HiDream-E1-Full

Image: HiDream-E1 Demo

HiDream-E1 is an image editing model built on HiDream-I1.

For more features and to experience the full capabilities of our product, please visit https://vivago.ai/.

Project Updates#

• 🌟 July 16, 2025: We've open-sourced the updated image editing model HiDream-E1.1.
• 📝 May 28, 2025: We've released our technical report HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformer.
• 🚀 April 28, 2025: We've open-sourced the instruction-based image editing model HiDream-E1.

Quick Start#

Please make sure you have installed Flash Attention and latest Diffusers. We recommend CUDA versions 12.4 for the manual installation.

pip install -r requirements.txt
pip install -U flash-attn --no-build-isolation
pip install -U git+https://github.com/huggingface/diffusers.git

Then you can run the inference scripts to generate images:

python ./inference.py

Alternatively, you can use the model in your own code:

import torch
from transformers import PreTrainedTokenizerFast, LlamaForCausalLM
from pipeline_hidream_image_editing import HiDreamImageEditingPipeline
from PIL import Image

# Load the tokenizer and text encoder
tokenizer_4 = PreTrainedTokenizerFast.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
text_encoder_4 = LlamaForCausalLM.from_pretrained(
    "meta-llama/Llama-3.1-8B-Instruct",
    output_hidden_states=True,
    output_attentions=True,
    torch_dtype=torch.bfloat16,
)

# Load the HiDream pipeline
pipe = HiDreamImageEditingPipeline.from_pretrained(
    "HiDream-ai/HiDream-E1-Full",
    tokenizer_4=tokenizer_4,
    text_encoder_4=text_encoder_4,
    torch_dtype=torch.bfloat16,
)

# Load and prepare input image
test_image = Image.open("your_image.jpg")
test_image = test_image.resize((768, 768))

# Move pipeline to GPU
pipe = pipe.to("cuda", torch.bfloat16)

# Generate edited image
image = pipe(
    prompt = 'Editing Instruction: Convert the image into a Ghibli style. Target Image Description: A person in a light pink t-shirt with short dark hair, depicted in a Ghibli style against a plain background.',
    negative_prompt = "low resolution, blur",
    image = test_image,
    guidance_scale=5.0,
    image_guidance_scale=4.0,
    num_inference_steps=28,
    generator=torch.Generator("cuda").manual_seed(3),
).images[0]

# Save output image
image.save("output.jpg")

Note. The inference script will try to automatically download meta-llama/Llama-3.1-8B-Instruct model files. You need to agree to the license of the Llama model on your HuggingFace account and login using huggingface-cli login in order to use the automatic downloader.

Note. The model accepts instructions in the following format:

Editing Instruction: {instruction}. Target Image Description: {description}

Copy

Example:

Editing Instruction: Convert the image into a Ghibli style. Target Image Description: A person in a light pink t-shirt with short dark hair, depicted in a Ghibli style against a plain background.

Copy

To refine your instructions, use the provided script:

python ./instruction_refinement.py --src_image ./test.jpeg --src_instruction "convert the image into a Ghibli style"

Copy

The instruction refinement script requires a VLM API key - you can either run vllm locally or use OpenAI's API.

Gradio Demo#

We also provide a Gradio demo for interactive image editing. You can run the demo with:

python gradio_demo.py 

Evaluation Metrics#

Evaluation results on EmuEdit and ReasonEdit Benchmarks. Higher is better.

License Agreement#

The Transformer models in this repository are licensed under the MIT License. The VAE is from FLUX.1 [schnell], and the text encoders from google/t5-v1_1-xxl and meta-llama/Meta-Llama-3.1-8B-Instruct. Please follow the license terms specified for these components. You own all content you create with this model. You can use your generated content freely, but you must comply with this license agreement. You are responsible for how you use the models. Do not create illegal content, harmful material, personal information that could harm others, false information, or content targeting vulnerable groups.

Acknowledgements#

• The VAE component is from FLUX.1 [schnell], licensed under Apache 2.0.
• The text encoders are from google/t5-v1_1-xxl (licensed under Apache 2.0) and meta-llama/Meta-Llama-3.1-8B-Instruct (licensed under the Llama 3.1 Community License Agreement).

Citation#

@article{hidreami1technicalreport,
  title={HiDream-I1: A High-Efficient Image Generative Foundation Model with Sparse Diffusion Transformer},
  author={Cai, Qi and Chen, Jingwen and Chen, Yang and Li, Yehao and Long, Fuchen and Pan, Yingwei and Qiu, Zhaofan and Zhang, Yiheng and Gao, Fengbin and Xu, Peihan and others},
  journal={arXiv preprint arXiv:2505.22705},
  year={2025}
}