRALoader
 
     def enable_lora_merger(self):

diffsynth/pipelines/ltx2_audio_video.py

@@ -76,6 +76,7 @@ class LTX2AudioVideoPipeline(BasePipeline):
             LTX2AudioVideoUnit_SetScheduleStage2(),
         ]
         self.model_fn = model_fn_ltx2
+        self.compilable_models = ["dit"]
 
         self.default_negative_prompt = {
             "LTX-2": (

diffsynth/pipelines/mova_audio_video.py

@@ -52,6 +52,7 @@ class MovaAudioVideoPipeline(BasePipeline):
             MovaAudioVideoUnit_UnifiedSequenceParallel(),
         ]
         self.model_fn = model_fn_mova_audio_video
+        self.compilable_models = ["video_dit", "video_dit2", "audio_dit"]
 
     def enable_usp(self):
         from ..utils.xfuser import get_sequence_parallel_world_size, usp_attn_forward

diffsynth/pipelines/qwen_image.py

@@ -56,6 +56,7 @@ class QwenImagePipeline(BasePipeline):
             QwenImageUnit_BlockwiseControlNet(),
         ]
         self.model_fn = model_fn_qwen_image
+        self.compilable_models = ["dit"]
     
     
     @staticmethod

diffsynth/pipelines/wan_video.py

@@ -83,10 +83,11 @@ class WanVideoPipeline(BasePipeline):
             WanVideoPostUnit_S2V(),
         ]
         self.model_fn = model_fn_wan_video
+        self.compilable_models = ["dit", "dit2"]
 
 
     def enable_usp(self):
-        from ..utils.xfuser import get_sequence_parallel_world_size, usp_attn_forward, usp_dit_forward
+        from ..utils.xfuser import get_sequence_parallel_world_size, usp_attn_forward, usp_dit_forward, usp_vace_forward
 
         for block in self.dit.blocks:
             block.self_attn.forward = types.MethodType(usp_attn_forward, block.self_attn)
@@ -95,6 +96,14 @@ class WanVideoPipeline(BasePipeline):
             for block in self.dit2.blocks:
                 block.self_attn.forward = types.MethodType(usp_attn_forward, block.self_attn)
             self.dit2.forward = types.MethodType(usp_dit_forward, self.dit2)
+        if self.vace is not None:
+            for block in self.vace.vace_blocks:
+                block.self_attn.forward = types.MethodType(usp_attn_forward, block.self_attn)
+            self.vace.forward = types.MethodType(usp_vace_forward, self.vace)
+        if self.vace2 is not None:
+            for block in self.vace2.vace_blocks:
+                block.self_attn.forward = types.MethodType(usp_attn_forward, block.self_attn)
+            self.vace2.forward = types.MethodType(usp_vace_forward, self.vace2)
         self.sp_size = get_sequence_parallel_world_size()
         self.use_unified_sequence_parallel = True
 
@@ -1451,13 +1460,6 @@ def model_fn_wan_video(
     else:
         tea_cache_update = False
 
-    if vace_context is not None:
-        vace_hints = vace(
-            x, vace_context, context, t_mod, freqs,
-            use_gradient_checkpointing=use_gradient_checkpointing,
-            use_gradient_checkpointing_offload=use_gradient_checkpointing_offload
-        )
-
     # WanToDance
     if hasattr(dit, "wantodance_enable_global") and dit.wantodance_enable_global:
         if wantodance_refimage_feature is not None:
@@ -1519,6 +1521,13 @@ def model_fn_wan_video(
             pad_shape = chunks[0].shape[1] - chunks[-1].shape[1]
             chunks = [torch.nn.functional.pad(chunk, (0, 0, 0, chunks[0].shape[1]-chunk.shape[1]), value=0) for chunk in chunks]
             x = chunks[get_sequence_parallel_rank()]
+
+    if vace_context is not None:
+        vace_hints = vace(
+            x, vace_context, context, t_mod, freqs,
+            use_gradient_checkpointing=use_gradient_checkpointing,
+            use_gradient_checkpointing_offload=use_gradient_checkpointing_offload
+        )
     if tea_cache_update:
         x = tea_cache.update(x)
     else:
@@ -1561,9 +1570,6 @@ def model_fn_wan_video(
             # VACE
             if vace_context is not None and block_id in vace.vace_layers_mapping:
                 current_vace_hint = vace_hints[vace.vace_layers_mapping[block_id]]
-                if use_unified_sequence_parallel and dist.is_initialized() and dist.get_world_size() > 1:
-                    current_vace_hint = torch.chunk(current_vace_hint, get_sequence_parallel_world_size(), dim=1)[get_sequence_parallel_rank()]
-                    current_vace_hint = torch.nn.functional.pad(current_vace_hint, (0, 0, 0, chunks[0].shape[1] - current_vace_hint.shape[1]), value=0)
                 x = x + current_vace_hint * vace_scale
             
             # Animate

diffsynth/pipelines/z_image.py

@@ -54,6 +54,7 @@ class ZImagePipeline(BasePipeline):
             ZImageUnit_PAIControlNet(),
         ]
         self.model_fn = model_fn_z_image
+        self.compilable_models = ["dit"]
     
     
     @staticmethod

diffsynth/utils/xfuser/__init__.py

@@ -1 +1 @@
-from .xdit_context_parallel import usp_attn_forward, usp_dit_forward, get_sequence_parallel_world_size, initialize_usp, get_current_chunk, gather_all_chunks
+from .xdit_context_parallel import usp_attn_forward, usp_dit_forward, usp_vace_forward, get_sequence_parallel_world_size, initialize_usp, get_current_chunk, gather_all_chunks

diffsynth/utils/xfuser/xdit_context_parallel.py

@@ -117,6 +117,39 @@ def usp_dit_forward(self,
     return x
 
 
+def usp_vace_forward(
+    self, x, vace_context, context, t_mod, freqs,
+    use_gradient_checkpointing: bool = False,
+    use_gradient_checkpointing_offload: bool = False,
+):
+    # Compute full sequence length from the sharded x
+    full_seq_len = x.shape[1] * get_sequence_parallel_world_size()
+
+    # Embed vace_context via patch embedding
+    c = [self.vace_patch_embedding(u.unsqueeze(0)) for u in vace_context]
+    c = [u.flatten(2).transpose(1, 2) for u in c]
+    c = torch.cat([
+        torch.cat([u, u.new_zeros(1, full_seq_len - u.size(1), u.size(2))],
+                  dim=1) for u in c
+    ])
+
+    # Chunk VACE context along sequence dim BEFORE processing through blocks
+    c = torch.chunk(c, get_sequence_parallel_world_size(), dim=1)[get_sequence_parallel_rank()]
+
+    # Process through vace_blocks (self_attn already monkey-patched to usp_attn_forward)
+    for block in self.vace_blocks:
+        c = gradient_checkpoint_forward(
+            block,
+            use_gradient_checkpointing,
+            use_gradient_checkpointing_offload,
+            c, x, context, t_mod, freqs
+        )
+
+    # Hints are already sharded per-rank
+    hints = torch.unbind(c)[:-1]
+    return hints
+
+
 def usp_attn_forward(self, x, freqs):
     q = self.norm_q(self.q(x))
     k = self.norm_k(self.k(x))

docs/en/Model_Details/LTX-2.md

@@ -16,7 +16,7 @@ For more information about installation, please refer to [Installation Dependenc
 
 ## Quick Start
 
-Run the following code to quickly load the [Lightricks/LTX-2](https://www.modelscope.cn/models/Lightricks/LTX-2) model and perform inference. VRAM management has been enabled, and the framework will automatically control model parameter loading based on remaining VRAM. It can run with a minimum of 8GB VRAM.
+Run the following code to quickly load the [Lightricks/LTX-2.3](https://www.modelscope.cn/models/Lightricks/LTX-2.3) model and perform inference. VRAM management has been enabled, and the framework will automatically control model parameter loading based on remaining VRAM. It can run with a minimum of 8GB VRAM.
 
 ```python
 import torch
@@ -24,88 +24,36 @@ from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelCo
 from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
 
 vram_config = {
-    "offload_dtype": torch.float8_e5m2,
+    "offload_dtype": torch.bfloat16,
     "offload_device": "cpu",
-    "onload_dtype": torch.float8_e5m2,
-    "onload_device": "cpu",
-    "preparing_dtype": torch.float8_e5m2,
+    "onload_dtype": torch.bfloat16,
+    "onload_device": "cuda",
+    "preparing_dtype": torch.bfloat16,
     "preparing_device": "cuda",
     "computation_dtype": torch.bfloat16,
     "computation_device": "cuda",
 }
-"""
-Offical model repo: https://www.modelscope.cn/models/Lightricks/LTX-2
-Repackaged model repo: https://www.modelscope.cn/models/DiffSynth-Studio/LTX-2-Repackage
-For base models of LTX-2, offical checkpoint (with model config ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors"))
-and repackaged checkpoints (with model config ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="*.safetensors")) are both supported.
-We have repackeged the official checkpoints in DiffSynth-Studio/LTX-2-Repackage repo to support separate loading of different submodules,
-and avoid redundant memory usage when users only want to use part of the model.
-"""
-# use the repackaged modelconfig from "DiffSynth-Studio/LTX-2-Repackage" to avoid redundant model loading
 pipe = LTX2AudioVideoPipeline.from_pretrained(
     torch_dtype=torch.bfloat16,
     device="cuda",
     model_configs=[
         ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="transformer.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="text_encoder_post_modules.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="video_vae_decoder.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="audio_vae_decoder.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="audio_vocoder.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="video_vae_encoder.safetensors", **vram_config),
-        ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
+        ModelConfig(model_id="Lightricks/LTX-2.3", origin_file_pattern="ltx-2.3-22b-dev.safetensors", **vram_config),
+        ModelConfig(model_id="Lightricks/LTX-2.3", origin_file_pattern="ltx-2.3-spatial-upscaler-x2-1.0.safetensors", **vram_config),
     ],
     tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
-    stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
-    vram_limit=torch.cuda.mem_get_info("cuda")[1] / (1024 ** 3) - 0.5,
+    stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2.3", origin_file_pattern="ltx-2.3-22b-distilled-lora-384.safetensors"),
 )
-
-# use the following modelconfig if you want to initialize model from offical checkpoints from "Lightricks/LTX-2"
-# pipe = LTX2AudioVideoPipeline.from_pretrained(
-#     torch_dtype=torch.bfloat16,
-#     device="cuda",
-#     model_configs=[
-#         ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
-#         ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
-#         ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
-#     ],
-#     tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
-#     stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
-#     vram_limit=torch.cuda.mem_get_info("cuda")[1] / (1024 ** 3) - 0.5,
-# )
-
-prompt = "A girl is very happy, she is speaking: \"I enjoy working with Diffsynth-Studio, it's a perfect framework.\""
-negative_prompt = (
-    "blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
-    "grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
-    "deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
-    "wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
-    "field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
-    "lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
-    "valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
-    "mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
-    "off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
-    "pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
-    "inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
-)
-height, width, num_frames = 512 * 2, 768 * 2, 121
+prompt = "Two cute orange cats, wearing boxing gloves, stand in a boxing ring and fight each other. They are punching each other fast and yelling: 'I will win!'"
+negative_prompt = pipe.default_negative_prompt["LTX-2.3"]
 video, audio = pipe(
     prompt=prompt,
     negative_prompt=negative_prompt,
     seed=43,
-    height=height,
-    width=width,
-    num_frames=num_frames,
-    tiled=True,
-    use_two_stage_pipeline=True,
-)
-write_video_audio_ltx2(
-    video=video,
-    audio=audio,
-    output_path='ltx2_twostage.mp4',
-    fps=24,
-    audio_sample_rate=24000,
+    height=1024, width=1536, num_frames=121,
+    tiled=True, use_two_stage_pipeline=True,
 )
+write_video_audio_ltx2(video=video, audio=audio, output_path='video.mp4', fps=24, audio_sample_rate=pipe.audio_vocoder.output_sampling_rate)
 ```
 
 ## Model Overview

docs/zh/Model_Details/LTX-2.md

@@ -16,7 +16,7 @@ pip install -e .
 
 ## 快速开始
 
-运行以下代码可以快速加载 [Lightricks/LTX-2](https://www.modelscope.cn/models/Lightricks/LTX-2) 模型并进行推理。显存管理已启动，框架会自动根据剩余显存控制模型参数的加载，最低 8GB 显存即可运行。
+运行以下代码可以快速加载 [Lightricks/LTX-2.3](https://www.modelscope.cn/models/Lightricks/LTX-2.3) 模型并进行推理。显存管理已启动，框架会自动根据剩余显存控制模型参数的加载，最低 8GB 显存即可运行。
 
 ```python
 import torch
@@ -24,88 +24,36 @@ from diffsynth.pipelines.ltx2_audio_video import LTX2AudioVideoPipeline, ModelCo
 from diffsynth.utils.data.media_io_ltx2 import write_video_audio_ltx2
 
 vram_config = {
-    "offload_dtype": torch.float8_e5m2,
+    "offload_dtype": torch.bfloat16,
     "offload_device": "cpu",
-    "onload_dtype": torch.float8_e5m2,
-    "onload_device": "cpu",
-    "preparing_dtype": torch.float8_e5m2,
+    "onload_dtype": torch.bfloat16,
+    "onload_device": "cuda",
+    "preparing_dtype": torch.bfloat16,
     "preparing_device": "cuda",
     "computation_dtype": torch.bfloat16,
     "computation_device": "cuda",
 }
-"""
-Offical model repo: https://www.modelscope.cn/models/Lightricks/LTX-2
-Repackaged model repo: https://www.modelscope.cn/models/DiffSynth-Studio/LTX-2-Repackage
-For base models of LTX-2, offical checkpoint (with model config ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors"))
-and repackaged checkpoints (with model config ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="*.safetensors")) are both supported.
-We have repackeged the official checkpoints in DiffSynth-Studio/LTX-2-Repackage repo to support separate loading of different submodules,
-and avoid redundant memory usage when users only want to use part of the model.
-"""
-# use the repackaged modelconfig from "DiffSynth-Studio/LTX-2-Repackage" to avoid redundant model loading
 pipe = LTX2AudioVideoPipeline.from_pretrained(
     torch_dtype=torch.bfloat16,
     device="cuda",
     model_configs=[
         ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="transformer.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="text_encoder_post_modules.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="video_vae_decoder.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="audio_vae_decoder.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="audio_vocoder.safetensors", **vram_config),
-        ModelConfig(model_id="DiffSynth-Studio/LTX-2-Repackage", origin_file_pattern="video_vae_encoder.safetensors", **vram_config),
-        ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
+        ModelConfig(model_id="Lightricks/LTX-2.3", origin_file_pattern="ltx-2.3-22b-dev.safetensors", **vram_config),
+        ModelConfig(model_id="Lightricks/LTX-2.3", origin_file_pattern="ltx-2.3-spatial-upscaler-x2-1.0.safetensors", **vram_config),
     ],
     tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
-    stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
-    vram_limit=torch.cuda.mem_get_info("cuda")[1] / (1024 ** 3) - 0.5,
+    stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2.3", origin_file_pattern="ltx-2.3-22b-distilled-lora-384.safetensors"),
 )
-
-# use the following modelconfig if you want to initialize model from offical checkpoints from "Lightricks/LTX-2"
-# pipe = LTX2AudioVideoPipeline.from_pretrained(
-#     torch_dtype=torch.bfloat16,
-#     device="cuda",
-#     model_configs=[
-#         ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized", origin_file_pattern="model-*.safetensors", **vram_config),
-#         ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-dev.safetensors", **vram_config),
-#         ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-spatial-upscaler-x2-1.0.safetensors", **vram_config),
-#     ],
-#     tokenizer_config=ModelConfig(model_id="google/gemma-3-12b-it-qat-q4_0-unquantized"),
-#     stage2_lora_config=ModelConfig(model_id="Lightricks/LTX-2", origin_file_pattern="ltx-2-19b-distilled-lora-384.safetensors"),
-#     vram_limit=torch.cuda.mem_get_info("cuda")[1] / (1024 ** 3) - 0.5,
-# )
-
-prompt = "A girl is very happy, she is speaking: \"I enjoy working with Diffsynth-Studio, it's a perfect framework.\""
-negative_prompt = (
-    "blurry, out of focus, overexposed, underexposed, low contrast, washed out colors, excessive noise, "
-    "grainy texture, poor lighting, flickering, motion blur, distorted proportions, unnatural skin tones, "
-    "deformed facial features, asymmetrical face, missing facial features, extra limbs, disfigured hands, "
-    "wrong hand count, artifacts around text, inconsistent perspective, camera shake, incorrect depth of "
-    "field, background too sharp, background clutter, distracting reflections, harsh shadows, inconsistent "
-    "lighting direction, color banding, cartoonish rendering, 3D CGI look, unrealistic materials, uncanny "
-    "valley effect, incorrect ethnicity, wrong gender, exaggerated expressions, wrong gaze direction, "
-    "mismatched lip sync, silent or muted audio, distorted voice, robotic voice, echo, background noise, "
-    "off-sync audio, incorrect dialogue, added dialogue, repetitive speech, jittery movement, awkward "
-    "pauses, incorrect timing, unnatural transitions, inconsistent framing, tilted camera, flat lighting, "
-    "inconsistent tone, cinematic oversaturation, stylized filters, or AI artifacts."
-)
-height, width, num_frames = 512 * 2, 768 * 2, 121
+prompt = "Two cute orange cats, wearing boxing gloves, stand in a boxing ring and fight each other. They are punching each other fast and yelling: 'I will win!'"
+negative_prompt = pipe.default_negative_prompt["LTX-2.3"]
 video, audio = pipe(
     prompt=prompt,
     negative_prompt=negative_prompt,
     seed=43,
-    height=height,
-    width=width,
-    num_frames=num_frames,
-    tiled=True,
-    use_two_stage_pipeline=True,
-)
-write_video_audio_ltx2(
-    video=video,
-    audio=audio,
-    output_path='ltx2_twostage.mp4',
-    fps=24,
-    audio_sample_rate=24000,
+    height=1024, width=1536, num_frames=121,
+    tiled=True, use_two_stage_pipeline=True,
 )
+write_video_audio_ltx2(video=video, audio=audio, output_path='video.mp4', fps=24, audio_sample_rate=pipe.audio_vocoder.output_sampling_rate)
 ```
 
 ## 模型总览

examples/flux2/model_inference/FLUX.2-dev.py

@@ -1,6 +1,6 @@
 from diffsynth.pipelines.flux2_image import Flux2ImagePipeline, ModelConfig
 import torch
-
+from PIL import Image
 
 vram_config = {
     "offload_dtype": torch.bfloat16,
@@ -25,3 +25,8 @@ pipe = Flux2ImagePipeline.from_pretrained(
 prompt = "Realistic macro photograph of a hermit crab using a soda can as its shell, partially emerging from the can, captured with sharp detail and natural colors, on a sunlit beach with soft shadows and a shallow depth of field, with blurred ocean waves in the background. The can has the text `BFL Diffusers` on it and it has a color gradient that start with #FF5733 at the top and transitions to #33FF57 at the bottom."
 image = pipe(prompt, seed=42, rand_device="cuda", num_inference_steps=50)
 image.save("image_FLUX.2-dev.jpg")
+
+prompt  = "Transform the image into Japanese anime style"
+edit_image = [Image.open("image_FLUX.2-dev.jpg")]
+image = pipe(prompt, seed=42, rand_device="cuda", edit_image=edit_image, num_inference_steps=50, embedded_guidance=2.5)
+image.save("image_FLUX.2-dev_edit.jpg")

examples/flux2/model_inference_low_vram/FLUX.2-dev.py

@@ -1,5 +1,6 @@
 from diffsynth.pipelines.flux2_image import Flux2ImagePipeline, ModelConfig
 import torch
+from PIL import Image
 
 vram_config = {
     "offload_dtype": "disk",
@@ -25,3 +26,8 @@ pipe = Flux2ImagePipeline.from_pretrained(
 prompt = "High resolution. A dreamy underwater portrait of a serene young woman in a flowing blue dress. Her hair floats softly around her face, strands delicately suspended in the water. Clear, shimmering light filters through, casting gentle highlights, while tiny bubbles rise around her. Her expression is calm, her features finely detailed—creating a tranquil, ethereal scene."
 image = pipe(prompt, seed=42, rand_device="cuda", num_inference_steps=50)
 image.save("image.jpg")
+
+prompt  = "Transform the image into Japanese anime style"
+edit_image = [Image.open("image.jpg")]
+image = pipe(prompt, seed=42, rand_device="cuda", edit_image=edit_image, num_inference_steps=50, embedded_guidance=2.5)
+image.save("image_edit.jpg")

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "diffsynth"
-version = "2.0.6"
+version = "2.0.7"
 description = "Enjoy the magic of Diffusion models!"
 authors = [{name = "ModelScope Team"}]
 license = {text = "Apache-2.0"}