Hunyuan-DiT (original) (raw)

Hunyuan-DiT : A Powerful Multi-Resolution Diffusion Transformer with Fine-Grained Chinese Understanding from Tencent Hunyuan.

The abstract from the paper is:

We present Hunyuan-DiT, a text-to-image diffusion transformer with fine-grained understanding of both English and Chinese. To construct Hunyuan-DiT, we carefully design the transformer structure, text encoder, and positional encoding. We also build from scratch a whole data pipeline to update and evaluate data for iterative model optimization. For fine-grained language understanding, we train a Multimodal Large Language Model to refine the captions of the images. Finally, Hunyuan-DiT can perform multi-turn multimodal dialogue with users, generating and refining images according to the context. Through our holistic human evaluation protocol with more than 50 professional human evaluators, Hunyuan-DiT sets a new state-of-the-art in Chinese-to-image generation compared with other open-source models.

You can find the original codebase at Tencent/HunyuanDiT and all the available checkpoints at Tencent-Hunyuan.

Highlights: HunyuanDiT supports Chinese/English-to-image, multi-resolution generation.

HunyuanDiT has the following components:

• It uses a diffusion transformer as the backbone
• It combines two text encoders, a bilingual CLIP and a multilingual T5 encoder

Make sure to check out the Schedulers guide to learn how to explore the tradeoff between scheduler speed and quality, and see the reuse components across pipelines section to learn how to efficiently load the same components into multiple pipelines.

You can further improve generation quality by passing the generated image from HungyuanDiTPipeline to the SDXL refiner model.

Optimization

You can optimize the pipeline’s runtime and memory consumption with torch.compile and feed-forward chunking. To learn about other optimization methods, check out the Speed up inference and Reduce memory usage guides.

Inference

Use torch.compile to reduce the inference latency.

First, load the pipeline:

from diffusers import HunyuanDiTPipeline import torch

pipeline = HunyuanDiTPipeline.from_pretrained( "Tencent-Hunyuan/HunyuanDiT-Diffusers", torch_dtype=torch.float16 ).to("cuda")

Then change the memory layout of the pipelines transformer and vae components to torch.channels-last:

pipeline.transformer.to(memory_format=torch.channels_last) pipeline.vae.to(memory_format=torch.channels_last)

Finally, compile the components and run inference:

pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=True) pipeline.vae.decode = torch.compile(pipeline.vae.decode, mode="max-autotune", fullgraph=True)

image = pipeline(prompt="一个宇航员在骑马").images[0]

The benchmark results on a 80GB A100 machine are:

With torch.compile(): Average inference time: 12.470 seconds. Without torch.compile(): Average inference time: 20.570 seconds.

Memory optimization

By loading the T5 text encoder in 8 bits, you can run the pipeline in just under 6 GBs of GPU VRAM. Refer to this script for details.

Furthermore, you can use the enable_forward_chunking() method to reduce memory usage. Feed-forward chunking runs the feed-forward layers in a transformer block in a loop instead of all at once. This gives you a trade-off between memory consumption and inference runtime.

• pipeline.transformer.enable_forward_chunking(chunk_size=1, dim=1)

HunyuanDiTPipeline

class diffusers.HunyuanDiTPipeline

< source >

( vae: AutoencoderKL text_encoder: BertModel tokenizer: BertTokenizer transformer: HunyuanDiT2DModel scheduler: DDPMScheduler safety_checker: StableDiffusionSafetyChecker feature_extractor: CLIPImageProcessor requires_safety_checker: bool = True text_encoder_2: typing.Optional[transformers.models.t5.modeling_t5.T5EncoderModel] = None tokenizer_2: typing.Optional[transformers.models.mt5.tokenization_mt5.MT5Tokenizer] = None )

Parameters

• vae (AutoencoderKL) — Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. We usesdxl-vae-fp16-fix.
• text_encoder (Optional[~transformers.BertModel, ~transformers.CLIPTextModel]) — Frozen text-encoder (clip-vit-large-patch14). HunyuanDiT uses a fine-tuned [bilingual CLIP].
• tokenizer (Optional[~transformers.BertTokenizer, ~transformers.CLIPTokenizer]) — A BertTokenizer or CLIPTokenizer to tokenize text.
• transformer (HunyuanDiT2DModel) — The HunyuanDiT model designed by Tencent Hunyuan.
• text_encoder_2 (T5EncoderModel) — The mT5 embedder. Specifically, it is ‘t5-v1_1-xxl’.
• tokenizer_2 (MT5Tokenizer) — The tokenizer for the mT5 embedder.
• scheduler (DDPMScheduler) — A scheduler to be used in combination with HunyuanDiT to denoise the encoded image latents.

Pipeline for English/Chinese-to-image generation using HunyuanDiT.

This model inherits from DiffusionPipeline. Check the superclass documentation for the generic methods the library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)

HunyuanDiT uses two text encoders: mT5 and [bilingual CLIP](fine-tuned by ourselves)

__call__

< source >

( prompt: typing.Union[str, typing.List[str]] = None height: typing.Optional[int] = None width: typing.Optional[int] = None num_inference_steps: typing.Optional[int] = 50 guidance_scale: typing.Optional[float] = 5.0 negative_prompt: typing.Union[str, typing.List[str], NoneType] = None num_images_per_prompt: typing.Optional[int] = 1 eta: typing.Optional[float] = 0.0 generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None latents: typing.Optional[torch.Tensor] = None prompt_embeds: typing.Optional[torch.Tensor] = None prompt_embeds_2: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds_2: typing.Optional[torch.Tensor] = None prompt_attention_mask: typing.Optional[torch.Tensor] = None prompt_attention_mask_2: typing.Optional[torch.Tensor] = None negative_prompt_attention_mask: typing.Optional[torch.Tensor] = None negative_prompt_attention_mask_2: typing.Optional[torch.Tensor] = None output_type: typing.Optional[str] = 'pil' return_dict: bool = True callback_on_step_end: typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None callback_on_step_end_tensor_inputs: typing.List[str] = ['latents'] guidance_rescale: float = 0.0 original_size: typing.Optional[typing.Tuple[int, int]] = (1024, 1024) target_size: typing.Optional[typing.Tuple[int, int]] = None crops_coords_top_left: typing.Tuple[int, int] = (0, 0) use_resolution_binning: bool = True ) → StableDiffusionPipelineOutput or tuple

Parameters

• prompt (str or List[str], optional) — The prompt or prompts to guide image generation. If not defined, you need to pass prompt_embeds.
• height (int) — The height in pixels of the generated image.
• width (int) — The width in pixels of the generated image.
• num_inference_steps (int, optional, defaults to 50) — The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. This parameter is modulated by strength.
• guidance_scale (float, optional, defaults to 7.5) — A higher guidance scale value encourages the model to generate images closely linked to the textprompt at the expense of lower image quality. Guidance scale is enabled when guidance_scale > 1.
• negative_prompt (str or List[str], optional) — The prompt or prompts to guide what to not include in image generation. If not defined, you need to pass negative_prompt_embeds instead. Ignored when not using guidance (guidance_scale < 1).
• num_images_per_prompt (int, optional, defaults to 1) — The number of images to generate per prompt.
• eta (float, optional, defaults to 0.0) — Corresponds to parameter eta (η) from the DDIM paper. Only applies to the DDIMScheduler, and is ignored in other schedulers.
• generator (torch.Generator or List[torch.Generator], optional) — A torch.Generator to make generation deterministic.
• prompt_embeds (torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from the prompt input argument.
• prompt_embeds_2 (torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from the prompt input argument.
• negative_prompt_embeds (torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, negative_prompt_embeds are generated from the negative_prompt input argument.
• negative_prompt_embeds_2 (torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, negative_prompt_embeds are generated from the negative_prompt input argument.
• prompt_attention_mask (torch.Tensor, optional) — Attention mask for the prompt. Required when prompt_embeds is passed directly.
• prompt_attention_mask_2 (torch.Tensor, optional) — Attention mask for the prompt. Required when prompt_embeds_2 is passed directly.
• negative_prompt_attention_mask (torch.Tensor, optional) — Attention mask for the negative prompt. Required when negative_prompt_embeds is passed directly.
• negative_prompt_attention_mask_2 (torch.Tensor, optional) — Attention mask for the negative prompt. Required when negative_prompt_embeds_2 is passed directly.
• output_type (str, optional, defaults to "pil") — The output format of the generated image. Choose between PIL.Image or np.array.
• return_dict (bool, optional, defaults to True) — Whether or not to return a StableDiffusionPipelineOutput instead of a plain tuple.
• callback_on_step_end (Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks, optional) — A callback function or a list of callback functions to be called at the end of each denoising step.
• callback_on_step_end_tensor_inputs (List[str], optional) — A list of tensor inputs that should be passed to the callback function. If not defined, all tensor inputs will be passed.
• guidance_rescale (float, optional, defaults to 0.0) — Rescale the noise_cfg according to guidance_rescale. Based on findings of Common Diffusion Noise Schedules and Sample Steps are Flawed. See Section 3.4
• original_size (Tuple[int, int], optional, defaults to (1024, 1024)) — The original size of the image. Used to calculate the time ids.
• target_size (Tuple[int, int], optional) — The target size of the image. Used to calculate the time ids.
• crops_coords_top_left (Tuple[int, int], optional, defaults to (0, 0)) — The top left coordinates of the crop. Used to calculate the time ids.
• use_resolution_binning (bool, optional, defaults to True) — Whether to use resolution binning or not. If True, the input resolution will be mapped to the closest standard resolution. Supported resolutions are 1024x1024, 1280x1280, 1024x768, 1152x864, 1280x960, 768x1024, 864x1152, 960x1280, 1280x768, and 768x1280. It is recommended to set this to True.

If return_dict is True, StableDiffusionPipelineOutput is returned, otherwise a tuple is returned where the first element is a list with the generated images and the second element is a list of bools indicating whether the corresponding generated image contains “not-safe-for-work” (nsfw) content.

The call function to the pipeline for generation with HunyuanDiT.

Examples:

import torch from diffusers import HunyuanDiTPipeline

pipe = HunyuanDiTPipeline.from_pretrained( ... "Tencent-Hunyuan/HunyuanDiT-Diffusers", torch_dtype=torch.float16 ... ) pipe.to("cuda")

prompt = "一个宇航员在骑马" image = pipe(prompt).images[0]

encode_prompt

< source >

( prompt: str device: device = None dtype: dtype = None num_images_per_prompt: int = 1 do_classifier_free_guidance: bool = True negative_prompt: typing.Optional[str] = None prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None prompt_attention_mask: typing.Optional[torch.Tensor] = None negative_prompt_attention_mask: typing.Optional[torch.Tensor] = None max_sequence_length: typing.Optional[int] = None text_encoder_index: int = 0 )

Parameters

• prompt (str or List[str], optional) — prompt to be encoded
• device — (torch.device): torch device
• dtype (torch.dtype) — torch dtype
• num_images_per_prompt (int) — number of images that should be generated per prompt
• do_classifier_free_guidance (bool) — whether to use classifier free guidance or not
• negative_prompt (str or List[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embeds instead. Ignored when not using guidance (i.e., ignored if guidance_scale is less than 1).
• prompt_embeds (torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument.
• negative_prompt_embeds (torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated from negative_prompt input argument.
• prompt_attention_mask (torch.Tensor, optional) — Attention mask for the prompt. Required when prompt_embeds is passed directly.
• negative_prompt_attention_mask (torch.Tensor, optional) — Attention mask for the negative prompt. Required when negative_prompt_embeds is passed directly.
• max_sequence_length (int, optional) — maximum sequence length to use for the prompt.
• text_encoder_index (int, optional) — Index of the text encoder to use. 0 for clip and 1 for T5.

Encodes the prompt into text encoder hidden states.

< > Update on GitHub