Spaces:
Runtime error
Runtime error
Remove the word "pixart" from code.
Browse files
xora/examples/image_to_video.py
CHANGED
|
@@ -3,7 +3,7 @@ from xora.models.autoencoders.causal_video_autoencoder import CausalVideoAutoenc
|
|
| 3 |
from xora.models.transformers.transformer3d import Transformer3DModel
|
| 4 |
from xora.models.transformers.symmetric_patchifier import SymmetricPatchifier
|
| 5 |
from xora.schedulers.rf import RectifiedFlowScheduler
|
| 6 |
-
from xora.pipelines.
|
| 7 |
from pathlib import Path
|
| 8 |
from transformers import T5EncoderModel, T5Tokenizer
|
| 9 |
import safetensors.torch
|
|
@@ -180,7 +180,7 @@ def main():
|
|
| 180 |
"vae": vae,
|
| 181 |
}
|
| 182 |
|
| 183 |
-
pipeline =
|
| 184 |
|
| 185 |
# Load media (video or image)
|
| 186 |
if args.video_path:
|
|
|
|
| 3 |
from xora.models.transformers.transformer3d import Transformer3DModel
|
| 4 |
from xora.models.transformers.symmetric_patchifier import SymmetricPatchifier
|
| 5 |
from xora.schedulers.rf import RectifiedFlowScheduler
|
| 6 |
+
from xora.pipelines.pipeline_xora_video import XoraVideoPipeline
|
| 7 |
from pathlib import Path
|
| 8 |
from transformers import T5EncoderModel, T5Tokenizer
|
| 9 |
import safetensors.torch
|
|
|
|
| 180 |
"vae": vae,
|
| 181 |
}
|
| 182 |
|
| 183 |
+
pipeline = XoraVideoPipeline(**submodel_dict).to("cuda")
|
| 184 |
|
| 185 |
# Load media (video or image)
|
| 186 |
if args.video_path:
|
xora/examples/text_to_video.py
CHANGED
|
@@ -3,7 +3,7 @@ from xora.models.autoencoders.causal_video_autoencoder import CausalVideoAutoenc
|
|
| 3 |
from xora.models.transformers.transformer3d import Transformer3DModel
|
| 4 |
from xora.models.transformers.symmetric_patchifier import SymmetricPatchifier
|
| 5 |
from xora.schedulers.rf import RectifiedFlowScheduler
|
| 6 |
-
from xora.pipelines.
|
| 7 |
from pathlib import Path
|
| 8 |
from transformers import T5EncoderModel, T5Tokenizer
|
| 9 |
import safetensors.torch
|
|
@@ -82,7 +82,7 @@ def main():
|
|
| 82 |
"vae": vae,
|
| 83 |
}
|
| 84 |
|
| 85 |
-
pipeline =
|
| 86 |
|
| 87 |
# Sample input
|
| 88 |
num_inference_steps = 20
|
|
|
|
| 3 |
from xora.models.transformers.transformer3d import Transformer3DModel
|
| 4 |
from xora.models.transformers.symmetric_patchifier import SymmetricPatchifier
|
| 5 |
from xora.schedulers.rf import RectifiedFlowScheduler
|
| 6 |
+
from xora.pipelines.pipeline_xora_video import XoraVideoPipeline
|
| 7 |
from pathlib import Path
|
| 8 |
from transformers import T5EncoderModel, T5Tokenizer
|
| 9 |
import safetensors.torch
|
|
|
|
| 82 |
"vae": vae,
|
| 83 |
}
|
| 84 |
|
| 85 |
+
pipeline = XoraVideoPipeline(**submodel_dict).to("cuda")
|
| 86 |
|
| 87 |
# Sample input
|
| 88 |
num_inference_steps = 20
|
xora/models/transformers/symmetric_patchifier.py
CHANGED
|
@@ -60,26 +60,19 @@ class Patchifier(ConfigMixin, ABC):
|
|
| 60 |
return grid
|
| 61 |
|
| 62 |
|
| 63 |
-
def pixart_alpha_patchify(
|
| 64 |
-
latents: Tensor,
|
| 65 |
-
patch_size: int,
|
| 66 |
-
) -> Tuple[Tensor, Tensor]:
|
| 67 |
-
latents = rearrange(
|
| 68 |
-
latents,
|
| 69 |
-
"b c (f p1) (h p2) (w p3) -> b (f h w) (c p1 p2 p3)",
|
| 70 |
-
p1=patch_size[0],
|
| 71 |
-
p2=patch_size[1],
|
| 72 |
-
p3=patch_size[2],
|
| 73 |
-
)
|
| 74 |
-
return latents
|
| 75 |
-
|
| 76 |
-
|
| 77 |
class SymmetricPatchifier(Patchifier):
|
| 78 |
def patchify(
|
| 79 |
self,
|
| 80 |
latents: Tensor,
|
| 81 |
) -> Tuple[Tensor, Tensor]:
|
| 82 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
|
| 84 |
def unpatchify(
|
| 85 |
self,
|
|
|
|
| 60 |
return grid
|
| 61 |
|
| 62 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
class SymmetricPatchifier(Patchifier):
|
| 64 |
def patchify(
|
| 65 |
self,
|
| 66 |
latents: Tensor,
|
| 67 |
) -> Tuple[Tensor, Tensor]:
|
| 68 |
+
latents = rearrange(
|
| 69 |
+
latents,
|
| 70 |
+
"b c (f p1) (h p2) (w p3) -> b (f h w) (c p1 p2 p3)",
|
| 71 |
+
p1=self._patch_size[0],
|
| 72 |
+
p2=self._patch_size[1],
|
| 73 |
+
p3=self._patch_size[2],
|
| 74 |
+
)
|
| 75 |
+
return latents
|
| 76 |
|
| 77 |
def unpatchify(
|
| 78 |
self,
|
xora/models/transformers/transformer3d.py
CHANGED
|
@@ -141,12 +141,10 @@ class Transformer3DModel(ModelMixin, ConfigMixin):
|
|
| 141 |
)
|
| 142 |
self.proj_out = nn.Linear(inner_dim, self.out_channels)
|
| 143 |
|
| 144 |
-
# 5. PixArt-Alpha blocks.
|
| 145 |
self.adaln_single = AdaLayerNormSingle(
|
| 146 |
inner_dim, use_additional_conditions=False
|
| 147 |
)
|
| 148 |
if adaptive_norm == "single_scale":
|
| 149 |
-
# Use 4 channels instead of the 6 for the PixArt-Alpha scale + shift ada norm.
|
| 150 |
self.adaln_single.linear = nn.Linear(inner_dim, 4 * inner_dim, bias=True)
|
| 151 |
|
| 152 |
self.caption_projection = None
|
|
@@ -170,7 +168,7 @@ class Transformer3DModel(ModelMixin, ConfigMixin):
|
|
| 170 |
for block in self.transformer_blocks:
|
| 171 |
block.set_use_tpu_flash_attention(self.device.type)
|
| 172 |
|
| 173 |
-
def initialize(self, embedding_std: float, mode: Literal["xora", "
|
| 174 |
def _basic_init(module):
|
| 175 |
if isinstance(module, nn.Linear):
|
| 176 |
torch.nn.init.xavier_uniform_(module.weight)
|
|
@@ -211,7 +209,6 @@ class Transformer3DModel(ModelMixin, ConfigMixin):
|
|
| 211 |
nn.init.normal_(self.caption_projection.linear_1.weight, std=embedding_std)
|
| 212 |
nn.init.normal_(self.caption_projection.linear_1.weight, std=embedding_std)
|
| 213 |
|
| 214 |
-
# Zero-out adaLN modulation layers in PixArt blocks:
|
| 215 |
for block in self.transformer_blocks:
|
| 216 |
if mode.lower() == "xora":
|
| 217 |
nn.init.constant_(block.attn1.to_out[0].weight, 0)
|
|
|
|
| 141 |
)
|
| 142 |
self.proj_out = nn.Linear(inner_dim, self.out_channels)
|
| 143 |
|
|
|
|
| 144 |
self.adaln_single = AdaLayerNormSingle(
|
| 145 |
inner_dim, use_additional_conditions=False
|
| 146 |
)
|
| 147 |
if adaptive_norm == "single_scale":
|
|
|
|
| 148 |
self.adaln_single.linear = nn.Linear(inner_dim, 4 * inner_dim, bias=True)
|
| 149 |
|
| 150 |
self.caption_projection = None
|
|
|
|
| 168 |
for block in self.transformer_blocks:
|
| 169 |
block.set_use_tpu_flash_attention(self.device.type)
|
| 170 |
|
| 171 |
+
def initialize(self, embedding_std: float, mode: Literal["xora", "legacy"]):
|
| 172 |
def _basic_init(module):
|
| 173 |
if isinstance(module, nn.Linear):
|
| 174 |
torch.nn.init.xavier_uniform_(module.weight)
|
|
|
|
| 209 |
nn.init.normal_(self.caption_projection.linear_1.weight, std=embedding_std)
|
| 210 |
nn.init.normal_(self.caption_projection.linear_1.weight, std=embedding_std)
|
| 211 |
|
|
|
|
| 212 |
for block in self.transformer_blocks:
|
| 213 |
if mode.lower() == "xora":
|
| 214 |
nn.init.constant_(block.attn1.to_out[0].weight, 0)
|
xora/pipelines/{pipeline_video_pixart_alpha.py → pipeline_xora_video.py}
RENAMED
|
@@ -1,4 +1,4 @@
|
|
| 1 |
-
#
|
| 2 |
import html
|
| 3 |
import inspect
|
| 4 |
import math
|
|
@@ -19,7 +19,6 @@ from diffusers.utils import (
|
|
| 19 |
is_bs4_available,
|
| 20 |
is_ftfy_available,
|
| 21 |
logging,
|
| 22 |
-
replace_example_docstring,
|
| 23 |
)
|
| 24 |
from diffusers.utils.torch_utils import randn_tensor
|
| 25 |
from einops import rearrange
|
|
@@ -44,22 +43,6 @@ if is_bs4_available():
|
|
| 44 |
if is_ftfy_available():
|
| 45 |
import ftfy
|
| 46 |
|
| 47 |
-
EXAMPLE_DOC_STRING = """
|
| 48 |
-
Examples:
|
| 49 |
-
```py
|
| 50 |
-
>>> import torch
|
| 51 |
-
>>> from diffusers import PixArtAlphaPipeline
|
| 52 |
-
|
| 53 |
-
>>> # You can replace the checkpoint id with "PixArt-alpha/PixArt-XL-2-512x512" too.
|
| 54 |
-
>>> pipe = PixArtAlphaPipeline.from_pretrained("PixArt-alpha/PixArt-XL-2-1024-MS", torch_dtype=torch.float16)
|
| 55 |
-
>>> # Enable memory optimizations.
|
| 56 |
-
>>> pipe.enable_model_cpu_offload()
|
| 57 |
-
|
| 58 |
-
>>> prompt = "A small cactus with a happy face in the Sahara desert."
|
| 59 |
-
>>> image = pipe(prompt).images[0]
|
| 60 |
-
```
|
| 61 |
-
"""
|
| 62 |
-
|
| 63 |
ASPECT_RATIO_1024_BIN = {
|
| 64 |
"0.25": [512.0, 2048.0],
|
| 65 |
"0.28": [512.0, 1856.0],
|
|
@@ -180,9 +163,9 @@ def retrieve_timesteps(
|
|
| 180 |
return timesteps, num_inference_steps
|
| 181 |
|
| 182 |
|
| 183 |
-
class
|
| 184 |
r"""
|
| 185 |
-
Pipeline for text-to-image generation using
|
| 186 |
|
| 187 |
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
| 188 |
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
|
@@ -191,7 +174,7 @@ class VideoPixArtAlphaPipeline(DiffusionPipeline):
|
|
| 191 |
vae ([`AutoencoderKL`]):
|
| 192 |
Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
|
| 193 |
text_encoder ([`T5EncoderModel`]):
|
| 194 |
-
Frozen text-encoder.
|
| 195 |
[T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel), specifically the
|
| 196 |
[t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
|
| 197 |
tokenizer (`T5Tokenizer`):
|
|
@@ -247,7 +230,6 @@ class VideoPixArtAlphaPipeline(DiffusionPipeline):
|
|
| 247 |
)
|
| 248 |
self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
|
| 249 |
|
| 250 |
-
# Adapted from https://github.com/PixArt-alpha/PixArt-alpha/blob/master/diffusion/model/utils.py
|
| 251 |
def mask_text_embeddings(self, emb, mask):
|
| 252 |
if emb.shape[0] == 1:
|
| 253 |
keep_index = mask.sum().item()
|
|
@@ -280,7 +262,7 @@ class VideoPixArtAlphaPipeline(DiffusionPipeline):
|
|
| 280 |
negative_prompt (`str` or `List[str]`, *optional*):
|
| 281 |
The prompt not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`
|
| 282 |
instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). For
|
| 283 |
-
|
| 284 |
do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
|
| 285 |
whether to use classifier free guidance or not
|
| 286 |
num_images_per_prompt (`int`, *optional*, defaults to 1):
|
|
@@ -291,8 +273,7 @@ class VideoPixArtAlphaPipeline(DiffusionPipeline):
|
|
| 291 |
Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
|
| 292 |
provided, text embeddings will be generated from `prompt` input argument.
|
| 293 |
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
| 294 |
-
Pre-generated negative text embeddings.
|
| 295 |
-
string.
|
| 296 |
clean_caption (bool, defaults to `False`):
|
| 297 |
If `True`, the function will preprocess and clean the provided caption before encoding.
|
| 298 |
"""
|
|
@@ -753,7 +734,6 @@ class VideoPixArtAlphaPipeline(DiffusionPipeline):
|
|
| 753 |
return samples
|
| 754 |
|
| 755 |
@torch.no_grad()
|
| 756 |
-
@replace_example_docstring(EXAMPLE_DOC_STRING)
|
| 757 |
def __call__(
|
| 758 |
self,
|
| 759 |
height: int,
|
|
@@ -824,7 +804,7 @@ class VideoPixArtAlphaPipeline(DiffusionPipeline):
|
|
| 824 |
provided, text embeddings will be generated from `prompt` input argument.
|
| 825 |
prompt_attention_mask (`torch.FloatTensor`, *optional*): Pre-generated attention mask for text embeddings.
|
| 826 |
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
| 827 |
-
Pre-generated negative text embeddings.
|
| 828 |
provided, negative_prompt_embeds will be generated from `negative_prompt` input argument.
|
| 829 |
negative_prompt_attention_mask (`torch.FloatTensor`, *optional*):
|
| 830 |
Pre-generated attention mask for negative text embeddings.
|
|
|
|
| 1 |
+
# Adapted from: https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pixart_alpha/pipeline_pixart_alpha.py
|
| 2 |
import html
|
| 3 |
import inspect
|
| 4 |
import math
|
|
|
|
| 19 |
is_bs4_available,
|
| 20 |
is_ftfy_available,
|
| 21 |
logging,
|
|
|
|
| 22 |
)
|
| 23 |
from diffusers.utils.torch_utils import randn_tensor
|
| 24 |
from einops import rearrange
|
|
|
|
| 43 |
if is_ftfy_available():
|
| 44 |
import ftfy
|
| 45 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
ASPECT_RATIO_1024_BIN = {
|
| 47 |
"0.25": [512.0, 2048.0],
|
| 48 |
"0.28": [512.0, 1856.0],
|
|
|
|
| 163 |
return timesteps, num_inference_steps
|
| 164 |
|
| 165 |
|
| 166 |
+
class XoraVideoPipeline(DiffusionPipeline):
|
| 167 |
r"""
|
| 168 |
+
Pipeline for text-to-image generation using Xora.
|
| 169 |
|
| 170 |
This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
|
| 171 |
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
|
|
|
|
| 174 |
vae ([`AutoencoderKL`]):
|
| 175 |
Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
|
| 176 |
text_encoder ([`T5EncoderModel`]):
|
| 177 |
+
Frozen text-encoder. This uses
|
| 178 |
[T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel), specifically the
|
| 179 |
[t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
|
| 180 |
tokenizer (`T5Tokenizer`):
|
|
|
|
| 230 |
)
|
| 231 |
self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
|
| 232 |
|
|
|
|
| 233 |
def mask_text_embeddings(self, emb, mask):
|
| 234 |
if emb.shape[0] == 1:
|
| 235 |
keep_index = mask.sum().item()
|
|
|
|
| 262 |
negative_prompt (`str` or `List[str]`, *optional*):
|
| 263 |
The prompt not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`
|
| 264 |
instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`). For
|
| 265 |
+
This should be "".
|
| 266 |
do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
|
| 267 |
whether to use classifier free guidance or not
|
| 268 |
num_images_per_prompt (`int`, *optional*, defaults to 1):
|
|
|
|
| 273 |
Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
|
| 274 |
provided, text embeddings will be generated from `prompt` input argument.
|
| 275 |
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
| 276 |
+
Pre-generated negative text embeddings.
|
|
|
|
| 277 |
clean_caption (bool, defaults to `False`):
|
| 278 |
If `True`, the function will preprocess and clean the provided caption before encoding.
|
| 279 |
"""
|
|
|
|
| 734 |
return samples
|
| 735 |
|
| 736 |
@torch.no_grad()
|
|
|
|
| 737 |
def __call__(
|
| 738 |
self,
|
| 739 |
height: int,
|
|
|
|
| 804 |
provided, text embeddings will be generated from `prompt` input argument.
|
| 805 |
prompt_attention_mask (`torch.FloatTensor`, *optional*): Pre-generated attention mask for text embeddings.
|
| 806 |
negative_prompt_embeds (`torch.FloatTensor`, *optional*):
|
| 807 |
+
Pre-generated negative text embeddings. This negative prompt should be "". If not
|
| 808 |
provided, negative_prompt_embeds will be generated from `negative_prompt` input argument.
|
| 809 |
negative_prompt_attention_mask (`torch.FloatTensor`, *optional*):
|
| 810 |
Pre-generated attention mask for negative text embeddings.
|