877 lines
29 KiB
Python
877 lines
29 KiB
Python
import os
|
|
import torch
|
|
import gc
|
|
from typing import cast, Any
|
|
import logging
|
|
import json
|
|
import numpy as np
|
|
import comfy.model_management as mm
|
|
from comfy.utils import ProgressBar
|
|
from pathlib import Path
|
|
|
|
# import folder_paths
|
|
from tqdm import tqdm
|
|
from easydict import EasyDict
|
|
|
|
|
|
# Import LightX2V modules
|
|
from .lightx2v.lightx2v.utils.profiler import ProfilingContext
|
|
from .lightx2v.lightx2v.models.input_encoders.hf.t5.model import T5EncoderModel
|
|
from .lightx2v.lightx2v.models.input_encoders.hf.xlm_roberta.model import (
|
|
CLIPModel as ClipVisionModel,
|
|
)
|
|
from .lightx2v.lightx2v.models.video_encoders.hf.wan.vae import WanVAE
|
|
from .lightx2v.lightx2v.models.networks.wan.model import WanModel
|
|
from .lightx2v.lightx2v.models.networks.wan.lora_adapter import WanLoraWrapper
|
|
from .lightx2v.lightx2v.models.schedulers.wan.scheduler import WanScheduler
|
|
from .lightx2v.lightx2v.models.schedulers.wan.feature_caching.scheduler import (
|
|
WanSchedulerTeaCaching,
|
|
)
|
|
|
|
from .lightx2v.lightx2v.common.ops import * # noqa: F401, F403 for import global register
|
|
|
|
|
|
class WanVideoTeaCache:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"rel_l1_thresh": (
|
|
"FLOAT",
|
|
{
|
|
"default": 0.275,
|
|
"min": 0.0,
|
|
"max": 10.0,
|
|
"step": 0.001,
|
|
"tooltip": "Threshold for to determine when to apply the cache, compromise between speed and accuracy. When using coefficients a good value range is something between 0.2-0.4 for all but 1.3B model, which should be about 10 times smaller, same as when not using coefficients.",
|
|
},
|
|
),
|
|
"start_percent": (
|
|
"FLOAT",
|
|
{
|
|
"default": 0.1,
|
|
"min": 0.0,
|
|
"max": 1.0,
|
|
"step": 0.01,
|
|
"tooltip": "The start percentage of the steps to use with TeaCache.",
|
|
},
|
|
),
|
|
"end_percent": (
|
|
"FLOAT",
|
|
{
|
|
"default": 1.0,
|
|
"min": 0.0,
|
|
"max": 1.0,
|
|
"step": 0.01,
|
|
"tooltip": "The end percentage of the steps to use with TeaCache.",
|
|
},
|
|
),
|
|
"cache_device": (
|
|
["main_device", "offload_device"],
|
|
{"default": "offload_device", "tooltip": "Device to cache to"},
|
|
),
|
|
"coefficients": (
|
|
[
|
|
"i2v-14B-720p",
|
|
"i2v-14B-480p",
|
|
"1.3B",
|
|
"14B",
|
|
"disabled",
|
|
],
|
|
{
|
|
"default": "i2v-14B-720p",
|
|
"tooltip": "Use coefficients for TeaCache. 'i2v-14B-720p' will use the default coefficients, 'disabled' will disable coefficients.",
|
|
},
|
|
),
|
|
},
|
|
"optional": {
|
|
"mode": (
|
|
["e", "e0"],
|
|
{
|
|
"default": "e",
|
|
"tooltip": "Choice between using e (time embeds, default) or e0 (modulated time embeds)",
|
|
},
|
|
),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_TEACACHEARGS",)
|
|
RETURN_NAMES = ("teacache_args",)
|
|
FUNCTION = "process"
|
|
CATEGORY = "LightX2V"
|
|
|
|
EXPERIMENTAL = True
|
|
|
|
def process(
|
|
self,
|
|
rel_l1_thresh,
|
|
start_step,
|
|
end_step,
|
|
cache_device,
|
|
coefficients,
|
|
mode="e",
|
|
):
|
|
if cache_device == "main_device":
|
|
teacache_device = mm.get_torch_device()
|
|
else:
|
|
teacache_device = mm.unet_offload_device()
|
|
teacache_args = {
|
|
"rel_l1_thresh": rel_l1_thresh,
|
|
"start_step": start_step,
|
|
"end_step": end_step,
|
|
"cache_device": teacache_device,
|
|
"coefficients": coefficients,
|
|
"mode": mode,
|
|
}
|
|
return (teacache_args,)
|
|
|
|
|
|
class Lightx2vWanVideoT5EncoderLoader:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"t5_model_path": (
|
|
"STRING",
|
|
{
|
|
"default": "/mnt/aigc/users/lijiaqi2/wan_model/Wan2.1-I2V-14B-480P/models_t5_umt5-xxl-enc-bf16.pth"
|
|
},
|
|
),
|
|
"tokenizer_path": (
|
|
"STRING",
|
|
{
|
|
"default": "/mnt/aigc/users/lijiaqi2/wan_model/Wan2.1-I2V-14B-480P/google/umt5-xxl"
|
|
},
|
|
),
|
|
"text_len": (
|
|
"INT",
|
|
{"default": 512, "min": 64, "max": 2048, "step": 1},
|
|
),
|
|
"precision": (["bf16", "fp16", "fp32"], {"default": "bf16"}),
|
|
"device": (["cuda", "cpu"], {"default": "cuda"}),
|
|
"additional_param": ("STRING", {"default": "default_value"}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_T5_ENCODER",)
|
|
RETURN_NAMES = ("t5_encoder",)
|
|
FUNCTION = "load_t5_encoder"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def load_t5_encoder(
|
|
self,
|
|
t5_model_path,
|
|
tokenizer_path,
|
|
text_len,
|
|
precision,
|
|
device,
|
|
additional_param,
|
|
):
|
|
# Map precision to torch dtype
|
|
dtype_map = {
|
|
"bf16": torch.bfloat16,
|
|
"fp16": torch.float16,
|
|
"fp32": torch.float32,
|
|
}
|
|
dtype = dtype_map[precision]
|
|
|
|
# Resolve device
|
|
if device == "cuda":
|
|
device = mm.get_torch_device()
|
|
else:
|
|
device = torch.device("cpu")
|
|
|
|
# Load the T5 encoder
|
|
t5_encoder = T5EncoderModel(
|
|
text_len=text_len,
|
|
dtype=dtype,
|
|
device=device, # type:ignore
|
|
checkpoint_path=t5_model_path,
|
|
tokenizer_path=tokenizer_path,
|
|
shard_fn=None,
|
|
)
|
|
|
|
return (t5_encoder,)
|
|
|
|
|
|
class Lightx2vWanVideoT5Encoder:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"t5_encoder": ("LIGHT_T5_ENCODER",),
|
|
"prompt": (
|
|
"STRING",
|
|
{
|
|
"multiline": True,
|
|
"default": "Summer beach vacation style, a white cat wearing sunglasses sits on a surfboard. The fluffy-furred feline gazes directly at the camera with a relaxed expression. Blurred beach scenery forms the background featuring crystal-clear waters, distant green hills, and a blue sky dotted with white clouds. The cat assumes a naturally relaxed posture, as if savoring the sea breeze and warm sunlight. A close-up shot highlights the feline's intricate details and the refreshing atmosphere of the seaside.",
|
|
},
|
|
),
|
|
"negative_prompt": (
|
|
"STRING",
|
|
{
|
|
"multiline": True,
|
|
"default": "色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
|
|
},
|
|
),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_TEXT_EMBEDDINGS",)
|
|
RETURN_NAMES = ("text_embeddings",)
|
|
FUNCTION = "encode_text"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def encode_text(self, t5_encoder, prompt, negative_prompt):
|
|
# Create a config object with required attributes
|
|
class Config:
|
|
def __init__(self, cpu_offload=False):
|
|
self.cpu_offload = cpu_offload
|
|
|
|
# NOTE(xxx): adapt the config if cpu_offload is set, t5 model must be on cuda device
|
|
# Encode the text
|
|
context = t5_encoder.infer([prompt])
|
|
context_null = t5_encoder.infer([negative_prompt if negative_prompt else ""])
|
|
|
|
# Create text embeddings dictionary
|
|
text_embeddings = {"context": context, "context_null": context_null}
|
|
|
|
print(f"Text Encoder Output Shape: {context[0].shape}")
|
|
|
|
return (text_embeddings,)
|
|
|
|
|
|
class Lightx2vWanVideoVaeLoader:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"vae_model_path": (
|
|
"STRING",
|
|
{
|
|
"default": "/mnt/aigc/users/lijiaqi2/wan_model/Wan2.1-I2V-14B-480P/Wan2.1_VAE.pth"
|
|
},
|
|
),
|
|
"precision": (["bf16", "fp16", "fp32"], {"default": "fp16"}),
|
|
"device": (["cuda", "cpu"], {"default": "cuda"}),
|
|
"parallel": ("BOOLEAN", {"default": False}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_WAN_VAE",)
|
|
RETURN_NAMES = ("wan_vae",)
|
|
FUNCTION = "load_vae"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def load_vae(self, vae_model_path, precision, device, parallel):
|
|
# Map precision to torch dtype
|
|
dtype_map = {
|
|
"bf16": torch.bfloat16,
|
|
"fp16": torch.float16,
|
|
"fp32": torch.float32,
|
|
}
|
|
dtype = dtype_map[precision]
|
|
|
|
# Resolve device
|
|
if device == "cuda":
|
|
device = mm.get_torch_device()
|
|
else:
|
|
device = torch.device("cpu")
|
|
|
|
# Load the VAE
|
|
vae = WanVAE(
|
|
z_dim=16,
|
|
vae_pth=vae_model_path,
|
|
dtype=dtype,
|
|
device=device, # type:ignore
|
|
parallel=parallel,
|
|
)
|
|
|
|
return (vae,)
|
|
|
|
|
|
class Lightx2vWanVideoVaeDecoder:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"wan_vae": ("LIGHT_WAN_VAE",),
|
|
"latent": ("LIGHT_LATENT",),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("IMAGE",)
|
|
RETURN_NAMES = ("images",)
|
|
FUNCTION = "decode_latent"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def decode_latent(self, wan_vae, latent):
|
|
config = EasyDict({"cpu_offload": False})
|
|
|
|
# 获取潜在表示和生成器
|
|
latents = latent["samples"]
|
|
generator = latent["generator"]
|
|
|
|
# 使用VAE解码潜在表示
|
|
with torch.no_grad():
|
|
# 解码得到视频帧
|
|
decoded_images = wan_vae.decode(latents, generator=generator, config=config)
|
|
|
|
# 将像素值从 [-1, 1] 归一化到 [0, 1]
|
|
images = (decoded_images + 1) / 2
|
|
|
|
# 重新排列维度为ComfyUI标准的图像格式 [T, H, W, C]
|
|
# 从 [1, C, T, H, W] 转换为 [T, H, W, C]
|
|
images = images.squeeze(0).permute(1, 2, 3, 0).cpu()
|
|
|
|
# 确保像素值在有效范围内
|
|
images = torch.clamp(images, 0, 1)
|
|
|
|
# 清理缓存以释放GPU内存
|
|
torch.cuda.empty_cache()
|
|
gc.collect()
|
|
|
|
return (images,)
|
|
|
|
|
|
class Lightx2vWanVideoClipVisionEncoderLoader:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"clip_model_path": (
|
|
"STRING",
|
|
{
|
|
"default": "/mnt/aigc/users/lijiaqi2/wan_model/Wan2.1-I2V-14B-480P/models_clip_open-clip-xlm-roberta-large-vit-huge-14.pth"
|
|
},
|
|
),
|
|
"tokenizer_path": (
|
|
"STRING",
|
|
{
|
|
"default": "/mnt/aigc/users/lijiaqi2/wan_model/Wan2.1-I2V-14B-480P/xlm-roberta-large"
|
|
},
|
|
),
|
|
"precision": (["fp16", "fp32"], {"default": "fp16"}),
|
|
"device": (["cuda", "cpu"], {"default": "cuda"}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_CLIP_VISION_ENCODER",)
|
|
RETURN_NAMES = ("clip_vision_encoder",)
|
|
FUNCTION = "load_clip_vision_encoder"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def load_clip_vision_encoder(
|
|
self, clip_model_path, tokenizer_path, precision, device
|
|
):
|
|
# Map precision to torch dtype
|
|
dtype_map = {"fp16": torch.float16, "fp32": torch.float32}
|
|
dtype = dtype_map[precision]
|
|
|
|
# Resolve device
|
|
if device == "cuda":
|
|
device = mm.get_torch_device()
|
|
else:
|
|
device = torch.device("cpu")
|
|
|
|
# Load the CLIP vision encoder
|
|
clip_vision_encoder = ClipVisionModel(
|
|
dtype=dtype,
|
|
device=device,
|
|
checkpoint_path=clip_model_path,
|
|
tokenizer_path=tokenizer_path,
|
|
)
|
|
|
|
return (clip_vision_encoder,)
|
|
|
|
|
|
class Lightx2vWanVideoImageEncoder:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"vae": ("LIGHT_WAN_VAE",),
|
|
"clip_vision_encoder": ("LIGHT_CLIP_VISION_ENCODER",),
|
|
"image": ("IMAGE",),
|
|
"width": (
|
|
"INT",
|
|
{
|
|
"default": 832,
|
|
"min": 64,
|
|
"max": 2048,
|
|
"step": 8,
|
|
"tooltip": "Width of the image to encode",
|
|
},
|
|
),
|
|
"height": (
|
|
"INT",
|
|
{
|
|
"default": 480,
|
|
"min": 64,
|
|
"max": 29048,
|
|
"step": 8,
|
|
"tooltip": "Height of the image to encode",
|
|
},
|
|
),
|
|
"num_frames": (
|
|
"INT",
|
|
{
|
|
"default": 81,
|
|
"min": 1,
|
|
"max": 10000,
|
|
"step": 4,
|
|
"tooltip": "Number of frames to encode",
|
|
},
|
|
),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_IMAGE_EMBEDDINGS",)
|
|
RETURN_NAMES = ("image_embeddings",)
|
|
FUNCTION = "encode_image"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def encode_image(
|
|
self,
|
|
vae: WanVAE,
|
|
image,
|
|
clip_vision_encoder: ClipVisionModel,
|
|
height,
|
|
width,
|
|
num_frames,
|
|
):
|
|
# 创建配置对象
|
|
|
|
config = EasyDict(
|
|
{
|
|
"cpu_offload": False,
|
|
"target_height": height,
|
|
"target_width": width,
|
|
"target_video_length": num_frames,
|
|
"vae_stride": (4, 8, 8),
|
|
"patch_size": (1, 2, 2),
|
|
}
|
|
)
|
|
# skip lint
|
|
config = cast(Any, config)
|
|
|
|
# 将图像转换为期望的张量格式
|
|
device = mm.get_torch_device()
|
|
img = image[0].permute(2, 0, 1).to(device) # [C, H, W]
|
|
img = img.sub_(0.5).div_(0.5) # 归一化到 [-1, 1]
|
|
|
|
# 使用CLIP视觉编码器编码图像
|
|
clip_encoder_out = (
|
|
clip_vision_encoder.visual([img[:, None, :, :]], config)
|
|
.squeeze(0)
|
|
.to(torch.bfloat16)
|
|
)
|
|
|
|
# 计算宽高比和尺寸
|
|
h, w = img.shape[1:]
|
|
aspect_ratio = h / w
|
|
max_area = config.target_height * config.target_width
|
|
lat_h = round(
|
|
np.sqrt(max_area * aspect_ratio)
|
|
// config.vae_stride[1]
|
|
// config.patch_size[1]
|
|
* config.patch_size[1]
|
|
)
|
|
lat_w = round(
|
|
np.sqrt(max_area / aspect_ratio)
|
|
// config.vae_stride[2]
|
|
// config.patch_size[2]
|
|
* config.patch_size[2]
|
|
)
|
|
|
|
# XXX: trick
|
|
config.lat_h = lat_h
|
|
config.lat_w = lat_w
|
|
|
|
h = lat_h * config.vae_stride[1]
|
|
w = lat_w * config.vae_stride[2]
|
|
|
|
msk = torch.ones(
|
|
1, config.target_video_length, lat_h, lat_w, device=torch.device("cuda")
|
|
)
|
|
msk[:, 1:] = 0
|
|
msk = torch.concat(
|
|
[torch.repeat_interleave(msk[:, 0:1], repeats=4, dim=1), msk[:, 1:]], dim=1
|
|
)
|
|
msk = msk.view(1, msk.shape[1] // 4, 4, lat_h, lat_w)
|
|
msk = msk.transpose(1, 2)[0]
|
|
vae_encode_out = vae.encode(
|
|
[
|
|
torch.concat(
|
|
[
|
|
torch.nn.functional.interpolate(
|
|
img[None].cpu(), size=(h, w), mode="bicubic"
|
|
).transpose(0, 1),
|
|
torch.zeros(3, config.target_video_length - 1, h, w),
|
|
],
|
|
dim=1,
|
|
).cuda()
|
|
],
|
|
config,
|
|
)[0]
|
|
# TODO(xxx): hard code
|
|
vae_encode_out = torch.concat([msk, vae_encode_out]).to(torch.bfloat16)
|
|
|
|
image_embeddings = {
|
|
"clip_encoder_out": clip_encoder_out,
|
|
"vae_encode_out": vae_encode_out,
|
|
"config": config,
|
|
}
|
|
|
|
print(f"Image Encoder Output Shape: {clip_encoder_out.shape}")
|
|
print(f"VAE Encoder Output Shape: {vae_encode_out.shape}")
|
|
print(f"Latent Height: {lat_h}, Latent Width: {lat_w}")
|
|
print(f"Image Shape: {img.shape}")
|
|
print(f"Configuration: {config}")
|
|
|
|
return (image_embeddings,)
|
|
|
|
|
|
class Lightx2vWanVideoEmptyEmbeds:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"width": (
|
|
"INT",
|
|
{
|
|
"default": 832,
|
|
"min": 64,
|
|
"max": 2048,
|
|
"step": 8,
|
|
"tooltip": "Width of the image to encode",
|
|
},
|
|
),
|
|
"height": (
|
|
"INT",
|
|
{
|
|
"default": 480,
|
|
"min": 64,
|
|
"max": 29048,
|
|
"step": 8,
|
|
"tooltip": "Height of the image to encode",
|
|
},
|
|
),
|
|
"num_frames": (
|
|
"INT",
|
|
{
|
|
"default": 81,
|
|
"min": 1,
|
|
"max": 10000,
|
|
"step": 4,
|
|
"tooltip": "Number of frames to encode",
|
|
},
|
|
),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_IMAGE_EMBEDDINGS",)
|
|
RETURN_NAMES = ("image_embeddings",)
|
|
FUNCTION = "process"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def process(self, num_frames, width, height, control_embeds=None):
|
|
config = EasyDict(
|
|
{
|
|
"target_height": height,
|
|
"target_width": width,
|
|
"target_video_length": num_frames,
|
|
"vae_stride": (4, 8, 8),
|
|
"patch_size": (1, 2, 2),
|
|
}
|
|
)
|
|
|
|
return ({"config": config},)
|
|
|
|
|
|
class Lightx2vWanVideoModelLoader:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"model_path": (
|
|
"STRING",
|
|
{
|
|
"default": "/mnt/aigc/users/lijiaqi2/wan_model/Wan2.1-I2V-14B-480P"
|
|
},
|
|
),
|
|
"model_type": (["t2v", "i2v"], {"default": "i2v"}),
|
|
"precision": (["bf16", "fp16", "fp32"], {"default": "bf16"}),
|
|
"device": (["cuda", "cpu"], {"default": "cuda"}),
|
|
"attention_type": (
|
|
["sdpa", "flash_attn2", "flash_attn3"],
|
|
{"default": "flash_attn3"},
|
|
),
|
|
"cpu_offload": ("BOOLEAN", {"default": False}),
|
|
"mm_type": ("STRING", {"default": "Default"}),
|
|
},
|
|
"optional": {
|
|
"teacache_args": ("LIGHT_TEACACHEARGS", {"default": None}),
|
|
"lora_path": ("STRING", {"default": None}),
|
|
"lora_strength": (
|
|
"FLOAT",
|
|
{"default": 1.0, "min": 0.0, "max": 2.0, "step": 0.01},
|
|
),
|
|
},
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_WAN_MODEL",)
|
|
RETURN_NAMES = ("wan_model",)
|
|
FUNCTION = "load_model"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def load_model(
|
|
self,
|
|
model_path,
|
|
model_type,
|
|
precision,
|
|
device,
|
|
attention_type,
|
|
mm_type,
|
|
lora_path=None,
|
|
lora_strength=1.0,
|
|
cpu_offload=False,
|
|
teacache_args=None,
|
|
):
|
|
# 映射精度到torch dtype
|
|
dtype_map = {
|
|
"bf16": torch.bfloat16,
|
|
"fp16": torch.float16,
|
|
"fp32": torch.float32,
|
|
}
|
|
dtype = dtype_map[precision]
|
|
|
|
# 解析设备
|
|
if device == "cuda":
|
|
device = mm.get_torch_device()
|
|
else:
|
|
device = torch.device("cpu")
|
|
|
|
model_path = Path(model_path)
|
|
if model_path.is_file():
|
|
model_path_dir = model_path.parent
|
|
else:
|
|
model_path_dir = model_path
|
|
# TODO(xxx）：每个模型config是固定的，可以配置，直接选取默认值
|
|
config_json_path = model_path_dir / "config.json"
|
|
|
|
config_json = {}
|
|
if config_json_path.exists():
|
|
with open(config_json_path, "r") as f:
|
|
config_json = json.load(f)
|
|
else:
|
|
logging.error(f"Config file not found at {config_json_path}")
|
|
raise FileNotFoundError(f"Config file not found at {config_json_path}")
|
|
|
|
feature_caching = "Tea" if teacache_args is not None else "NoCaching"
|
|
if teacache_args:
|
|
teacache_thresh = teacache_args["rel_l1_thresh"]
|
|
else:
|
|
teacache_thresh = 0.26
|
|
|
|
# 创建配置字典
|
|
config = {
|
|
"do_mm_calib": False,
|
|
"cpu_offload": cpu_offload,
|
|
"parallel_attn_type": None, # [None, "ulysses", "ring"]
|
|
"parallel_vae": False,
|
|
"max_area": False,
|
|
"vae_stride": (4, 8, 8),
|
|
"patch_size": (1, 2, 2),
|
|
"feature_caching": feature_caching, # ["NoCaching", "TaylorSeer", "Tea"]
|
|
"teacache_thresh": teacache_thresh,
|
|
"use_ret_steps": False,
|
|
"use_bfloat16": dtype == torch.bfloat16,
|
|
"mm_config": {
|
|
"mm_type": mm_type,
|
|
"weight_auto_quant": False if mm_type == "Default" else True,
|
|
},
|
|
"model_path": model_path,
|
|
"task": model_type,
|
|
"model_cls": "wan2.1",
|
|
"device": device,
|
|
"attention_type": attention_type,
|
|
"lora_path": lora_path if lora_path and lora_path.strip() else None,
|
|
"strength_model": lora_strength,
|
|
}
|
|
# merge model dir config.json and config dict
|
|
config.update(**config_json)
|
|
# NOTE(xxx): adapt to Lightx2v
|
|
config = EasyDict(config)
|
|
logging.info(f"Loaded config:\n {config}")
|
|
logging.info(f"Loading WanModel from {model_path} with type {model_type}")
|
|
model = WanModel(model_path, config, device)
|
|
|
|
# 如果指定了LoRA路径，应用LoRA
|
|
if lora_path and os.path.exists(lora_path):
|
|
logging.info(
|
|
f"Applying LoRA from {lora_path} with strength {lora_strength}"
|
|
)
|
|
lora_wrapper = WanLoraWrapper(model)
|
|
lora_name = lora_wrapper.load_lora(lora_path)
|
|
lora_wrapper.apply_lora(lora_name, lora_strength)
|
|
logging.info(f"LoRA {lora_name} applied successfully")
|
|
|
|
wan_model = {"wan_model": model, "config": config}
|
|
|
|
return (wan_model,)
|
|
|
|
|
|
class Lightx2vWanVideoSampler:
|
|
@classmethod
|
|
def INPUT_TYPES(cls):
|
|
return {
|
|
"required": {
|
|
"model": ("LIGHT_WAN_MODEL",),
|
|
"text_embeddings": ("LIGHT_TEXT_EMBEDDINGS",),
|
|
"image_embeddings": ("LIGHT_IMAGE_EMBEDDINGS",),
|
|
"steps": ("INT", {"default": 20, "min": 1, "max": 100, "step": 1}),
|
|
"shift": ("FLOAT", {"default": 5.0}),
|
|
"cfg_scale": (
|
|
"FLOAT",
|
|
{"default": 5, "min": 1, "max": 20.0, "step": 0.1},
|
|
),
|
|
"seed": ("INT", {"default": 42}),
|
|
}
|
|
}
|
|
|
|
RETURN_TYPES = ("LIGHT_LATENT",)
|
|
RETURN_NAMES = ("latent",)
|
|
FUNCTION = "sample"
|
|
CATEGORY = "LightX2V"
|
|
|
|
def sample(
|
|
self,
|
|
model,
|
|
text_embeddings,
|
|
steps,
|
|
shift,
|
|
cfg_scale,
|
|
seed,
|
|
image_embeddings,
|
|
):
|
|
# Update model config
|
|
|
|
model_config = model.get("config")
|
|
image_config = image_embeddings.get("config")
|
|
model_config.update(image_config)
|
|
model_config = cast(Any, model_config)
|
|
|
|
logging.info(f"Loaded config:\n {model_config}")
|
|
|
|
# wan model
|
|
wan_model = cast(WanModel, model.get("wan_model"))
|
|
# clip vision result
|
|
clip_encoder_out = image_embeddings.get("clip_encoder_out", None)
|
|
# text result
|
|
vae_encode_out = image_embeddings.get("vae_encode_out", None)
|
|
|
|
if model_config.task == "i2v" and (
|
|
clip_encoder_out is None or vae_encode_out is None
|
|
):
|
|
raise ValueError("clip_encoder_out must be provided for i2v task")
|
|
|
|
model_config.infer_steps = steps
|
|
model_config.sample_shift = shift
|
|
model_config.sample_guide_scale = cfg_scale
|
|
model_config.seed = seed
|
|
|
|
model_config.enable_cfg = True
|
|
model_config.offload_granularity = "block"
|
|
|
|
# wan_runner.set_target_shape
|
|
num_channels_latents = model_config.get("num_channels_latents", 16)
|
|
|
|
if model_config.task == "i2v":
|
|
model_config.target_shape = (
|
|
num_channels_latents,
|
|
(model_config.target_video_length - 1) // model_config.vae_stride[0]
|
|
+ 1,
|
|
model_config.lat_h,
|
|
model_config.lat_w,
|
|
)
|
|
elif model_config.task == "t2v":
|
|
model_config.target_shape = (
|
|
16,
|
|
(model_config.target_video_length - 1) // 4 + 1,
|
|
int(model_config.target_height) // model_config.vae_stride[1],
|
|
int(model_config.target_width) // model_config.vae_stride[2],
|
|
)
|
|
|
|
# wan_runner.init_scheduler
|
|
if model_config.feature_caching == "NoCaching":
|
|
scheduler = WanScheduler(model_config)
|
|
elif model_config.feature_caching == "Tea":
|
|
scheduler = WanSchedulerTeaCaching(model_config)
|
|
else:
|
|
raise NotImplementedError(
|
|
f"Unsupported feature_caching type: {model_config.feature_caching}" # type:ignore
|
|
)
|
|
|
|
# setup scheduler
|
|
wan_model.set_scheduler(scheduler)
|
|
|
|
# Set up inputs
|
|
inputs = {
|
|
"text_encoder_output": text_embeddings,
|
|
"image_encoder_output": image_embeddings,
|
|
}
|
|
|
|
# Prepare for sampling
|
|
scheduler.prepare(inputs.get("image_encoder_output"))
|
|
|
|
# Run sampling
|
|
progress = ProgressBar(steps)
|
|
for step_index in tqdm(
|
|
range(scheduler.infer_steps), desc="inference", unit="step"
|
|
):
|
|
with ProfilingContext("scheduler.step_pre"):
|
|
scheduler.step_pre(step_index=step_index)
|
|
with ProfilingContext("model.infer"):
|
|
wan_model.infer(inputs)
|
|
with ProfilingContext("scheduler.step_post"):
|
|
scheduler.step_post()
|
|
|
|
progress.update(1)
|
|
|
|
scheduler.clear()
|
|
|
|
del inputs, wan_model, text_embeddings, image_embeddings
|
|
torch.cuda.empty_cache()
|
|
|
|
return ({"samples": scheduler.latents, "generator": scheduler.generator},)
|
|
|
|
|
|
# Register the nodes
|
|
NODE_CLASS_MAPPINGS = {
|
|
"Lightx2vWanVideoT5EncoderLoader": Lightx2vWanVideoT5EncoderLoader,
|
|
"Lightx2vWanVideoT5Encoder": Lightx2vWanVideoT5Encoder,
|
|
"Lightx2vWanVideoClipVisionEncoderLoader": Lightx2vWanVideoClipVisionEncoderLoader,
|
|
"Lightx2vWanVideoVaeLoader": Lightx2vWanVideoVaeLoader,
|
|
"Lightx2vTeaCache": WanVideoTeaCache,
|
|
"Lightx2vWanVideoEmptyEmbeds": Lightx2vWanVideoEmptyEmbeds,
|
|
"Lightx2vWanVideoImageEncoder": Lightx2vWanVideoImageEncoder,
|
|
"Lightx2vWanVideoVaeDecoder": Lightx2vWanVideoVaeDecoder,
|
|
"Lightx2vWanVideoModelLoader": Lightx2vWanVideoModelLoader,
|
|
"Lightx2vWanVideoSampler": Lightx2vWanVideoSampler,
|
|
}
|
|
|
|
NODE_DISPLAY_NAME_MAPPINGS = {
|
|
"Lightx2vWanVideoT5EncoderLoader": "LightX2V WAN T5 Encoder Loader",
|
|
"Lightx2vWanVideoT5Encoder": "LightX2V WAN T5 Encoder",
|
|
"Lightx2vWanVideoClipVisionEncoderLoader": "LightX2V WAN CLIP Vision Encoder Loader",
|
|
"Lightx2vWanVideoClipVisionEncoder": "LightX2V WAN CLIP Vision Encoder",
|
|
"Lightx2vWanVideoVaeLoader": "LightX2V WAN VAE Loader",
|
|
"Lightx2vWanVideoImageEncoder": "LightX2V WAN Image Encoder",
|
|
"Lightx2vWanVideoVaeDecoder": "LightX2V WAN VAE Decoder",
|
|
"Lightx2vWanVideoModelLoader": "LightX2V WAN Model Loader",
|
|
"Lightx2vWanVideoSampler": "LightX2V WAN Video Sampler",
|
|
"Lightx2vTeaCache": "LightX2V WAN Tea Cache",
|
|
"Lightx2vWanVideoEmptyEmbeds": "LightX2V WAN Video Empty Embeds",
|
|
}
|