Upload 10 files

.env

@@ -0,0 +1 @@
+OPENAI_API_KEY = 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

.gitattributes

@@ -1,35 +1,37 @@
-*.7z filter=lfs diff=lfs merge=lfs -text
-*.arrow filter=lfs diff=lfs merge=lfs -text
-*.bin filter=lfs diff=lfs merge=lfs -text
-*.bz2 filter=lfs diff=lfs merge=lfs -text
-*.ckpt filter=lfs diff=lfs merge=lfs -text
-*.ftz filter=lfs diff=lfs merge=lfs -text
-*.gz filter=lfs diff=lfs merge=lfs -text
-*.h5 filter=lfs diff=lfs merge=lfs -text
-*.joblib filter=lfs diff=lfs merge=lfs -text
-*.lfs.* filter=lfs diff=lfs merge=lfs -text
-*.mlmodel filter=lfs diff=lfs merge=lfs -text
-*.model filter=lfs diff=lfs merge=lfs -text
-*.msgpack filter=lfs diff=lfs merge=lfs -text
-*.npy filter=lfs diff=lfs merge=lfs -text
-*.npz filter=lfs diff=lfs merge=lfs -text
-*.onnx filter=lfs diff=lfs merge=lfs -text
-*.ot filter=lfs diff=lfs merge=lfs -text
-*.parquet filter=lfs diff=lfs merge=lfs -text
-*.pb filter=lfs diff=lfs merge=lfs -text
-*.pickle filter=lfs diff=lfs merge=lfs -text
-*.pkl filter=lfs diff=lfs merge=lfs -text
-*.pt filter=lfs diff=lfs merge=lfs -text
-*.pth filter=lfs diff=lfs merge=lfs -text
-*.rar filter=lfs diff=lfs merge=lfs -text
-*.safetensors filter=lfs diff=lfs merge=lfs -text
-saved_model/**/* filter=lfs diff=lfs merge=lfs -text
-*.tar.* filter=lfs diff=lfs merge=lfs -text
-*.tar filter=lfs diff=lfs merge=lfs -text
-*.tflite filter=lfs diff=lfs merge=lfs -text
-*.tgz filter=lfs diff=lfs merge=lfs -text
-*.wasm filter=lfs diff=lfs merge=lfs -text
-*.xz filter=lfs diff=lfs merge=lfs -text
-*.zip filter=lfs diff=lfs merge=lfs -text
-*.zst filter=lfs diff=lfs merge=lfs -text
-*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.jpg filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,183 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# UV
+#   Similar to Pipfile.lock, it is generally recommended to include uv.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#uv.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/latest/usage/project/#working-with-version-control
+.pdm.toml
+.pdm-python
+.pdm-build/
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+#.idea/
+
+# Ruff stuff:
+.ruff_cache/
+
+# PyPI configuration file
+.pypirc
+
+# User config files
+.vscode/
+output/
+
+# ckpt
+*.bin
+*.pt
+*.pth

.gitmodules

@@ -0,0 +1,3 @@
+[submodule "datasets/dreambooth"]
+	path = datasets/dreambooth
+	url = https://github.com/google/dreambooth.git

LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -1,13 +1,144 @@
----
-title: Uno Final
-emoji: 👁
-colorFrom: green
-colorTo: pink
-sdk: gradio
-sdk_version: 5.25.2
-app_file: app.py
-pinned: false
-short_description: uno-final
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+<h3 align="center">
+    <img src="assets/logo.png" alt="Logo" style="vertical-align: middle; width: 40px; height: 40px;">
+    Less-to-More Generalization: Unlocking More Controllability by In-Context Generation
+</h3>
+
+<p align="center"> 
+<a href="https://github.com/bytedance/UNO"><img alt="Build" src="https://img.shields.io/github/stars/bytedance/UNO"></a> 
+<a href="https://bytedance.github.io/UNO/"><img alt="Build" src="https://img.shields.io/badge/Project%20Page-UNO-yellow"></a> 
+<a href="https://arxiv.org/abs/2504.02160"><img alt="Build" src="https://img.shields.io/badge/arXiv%20paper-UNO-b31b1b.svg"></a>
+<a href="https://huggingface.co/bytedance-research/UNO"><img src="https://img.shields.io/static/v1?label=%F0%9F%A4%97%20Hugging%20Face&message=Model&color=orange"></a>
+<a href="https://huggingface.co/spaces/bytedance-research/UNO-FLUX"><img src="https://img.shields.io/static/v1?label=%F0%9F%A4%97%20Hugging%20Face&message=demo&color=orange"></a>
+</p>
+
+><p align="center"> <span style="color:#137cf3; font-family: Gill Sans">Shaojin Wu,</span><sup></sup></a>  <span style="color:#137cf3; font-family: Gill Sans">Mengqi Huang</span><sup>*</sup>,</a> <span style="color:#137cf3; font-family: Gill Sans">Wenxu Wu,</span><sup></sup></a>  <span style="color:#137cf3; font-family: Gill Sans">Yufeng Cheng,</span><sup></sup> </a>  <span style="color:#137cf3; font-family: Gill Sans">Fei Ding</span><sup>+</sup>,</a> <span style="color:#137cf3; font-family: Gill Sans">Qian He</span></a> <br> 
+><span style="font-size: 16px">Intelligent Creation Team, ByteDance</span></p>
+
+<p align="center">
+<img src="./assets/teaser.jpg" width=95% height=95% 
+class="center">
+</p>
+
+## 🔥 News
+- [04/2025] 🔥 Update fp8 mode as a primary low vmemory usage support. Gift for consumer-grade GPU users. The peak Vmemory usage is ~16GB now. We may try further inference optimization later.
+- [04/2025] 🔥 The [demo](https://huggingface.co/spaces/bytedance-research/UNO-FLUX) of UNO is released.
+- [04/2025] 🔥 The [training code](https://github.com/bytedance/UNO), [inference code](https://github.com/bytedance/UNO), and [model](https://huggingface.co/bytedance-research/UNO) of UNO are released.
+- [04/2025] 🔥 The [project page](https://bytedance.github.io/UNO) of UNO is created.
+- [04/2025] 🔥 The arXiv [paper](https://arxiv.org/abs/2504.02160) of UNO is released.
+
+## 📖 Introduction
+In this study, we propose a highly-consistent data synthesis pipeline to tackle this challenge. This pipeline harnesses the intrinsic in-context generation capabilities of diffusion transformers and generates high-consistency multi-subject paired data. Additionally, we introduce UNO, which consists of progressive cross-modal alignment and universal rotary position embedding. It is a multi-image conditioned subject-to-image model iteratively trained from a text-to-image model. Extensive experiments show that our method can achieve high consistency while ensuring controllability in both single-subject and multi-subject driven generation.
+
+
+## ⚡️ Quick Start
+
+### 🔧 Requirements and Installation
+
+Install the requirements
+```bash
+## create a virtual environment with python >= 3.10 <= 3.12, like
+# python -m venv uno_env
+# source uno_env/bin/activate
+# then install
+pip install -r requirements.txt
+```
+
+then download checkpoints in one of the three ways:
+1. Directly run the inference scripts, the checkpoints will be downloaded automatically by the `hf_hub_download` function in the code to your `$HF_HOME`(the default value is `~/.cache/huggingface`).
+2. use `huggingface-cli download <repo name>` to download `black-forest-labs/FLUX.1-dev`, `xlabs-ai/xflux_text_encoders`, `openai/clip-vit-large-patch14`, `bytedance-research/UNO`, then run the inference scripts. You can just download the checkpoint in need only to speed up your set up and save your disk space. i.e. for `black-forest-labs/FLUX.1-dev` use `huggingface-cli download black-forest-labs/FLUX.1-dev flux1-dev.safetensors` and `huggingface-cli download black-forest-labs/FLUX.1-dev ae.safetensors`, ignoreing the text encoder in `black-forest-labes/FLUX.1-dev` model repo(They are here for `diffusers` call). All of the checkpoints will take 37 GB of disk space.
+3. use `huggingface-cli download <repo name> --local-dir <LOCAL_DIR>` to download all the checkpoints mentioned in 2. to the directories your want. Then set the environment variable `AE`, `FLUX_DEV`(or `FLUX_DEV_FP8` if you use fp8 mode), `T5`, `CLIP`, `LORA` to the corresponding paths. Finally, run the inference scripts.
+4. **If you already have some of the checkpoints**, you can set the environment variable `AE`, `FLUX_DEV`, `T5`, `CLIP`, `LORA` to the corresponding paths. Finally, run the inference scripts.
+
+### 🌟 Gradio Demo
+
+```bash
+python app.py
+```
+
+**For low vmemory usage**, please pass the `--offload` and `--name flux-dev-fp8` args. The peak memory usage will be 16GB. Just for reference, the end2end inference time is 40s to 1min on RTX 3090 in fp8 and offload mode.
+
+```bash
+python app.py --offload --name flux-dev-fp8
+```
+
+
+### ✍️ Inference
+Start from the examples below to explore and spark your creativity. ✨
+```bash
+python inference.py --prompt "A clock on the beach is under a red sun umbrella" --image_paths "assets/clock.png" --width 704 --height 704
+python inference.py --prompt "The figurine is in the crystal ball" --image_paths "assets/figurine.png" "assets/crystal_ball.png" --width 704 --height 704
+python inference.py --prompt "The logo is printed on the cup" --image_paths "assets/cat_cafe.png" "assets/cup.png" --width 704 --height 704
+```
+
+Optional prepreration: If you want to test the inference on dreambench at the first time, you should clone the submodule `dreambench` to download the dataset.
+
+```bash
+git submodule update --init
+```
+Then running the following scripts:
+```bash
+# evaluated on dreambench
+## for single-subject
+python inference.py --eval_json_path ./datasets/dreambench_singleip.json
+## for multi-subject
+python inference.py --eval_json_path ./datasets/dreambench_multiip.json
+```
+
+
+
+### 🚄 Training
+
+```bash
+accelerate launch train.py
+```
+
+
+### 📌 Tips and Notes
+We integrate single-subject and multi-subject generation within a unified model. For single-subject scenarios, the longest side of the reference image is set to 512 by default, while for multi-subject scenarios, it is set to 320. UNO demonstrates remarkable flexibility across various aspect ratios, thanks to its training on a multi-scale dataset. Despite being trained within 512 buckets, it can handle higher resolutions, including 512, 568, and 704, among others.
+
+UNO excels in subject-driven generation but has room for improvement in generalization due to dataset constraints. We are actively developing an enhanced model—stay tuned for updates. Your feedback is valuable, so please feel free to share any suggestions.
+
+## 🎨 Application Scenarios
+<p align="center">
+<img src="./assets/simplecase.jpg" width=95% height=95% 
+class="center">
+</p>
+
+## 📄 Disclaimer
+<p>
+We open-source this project for academic research. The vast majority of images 
+used in this project are either generated or licensed. If you have any concerns, 
+please contact us, and we will promptly remove any inappropriate content. 
+Our code is released under the Apache 2.0 License,, while our models are under 
+the CC BY-NC 4.0 License. Any models related to <a href="https://huggingface.co/black-forest-labs/FLUX.1-dev" target="_blank">FLUX.1-dev</a> 
+base model must adhere to the original licensing terms.
+<br><br>This research aims to advance the field of generative AI. Users are free to 
+create images using this tool, provided they comply with local laws and exercise 
+responsible usage. The developers are not liable for any misuse of the tool by users.</p>
+
+## 🚀 Updates
+For the purpose of fostering research and the open-source community, we plan to open-source the entire project, encompassing training, inference, weights, etc. Thank you for your patience and support! 🌟
+- [x] Release github repo.
+- [x] Release inference code.
+- [x] Release training code.
+- [x] Release model checkpoints.
+- [x] Release arXiv paper.
+- [x] Release huggingface space demo.
+- [ ] Release in-context data generation pipelines.
+
+## Related resources
+
+- [https://github.com/jax-explorer/ComfyUI-UNO](https://github.com/jax-explorer/ComfyUI-UNO) a ComfyUI node implementation of UNO by jax-explorer.
+
+##  Citation
+If UNO is helpful, please help to ⭐ the repo.
+
+If you find this project useful for your research, please consider citing our paper:
+```bibtex
+@article{wu2025less,
+  title={Less-to-More Generalization: Unlocking More Controllability by In-Context Generation},
+  author={Wu, Shaojin and Huang, Mengqi and Wu, Wenxu and Cheng, Yufeng and Ding, Fei and He, Qian},
+  journal={arXiv preprint arXiv:2504.02160},
+  year={2025}
+}
+```

app.py

@@ -0,0 +1,186 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates. All rights reserved.
+
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+
+#     http://www.apache.org/licenses/LICENSE-2.0
+
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import dataclasses
+import json
+from pathlib import Path
+import gradio as gr
+import torch
+import openai
+import os
+
+from uno.flux.pipeline import UNOPipeline
+from uno.utils.prompt_enhancer import enhance_prompt_with_chatgpt
+
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
+
+openai.api_key = os.getenv("OPENAI_API_KEY")
+
+def get_examples(examples_dir: str = "assets/examples") -> list:
+    examples = Path(examples_dir)
+    ans = []
+    for example in examples.iterdir():
+        if not example.is_dir():
+            continue
+        with open(example / "config.json") as f:
+            example_dict = json.load(f)
+
+        example_list = [example_dict["useage"], example_dict["prompt"]]
+
+        for key in ["image_ref1", "image_ref2", "image_ref3", "image_ref4"]:
+            example_list.append(str(example / example_dict[key]) if key in example_dict else None)
+
+        example_list.append(example_dict["seed"])
+        ans.append(example_list)
+    return ans
+
+def create_demo(model_type: str, device: str = "cuda" if torch.cuda.is_available() else "cpu", offload: bool = False):
+    pipeline = UNOPipeline(model_type, device, offload, only_lora=True, lora_rank=512)
+
+    with gr.Blocks() as demo:
+        gr.Markdown("# UNO by UNO team")
+        gr.Markdown(
+            """
+            <div style="text-align: center; display: flex; justify-content: left; gap: 5px;">
+                <a href="https://github.com/bytedance/UNO"><img alt="Build" src="https://img.shields.io/github/stars/bytedance/UNO"></a> 
+                <a href="https://bytedance.github.io/UNO/"><img alt="Build" src="https://img.shields.io/badge/Project%20Page-UNO-yellow"></a> 
+                <a href="https://arxiv.org/abs/2504.02160"><img alt="Build" src="https://img.shields.io/badge/arXiv%20paper-UNO-b31b1b.svg"></a>
+                <a href="https://huggingface.co/bytedance-research/UNO"><img src="https://img.shields.io/static/v1?label=%F0%9F%A4%97%20Hugging%20Face&message=Model&color=orange"></a>
+                <a href="https://huggingface.co/spaces/bytedance-research/UNO-FLUX"><img src="https://img.shields.io/static/v1?label=%F0%9F%A4%97%20Hugging%20Face&message=demo&color=orange"></a>
+            </div>
+            """
+        )
+
+        with gr.Row():
+            with gr.Column():
+                prompt = gr.Textbox(label="Prompt", value="handsome woman in the city")
+                with gr.Row():
+                    image_prompt1 = gr.Image(label="Ref Img1", type="pil")
+                    image_prompt2 = gr.Image(label="Ref Img2", type="pil")
+                    image_prompt3 = gr.Image(label="Ref Img3", type="pil")
+                    image_prompt4 = gr.Image(label="Ref Img4", type="pil")
+
+                with gr.Row():
+                    with gr.Column():
+                        width = gr.Slider(512, 2048, 512, step=16, label="Generation Width")
+                        height = gr.Slider(512, 2048, 512, step=16, label="Generation Height")
+                    with gr.Column():
+                        gr.Markdown("📌 Trained on 512x512. Larger size = better quality, but less stable.")
+
+                with gr.Accordion("Advanced Options", open=False):
+                    with gr.Row():
+                        num_steps = gr.Slider(1, 50, 25, step=1, label="Number of steps")
+                        guidance = gr.Slider(1.0, 5.0, 4.0, step=0.1, label="Guidance")
+                        seed = gr.Number(-1, label="Seed (-1 for random)")
+                        num_outputs = gr.Slider(1, 5, 5, step=1, label="Number of Enhanced Prompts / Images")
+
+                generate_btn = gr.Button("Generate Enhanced Images")
+
+            with gr.Column():
+                outputs = []
+                for i in range(5):
+                    outputs.append(gr.Image(label=f"Image {i+1}"))
+                    outputs.append(gr.Textbox(label=f"Enhanced Prompt {i+1}"))
+
+        def run_generation(prompt, width, height, guidance, num_steps, seed,
+                           img1, img2, img3, img4, num_outputs):
+            uploaded_images = [img for img in [img1, img2, img3, img4] if img is not None]
+
+            print(f"\n📥 [DEBUG] User prompt: {prompt}")
+            prompts = enhance_prompt_with_chatgpt(
+                user_prompt=prompt,
+                num_prompts=num_outputs,
+                reference_images=uploaded_images
+            )
+
+            print(f"\n🧠 [DEBUG] Final Prompt List (len={len(prompts)}):")
+            for idx, p in enumerate(prompts):
+                print(f"  [{idx+1}] {p}")
+
+            while len(prompts) < num_outputs:
+                prompts.append(prompt)
+
+            results = []
+            for i in range(num_outputs):
+                try:
+                    seed_val = int(seed) if seed != -1 else torch.randint(0, 10**8, (1,)).item()
+                    print(f"🧪 [DEBUG] Using seed: {seed_val} for image {i+1}")
+                    gen_image, _ = pipeline.gradio_generate(
+                        prompt=prompts[i],
+                        width=width,
+                        height=height,
+                        guidance=guidance,
+                        num_steps=num_steps,
+                        seed=seed_val,
+                        image_prompt1=img1,
+                        image_prompt2=img2,
+                        image_prompt3=img3,
+                        image_prompt4=img4,
+                    )
+                    print(f"✅ [DEBUG] Image {i+1} generated using prompt: {prompts[i]}")
+                    results.append(gen_image)
+                    results.append(prompts[i])
+                except Exception as e:
+                    print(f"❌ [ERROR] Failed to generate image {i+1}: {e}")
+                    results.append(None)
+                    results.append(f"⚠️ Failed to generate: {e}")
+
+            # Pad to 10 outputs: 5 image + prompt pairs
+            while len(results) < 10:
+                results.append(None if len(results) % 2 == 0 else "")
+
+            return results
+
+        generate_btn.click(
+            fn=run_generation,
+            inputs=[
+                prompt, width, height, guidance, num_steps,
+                seed, image_prompt1, image_prompt2, image_prompt3, image_prompt4, num_outputs
+            ],
+            outputs=outputs
+        )
+
+        example_text = gr.Text("", visible=False, label="Case For:")
+        examples = get_examples("./assets/examples")
+
+        gr.Examples(
+            examples=examples,
+            inputs=[
+                example_text, prompt,
+                image_prompt1, image_prompt2, image_prompt3, image_prompt4,
+                seed, outputs[0]
+            ],
+        )
+
+    return demo
+
+if __name__ == "__main__":
+    from typing import Literal
+    from transformers import HfArgumentParser
+
+    @dataclasses.dataclass
+    class AppArgs:
+        name: Literal["flux-dev", "flux-dev-fp8", "flux-schnell"] = "flux-dev"
+        device: Literal["cuda", "cpu"] = "cuda" if torch.cuda.is_available() else "cpu"
+        offload: bool = dataclasses.field(
+            default=False,
+            metadata={"help": "If True, sequentially offload unused models to CPU"}
+        )
+        port: int = 7860
+
+    parser = HfArgumentParser([AppArgs])
+    args = parser.parse_args_into_dataclasses()[0]
+
+    demo = create_demo(args.name, args.device, args.offload)
+    demo.launch(server_port=args.port)

inference.py

@@ -0,0 +1,151 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates. All rights reserved.
+
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+
+#     http://www.apache.org/licenses/LICENSE-2.0
+
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import dataclasses
+from typing import Literal
+from accelerate import Accelerator
+from transformers import HfArgumentParser
+from PIL import Image
+import json
+import openai
+
+from uno.flux.pipeline import UNOPipeline, preprocess_ref
+from uno.utils.prompt_enhancer import enhance_prompt_with_chatgpt
+
+openai.api_key = os.getenv("OPENAI_API_KEY")
+
+def horizontal_concat(images):
+    widths, heights = zip(*(img.size for img in images))
+    total_width = sum(widths)
+    max_height = max(heights)
+    new_im = Image.new('RGB', (total_width, max_height))
+    x_offset = 0
+    for img in images:
+        new_im.paste(img, (x_offset, 0))
+        x_offset += img.size[0]
+    return new_im
+
+@dataclasses.dataclass
+class InferenceArgs:
+    prompt: str | None = None
+    image_paths: list[str] | None = None
+    eval_json_path: str | None = None
+    offload: bool = False
+    num_images_per_prompt: int = 1
+    model_type: Literal["flux-dev", "flux-dev-fp8", "flux-schnell"] = "flux-dev"
+    width: int = 512
+    height: int = 512
+    ref_size: int = -1
+    num_steps: int = 25
+    guidance: float = 4
+    seed: int = 3407
+    save_path: str = "output/inference"
+    only_lora: bool = True
+    concat_refs: bool = False
+    lora_rank: int = 512
+    data_resolution: int = 512
+    pe: Literal['d', 'h', 'w', 'o'] = 'd'
+
+def main(args: InferenceArgs):
+    accelerator = Accelerator()
+    pipeline = UNOPipeline(
+        args.model_type,
+        accelerator.device,
+        args.offload,
+        only_lora=args.only_lora,
+        lora_rank=args.lora_rank
+    )
+
+    assert args.prompt is not None or args.eval_json_path is not None, \
+        "Please provide either prompt or eval_json_path"
+
+    if args.eval_json_path:
+        with open(args.eval_json_path, "rt") as f:
+            data_dicts = json.load(f)
+        data_root = os.path.dirname(args.eval_json_path)
+    else:
+        data_root = "./"
+        data_dicts = [{"prompt": args.prompt, "image_paths": args.image_paths}]
+
+    for i, data_dict in enumerate(data_dicts):
+        try:
+            ref_imgs = [
+                Image.open(os.path.join(data_root, img_path))
+                for img_path in data_dict["image_paths"]
+            ]
+        except Exception as e:
+            print(f"❌ [ERROR] Failed to load reference images: {e}")
+            continue
+
+        if args.ref_size == -1:
+            args.ref_size = 512 if len(ref_imgs) == 1 else 320
+        ref_imgs = [preprocess_ref(img, args.ref_size) for img in ref_imgs]
+
+        print(f"\n🔧 [DEBUG] Enhancing prompt: '{data_dict['prompt']}'")
+        enhanced_prompts = enhance_prompt_with_chatgpt(
+            user_prompt=data_dict["prompt"],
+            num_prompts=args.num_images_per_prompt,
+            reference_images=ref_imgs
+        )
+
+        # Pad if needed
+        while len(enhanced_prompts) < args.num_images_per_prompt:
+            print(f"⚠️ [DEBUG] Padding prompts: returning user prompt as fallback.")
+            enhanced_prompts.append(data_dict["prompt"])
+
+        for j in range(args.num_images_per_prompt):
+            if (i * args.num_images_per_prompt + j) % accelerator.num_processes != accelerator.process_index:
+                continue
+
+            prompt_j = enhanced_prompts[j]
+            print(f"\n--- Generating image [{i}_{j}] ---")
+            print(f"Enhanced Prompt: {prompt_j}")
+            print(f"Image paths: {data_dict['image_paths']}")
+            print(f"Seed: {args.seed + j}")
+            print(f"Resolution: {args.width}x{args.height}")
+            print("------------------------------")
+
+            try:
+                image_gen = pipeline(
+                    prompt=prompt_j,
+                    width=args.width,
+                    height=args.height,
+                    guidance=args.guidance,
+                    num_steps=args.num_steps,
+                    seed=args.seed + j,
+                    ref_imgs=ref_imgs,
+                    pe=args.pe,
+                )
+
+                if args.concat_refs:
+                    image_gen = horizontal_concat([image_gen, *ref_imgs])
+
+                os.makedirs(args.save_path, exist_ok=True)
+                image_gen.save(os.path.join(args.save_path, f"{i}_{j}.png"))
+
+                # Save generation context
+                args_dict = vars(args)
+                args_dict['prompt'] = prompt_j
+                args_dict['image_paths'] = data_dict["image_paths"]
+                with open(os.path.join(args.save_path, f"{i}_{j}.json"), 'w') as f:
+                    json.dump(args_dict, f, indent=4)
+
+            except Exception as e:
+                print(f"❌ [ERROR] Failed to generate or save image {i}_{j}: {e}")
+
+if __name__ == "__main__":
+    parser = HfArgumentParser([InferenceArgs])
+    args = parser.parse_args_into_dataclasses()[0]
+    main(args)

requirements.txt

@@ -0,0 +1,16 @@
+--extra-index-url https://download.pytorch.org/whl/cu124
+torch==2.4.0
+torchvision==0.19.0
+
+accelerate==1.1.1
+deepspeed==0.14.4
+einops==0.8.0
+transformers==4.43.3
+huggingface-hub
+diffusers==0.30.1
+sentencepiece==0.2.0
+gradio==5.22.0
+
+
+openai>=1.14.0
+python-dotenv>=1.0.1

train.py

@@ -0,0 +1,482 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates. All rights reserved.
+
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+
+#     http://www.apache.org/licenses/LICENSE-2.0
+
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import dataclasses
+import gc
+import itertools
+import logging
+import os
+import random
+from copy import deepcopy
+from typing import TYPE_CHECKING, Literal
+
+import torch
+import torch.nn.functional as F
+import transformers
+from accelerate import Accelerator, DeepSpeedPlugin
+from accelerate.logging import get_logger
+from accelerate.utils import set_seed
+from diffusers.optimization import get_scheduler
+from einops import rearrange
+from PIL import Image
+from safetensors.torch import load_file
+from torch.utils.data import DataLoader
+from tqdm import tqdm
+
+from uno.dataset.uno import FluxPairedDatasetV2
+from uno.flux.sampling import denoise, get_noise, get_schedule, prepare_multi_ip, unpack
+from uno.flux.util import load_ae, load_clip, load_flow_model, load_t5, set_lora
+
+if TYPE_CHECKING:
+    from uno.flux.model import Flux
+    from uno.flux.modules.autoencoder import AutoEncoder
+    from uno.flux.modules.conditioner import HFEmbedder
+
+logger = get_logger(__name__)
+
+def get_models(name: str, device, offload: bool=False):
+    t5 = load_t5(device, max_length=512)
+    clip = load_clip(device)
+    model = load_flow_model(name, device="cpu")
+    vae = load_ae(name, device="cpu" if offload else device)
+    return model, vae, t5, clip
+
+def inference(
+    batch: dict,
+    model: "Flux", t5: "HFEmbedder", clip: "HFEmbedder", ae: "AutoEncoder",
+    accelerator: Accelerator,
+    seed: int = 0,
+    pe: Literal["d", "h", "w", "o"] = "d"
+) -> Image.Image:
+    ref_imgs = batch["ref_imgs"]
+    prompt = batch["txt"]
+    neg_prompt = ''
+    width, height = 512, 512
+    num_steps = 25
+    x = get_noise(
+        1, height, width,
+        device=accelerator.device,
+        dtype=torch.bfloat16,
+        seed=seed + accelerator.process_index
+    )
+    timesteps = get_schedule(
+        num_steps,
+        (width // 8) * (height // 8) // (16 * 16),
+        shift=True,
+    )
+    with torch.no_grad():
+        ref_imgs = [
+            ae.encode(ref_img_.to(accelerator.device, torch.float32)).to(torch.bfloat16)
+            for ref_img_ in ref_imgs
+        ]
+        inp_cond = prepare_multi_ip(
+            t5=t5, clip=clip, img=x, prompt=prompt,
+            ref_imgs=ref_imgs,
+            pe=pe
+        )
+        neg_inp_cond = prepare_multi_ip(
+            t5=t5, clip=clip, img=x, prompt=neg_prompt,
+            ref_imgs=ref_imgs,
+            pe=pe
+        )
+
+        x = denoise(
+            model,
+            **inp_cond,
+            timesteps=timesteps,
+            guidance=4,
+            timestep_to_start_cfg=30,
+            neg_txt=neg_inp_cond['txt'],
+            neg_txt_ids=neg_inp_cond['txt_ids'],
+            neg_vec=neg_inp_cond['vec'],
+            true_gs=3.5,
+            image_proj=None,
+            neg_image_proj=None,
+            ip_scale=1,
+            neg_ip_scale=1
+        )
+
+        x = unpack(x.float(), height, width)
+        x = ae.decode(x)
+
+    x1 = x.clamp(-1, 1)
+    x1 = rearrange(x1[-1], "c h w -> h w c")
+    output_img = Image.fromarray((127.5 * (x1 + 1.0)).cpu().byte().numpy())
+
+    return output_img
+
+
+def resume_from_checkpoint(
+    resume_from_checkpoint: str | None | Literal["latest"],
+    project_dir: str,
+    accelerator: Accelerator,
+    dit: "Flux",
+    optimizer: torch.optim.Optimizer,
+    lr_scheduler: torch.optim.lr_scheduler.LRScheduler,
+    dit_ema_dict: dict | None = None,
+) -> tuple["Flux", torch.optim.Optimizer, torch.optim.lr_scheduler.LRScheduler, dict | None, int]:
+    # Potentially load in the weights and states from a previous save
+    if resume_from_checkpoint is None:
+        return dit, optimizer, lr_scheduler, dit_ema_dict, 0
+
+    if resume_from_checkpoint == "latest":
+        # Get the most recent checkpoint
+        dirs = os.listdir(project_dir)
+        dirs = [d for d in dirs if d.startswith("checkpoint")]
+        dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
+        if len(dirs) == 0:
+            accelerator.print(
+                f"Checkpoint '{resume_from_checkpoint}' does not exist. Starting a new training run."
+            )
+            return dit, optimizer, lr_scheduler, dit_ema_dict, 0
+        path = dirs[-1]
+    else:
+        path = os.path.basename(resume_from_checkpoint)
+        
+
+    accelerator.print(f"Resuming from checkpoint {path}")
+    lora_state = load_file(os.path.join(project_dir, path, 'dit_lora.safetensors'), device=accelerator.device)
+    unwarp_dit = accelerator.unwrap_model(dit)
+    unwarp_dit.load_state_dict(lora_state, strict=False)
+    if dit_ema_dict is not None:
+        dit_ema_dict = load_file(
+            os.path.join(project_dir, path, 'dit_lora_ema.safetensors'),
+            device=accelerator.device
+        )
+        if dit is not unwarp_dit:
+            dit_ema_dict = {f"module.{k}": v for k, v in dit_ema_dict.items() if k in unwarp_dit.state_dict()}
+
+    global_step = int(path.split("-")[1])
+    
+    return dit, optimizer, lr_scheduler, dit_ema_dict, global_step
+
+@dataclasses.dataclass
+class TrainArgs:
+    ## accelerator
+    project_dir: str | None = None
+    mixed_precision: Literal["no", "fp16", "bf16"] = "bf16"
+    gradient_accumulation_steps: int = 1,
+    seed: int = 42
+    wandb_project_name: str | None = None
+    wandb_run_name: str | None = None
+
+    ## model
+    model_name: Literal["flux", "flux-schnell"] = "flux"
+    lora_rank: int = 512
+    double_blocks_indices: list[int] | None = dataclasses.field(
+        default=None,
+        metadata={"help": "Indices of double blocks to apply LoRA. None means all double blocks."}
+    )
+    single_blocks_indices: list[int] | None = dataclasses.field(
+        default=None,
+        metadata={"help": "Indices of double blocks to apply LoRA. None means all single blocks."}
+    )
+    pe: Literal["d", "h", "w", "o"] = "d",
+    gradient_checkpoint: bool = False
+
+    ## ema
+    ema: bool = False
+    ema_interval: int = 1
+    ema_decay: float = 0.99
+
+
+    ## optimizer
+    learning_rate: float = 1e-2
+    adam_betas: list[float] = dataclasses.field(default_factory=lambda: [0.9, 0.999])
+    adam_eps: float = 1e-8
+    adam_weight_decay: float = 0.01
+
+    ## lr_scheduler
+    lr_scheduler: str = "constant"
+    lr_warmup_steps: int = 100
+    max_train_steps: int = 100000
+
+    ## dataloader
+    train_data_json: str = "datasets/dreambench_singleip.json" # TODO: change to your own dataset, or use one data syenthsize pipeline comming in the future. stay tuned 
+    batch_size: int = 1
+    text_dropout: float = 0.1
+    resolution: int = 512
+    resolution_ref: int | None = None
+
+    eval_data_json: str = "datasets/dreambench_singleip.json"
+    eval_batch_size: int = 1
+
+    ## misc
+    resume_from_checkpoint: str | None | Literal["latest"] = None
+    checkpointing_steps: int = 1000
+
+def main(
+    args: TrainArgs,
+):
+    ## accelerator
+    deepspeed_plugins = {
+        "dit": DeepSpeedPlugin(hf_ds_config='config/deepspeed/zero2_config.json'),
+        "t5": DeepSpeedPlugin(hf_ds_config='config/deepspeed/zero3_config.json'),
+        "clip": DeepSpeedPlugin(hf_ds_config='config/deepspeed/zero3_config.json')
+    }
+    accelerator = Accelerator(
+        project_dir=args.project_dir,
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision,
+        deepspeed_plugins=deepspeed_plugins,
+        log_with="wandb",
+    )
+    set_seed(args.seed, device_specific=True)
+    accelerator.init_trackers(
+        project_name=args.wandb_project_name,
+        config=args.__dict__,
+        init_kwargs={
+            "wandb": {
+                "name": args.wandb_run_name,
+                "dir": accelerator.project_dir,
+            },
+        },
+    )
+    weight_dtype = {
+        "fp16": torch.float16,
+        "bf16": torch.bfloat16,
+        "no": torch.float32,
+    }.get(accelerator.mixed_precision, torch.float32)
+
+    ## logger
+    logging.basicConfig(
+        format=f"[RANK {accelerator.process_index}] " + "%(asctime)s - %(levelname)s - %(name)s - %(message)s",
+        datefmt="%m/%d/%Y %H:%M:%S",
+        level=logging.INFO,
+        force=True
+    )
+    logger.info(accelerator.state)
+    logger.info("Training script launched", main_process_only=False)
+
+    ## model
+    dit, vae, t5, clip = get_models(
+        name=args.model_name,
+        device=accelerator.device,
+    )
+    
+    vae.requires_grad_(False)
+    t5.requires_grad_(False)
+    clip.requires_grad_(False)
+
+    dit.requires_grad_(False)
+    dit = set_lora(dit, args.lora_rank, args.double_blocks_indices, args.single_blocks_indices, accelerator.device)
+    dit.train()
+    dit.gradient_checkpointing = args.gradient_checkpoint
+
+    ## optimizer and lr scheduler
+    optimizer = torch.optim.AdamW(
+        [p for p in dit.parameters() if p.requires_grad],
+        lr=args.learning_rate,
+        betas=args.adam_betas,
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_eps,
+    )
+    
+    lr_scheduler = get_scheduler(
+        args.lr_scheduler,
+        optimizer=optimizer,
+        num_warmup_steps=args.lr_warmup_steps * accelerator.num_processes,
+        num_training_steps=args.max_train_steps * accelerator.num_processes,
+    )
+
+    # dataloader
+    dataset = FluxPairedDatasetV2(
+        data_json=args.train_data_json,
+        resolution=args.resolution, resolution_ref=args.resolution_ref
+    )
+    dataloader = DataLoader(dataset, batch_size=args.batch_size, shuffle=True)
+    eval_dataset = FluxPairedDatasetV2(
+        data_json=args.eval_data_json,
+        resolution=args.resolution, resolution_ref=args.resolution_ref
+    )   
+    eval_dataloader = DataLoader(eval_dataset, batch_size=args.eval_batch_size, shuffle=False)
+
+    dataloader = accelerator.prepare_data_loader(dataloader)
+    eval_dataloader = accelerator.prepare_data_loader(eval_dataloader)
+    dataloader = itertools.cycle(dataloader)  # as infinite fetch data loader
+
+    ## parallel
+    dit = accelerator.prepare_model(dit)
+    optimizer = accelerator.prepare_optimizer(optimizer)
+    lr_scheduler = accelerator.prepare_scheduler(lr_scheduler)
+    accelerator.state.select_deepspeed_plugin("t5")
+    t5 = accelerator.prepare_model(t5)
+    accelerator.state.select_deepspeed_plugin("clip")
+    clip = accelerator.prepare_model(clip)
+    
+    ## ema
+    dit_ema_dict = {
+        k: deepcopy(v).requires_grad_(False) for k, v in dit.named_parameters() if v.requires_grad
+    } if args.ema else None
+
+    ## resume
+    (
+        dit,
+        optimizer,
+        lr_scheduler,
+        dit_ema_dict,
+        global_step
+    ) = resume_from_checkpoint(
+        args.resume_from_checkpoint,
+        project_dir=args.project_dir,
+        accelerator=accelerator,
+        dit=dit,
+        optimizer=optimizer,
+        lr_scheduler=lr_scheduler,
+        dit_ema_dict=dit_ema_dict
+    )
+
+    ## noise scheduler
+    timesteps = get_schedule(
+        999,
+        (args.resolution // 8) * (args.resolution // 8) // 4,
+        shift=True,
+    )
+    timesteps = torch.tensor(timesteps, device=accelerator.device)
+    total_batch_size = args.batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+
+    logger.info("***** Running training *****")
+    logger.info(f"  Instantaneous batch size per device = {args.batch_size}")
+    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
+    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
+    logger.info(f"  Total optimization steps = {args.max_train_steps}")
+    logger.info(f"  Total validation prompts = {len(eval_dataloader)}")
+
+    progress_bar = tqdm(
+        range(0, args.max_train_steps),
+        initial=global_step,
+        desc="Steps",
+        total=args.max_train_steps,
+        disable=not accelerator.is_local_main_process,
+    )
+
+    train_loss = 0.0
+    while global_step < (args.max_train_steps):
+        batch = next(dataloader)
+        prompts = [txt_ if random.random() > args.text_dropout else "" for txt_ in batch["txt"]]
+        img = batch["img"]
+        ref_imgs = batch["ref_imgs"]
+
+        with torch.no_grad():
+            x_1 = vae.encode(img.to(accelerator.device).to(torch.float32))
+            x_ref = [vae.encode(ref_img.to(accelerator.device).to(torch.float32)) for ref_img in ref_imgs]
+            inp = prepare_multi_ip(t5=t5, clip=clip, img=x_1, prompt=prompts, ref_imgs=tuple(x_ref), pe=args.pe)
+            x_1 = rearrange(x_1, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=2, pw=2)
+            x_ref = [rearrange(x, "b c (h ph) (w pw) -> b (h w) (c ph pw)", ph=2, pw=2) for x in x_ref]
+
+            bs = img.shape[0]
+            t = torch.randint(0, 1000, (bs,), device=accelerator.device)
+            t = timesteps[t]
+            x_0 = torch.randn_like(x_1, device=accelerator.device)
+            x_t = (1 - t[:, None, None]) * x_1 + t[:, None, None] * x_0
+            guidance_vec = torch.full((x_t.shape[0],), 1, device=x_t.device, dtype=x_t.dtype)
+        
+        with accelerator.accumulate(dit):
+            # Predict the noise residual and compute loss
+            model_pred = dit(
+                img=x_t.to(weight_dtype),
+                img_ids=inp['img_ids'].to(weight_dtype),
+                ref_img=[x.to(weight_dtype) for x in x_ref],
+                ref_img_ids=[ref_img_id.to(weight_dtype) for ref_img_id in inp['ref_img_ids']],
+                txt=inp['txt'].to(weight_dtype),
+                txt_ids=inp['txt_ids'].to(weight_dtype),
+                y=inp['vec'].to(weight_dtype),
+                timesteps=t.to(weight_dtype),
+                guidance=guidance_vec.to(weight_dtype)
+            )
+
+            loss = F.mse_loss(model_pred.float(), (x_0 - x_1).float(), reduction="mean")
+
+            # Gather the losses across all processes for logging (if we use distributed training).
+            avg_loss = accelerator.gather(loss.repeat(args.train_batch_size)).mean()
+            train_loss += avg_loss.item() / args.gradient_accumulation_steps
+
+            # Backpropagate
+            accelerator.backward(loss)
+            if accelerator.sync_gradients:
+                accelerator.clip_grad_norm_(dit.parameters(), args.max_grad_norm)
+            optimizer.step()
+            lr_scheduler.step()
+            optimizer.zero_grad()
+
+        # Checks if the accelerator has performed an optimization step behind the scenes
+        if accelerator.sync_gradients:
+            progress_bar.update(1)
+            global_step += 1
+            accelerator.log({"train_loss": train_loss}, step=global_step)
+            train_loss = 0.0
+
+        if accelerator.sync_gradients and dit_ema_dict is not None and global_step % args.ema_interval == 0:
+            src_dict = dit.state_dict()
+            for tgt_name in dit_ema_dict:
+                dit_ema_dict[tgt_name].data.lerp_(src_dict[tgt_name].to(dit_ema_dict[tgt_name]), 1 - args.ema_decay)
+
+        if accelerator.sync_gradients and accelerator.is_main_process and global_step % args.checkpointing_steps == 0:
+            logger.info(f"saving checkpoint in {global_step=}")
+            save_path = os.path.join(args.project_dir, f"checkpoint-{global_step}")
+            os.makedirs(save_path, exist_ok=True)
+
+            # save
+            accelerator.wait_for_everyone()
+            unwrapped_model = accelerator.unwrap_model(dit)
+            unwrapped_model_state = unwrapped_model.state_dict()
+            unwrapped_model_state = {k: v for k, v in unwrapped_model_state.items() if v.requires_grad}
+
+            accelerator.save(
+                unwrapped_model_state,
+                os.path.join(save_path, 'dit_lora.safetensors'),
+                safe_serialization=True
+            )
+            unwrapped_opt = accelerator.unwrap_model(optimizer)
+            accelerator.save(unwrapped_opt.state_dict(), os.path.join(save_path, 'optimizer.bin'))
+            logger.info(f"Saved state to {save_path}")
+
+            if args.ema:
+                accelerator.save(
+                    {k.split("module.")[-1]: v for k, v in dit_ema_dict.items()},
+                    os.path.join(save_path, 'dit_lora_ema.safetensors')
+                )
+
+            # validate
+            dit.eval()
+            torch.set_grad_enabled(False)
+            for i, batch in enumerate(eval_dataloader):
+                result = inference(batch, dit, t5, clip, vae, accelerator, seed=0)
+                accelerator.log({f"eval_gen_{i}": result}, step=global_step)
+
+
+            if args.ema:
+                original_state_dict = dit.state_dict()
+                dit.load_state_dict(dit_ema_dict, strict=False)
+                for batch in eval_dataloader:
+                    result = inference(batch, dit, t5, clip, vae, accelerator, seed=0)
+                    accelerator.log({f"eval_ema_gen_{i}": result}, step=global_step)
+                dit.load_state_dict(original_state_dict, strict=False)
+            
+            torch.cuda.empty_cache()
+            gc.collect()
+            torch.set_grad_enabled(True)
+            dit.train()
+            accelerator.wait_for_everyone()
+
+        logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
+        progress_bar.set_postfix(**logs)
+    
+    accelerator.wait_for_everyone()
+    accelerator.end_training()
+
+if __name__ == "__main__":
+    parser = transformers.HfArgumentParser([TrainArgs])
+    args_tuple = parser.parse_args_into_dataclasses(args_file_flag="--config")
+    main(*args_tuple)