initial commit

.gitattributes

@@ -33,3 +33,6 @@ saved_model/**/* 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
+*.webp filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,235 @@
+# Created by https://www.toptal.com/developers/gitignore/api/macos,python
+# Edit at https://www.toptal.com/developers/gitignore?templates=macos,python
+
+### macOS ###
+# General
+.DS_Store
+.AppleDouble
+.LSOverride
+
+# Icon must end with two \r
+Icon
+
+
+# Thumbnails
+._*
+
+# Files that might appear in the root of a volume
+.DocumentRevisions-V100
+.fseventsd
+.Spotlight-V100
+.TemporaryItems
+.Trashes
+.VolumeIcon.icns
+.com.apple.timemachine.donotpresent
+
+# Directories potentially created on remote AFP share
+.AppleDB
+.AppleDesktop
+Network Trash Folder
+Temporary Items
+.apdisk
+
+### macOS Patch ###
+# iCloud generated files
+*.icloud
+
+### Python ###
+# 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
+
+# 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/#use-with-ide
+.pdm.toml
+
+# 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/
+
+### Python Patch ###
+# Poetry local configuration file - https://python-poetry.org/docs/configuration/#local-configuration
+poetry.toml
+
+# ruff
+.ruff_cache/
+
+# LSP config files
+pyrightconfig.json
+
+# End of https://www.toptal.com/developers/gitignore/api/macos,python
+
+local_scripts/
+
+omnigen2/utils/vpn_utils.py
+
+test_tokenizer.py
+save_pipeline.py
+app.sh
+logs/
+results/
+test_jsonl*
+pbs_files/
+convert_ckpt_to_pipeline.py
+inference_test_efficiency.py
+upload_pipeline*
+example_images_resized/
+example_t2i_test_efficiency*.sh
+example_edit_test_efficiency*.sh
+example_in_context_generation_test_efficiency*.sh
+intro*
+resize_example_images.py
+save_pipeline.py
+outputs_gradio/*
+test.py
+data_configs/test
+scripts/test

README.md

@@ -1,6 +1,6 @@
 ---
 title: UMO OmniGen2
-emoji: 📊
+emoji: ⚡️
 colorFrom: green
 colorTo: purple
 sdk: gradio

app.py

@@ -0,0 +1,485 @@
+# Copyright (c) 2025 Bytedance Ltd. and/or its affiliates. All rights reserved.
+# Copyright (c) VectorSpaceLab and 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 dotenv
+dotenv.load_dotenv(override=True)
+
+import gradio as gr
+import spaces
+
+import argparse
+import json
+import random
+from datetime import datetime
+from glob import glob
+from typing import Literal
+
+import torch
+from torchvision.transforms.functional import to_pil_image, to_tensor
+
+from accelerate import Accelerator
+from peft import LoraConfig
+from safetensors.torch import load_file
+
+from omnigen2.pipelines.omnigen2.pipeline_omnigen2 import OmniGen2Pipeline
+from omnigen2.models.transformers.transformer_omnigen2 import OmniGen2Transformer2DModel
+from omnigen2.schedulers.scheduling_flow_match_euler_discrete import FlowMatchEulerDiscreteScheduler
+from omnigen2.schedulers.scheduling_dpmsolver_multistep import DPMSolverMultistepScheduler
+from omnigen2.utils.img_util import create_collage
+
+NEGATIVE_PROMPT = "(((deformed))), blurry, over saturation, bad anatomy, disfigured, poorly drawn face, mutation, mutated, (extra_limb), (ugly), (poorly drawn hands), fused fingers, messy drawing, broken legs censor, censored, censor_bar"
+ROOT_DIR = "projects/OmniGen2"
+SAVE_DIR = "output/gradio"
+
+pipeline = None
+accelerator = None
+save_images = False
+
+def load_pipeline(accelerator, weight_dtype, args):
+    pipeline = OmniGen2Pipeline.from_pretrained(
+        args.model_path,
+        torch_dtype=weight_dtype,
+        trust_remote_code=True,
+    )
+    pipeline.transformer = OmniGen2Transformer2DModel.from_pretrained(
+        args.model_path,
+        subfolder="transformer",
+        torch_dtype=weight_dtype,
+    )
+    if args.lora_path is not None:
+        target_modules = ["to_k", "to_q", "to_v", "to_out.0"]
+        lora_config = LoraConfig(
+            r=512,
+            lora_alpha=512,
+            lora_dropout=0,
+            init_lora_weights="gaussian",
+            target_modules=target_modules,
+        )
+        pipeline.transformer.add_adapter(lora_config)
+        lora_state_dict = load_file(args.lora_path, device=accelerator.device.__str__())
+        pipeline.transformer.load_state_dict(lora_state_dict, strict=False)
+        pipeline.transformer.fuse_lora(lora_scale=1, safe_fusing=False, adapter_names=["default"])
+        pipeline.transformer.unload_lora()
+    if args.enable_sequential_cpu_offload:
+        pipeline.enable_sequential_cpu_offload()
+    elif args.enable_model_cpu_offload:
+        pipeline.enable_model_cpu_offload()
+    else:
+        pipeline = pipeline.to(accelerator.device)
+    return pipeline
+
+@spaces.GPU(duration=120)
+def run(
+    instruction,
+    width_input,
+    height_input,
+    image_input_1,
+    image_input_2,
+    image_input_3,
+    scheduler: Literal["euler", "dpmsolver++"] = "euler",
+    num_inference_steps: int = 50,
+    negative_prompt: str = NEGATIVE_PROMPT,
+    guidance_scale_input: float = 5.0,
+    img_guidance_scale_input: float = 2.0,
+    cfg_range_start: float = 0.0,
+    cfg_range_end: float = 1.0,
+    num_images_per_prompt: int = 1,
+    max_input_image_side_length: int = 2048,
+    max_pixels: int = 1024 * 1024,
+    seed_input: int = -1,
+    align_res: bool = True,
+    progress=gr.Progress(),
+):
+    input_images = [image_input_1, image_input_2, image_input_3]
+    input_images = [img for img in input_images if img is not None]
+
+    if len(input_images) == 0:
+        input_images = None
+
+    if seed_input == -1:
+        seed_input = random.randint(0, 2**16 - 1)
+
+    generator = torch.Generator(device=accelerator.device).manual_seed(seed_input)
+
+    def progress_callback(cur_step, timesteps):
+        frac = (cur_step + 1) / float(timesteps)
+        progress(frac)
+
+    if scheduler == 'euler':
+        pipeline.scheduler = FlowMatchEulerDiscreteScheduler()
+    elif scheduler == 'dpmsolver++':
+        pipeline.scheduler = DPMSolverMultistepScheduler(
+            algorithm_type="dpmsolver++",
+            solver_type="midpoint",
+            solver_order=2,
+            prediction_type="flow_prediction",
+        )
+
+    results = pipeline(
+        prompt=instruction,
+        input_images=input_images,
+        width=width_input,
+        height=height_input,
+        align_res=align_res,
+        max_input_image_side_length=max_input_image_side_length,
+        max_pixels=max_pixels,
+        num_inference_steps=num_inference_steps,
+        max_sequence_length=1024,
+        text_guidance_scale=guidance_scale_input,
+        image_guidance_scale=img_guidance_scale_input,
+        cfg_range=(cfg_range_start, cfg_range_end),
+        negative_prompt=negative_prompt,
+        num_images_per_prompt=num_images_per_prompt,
+        generator=generator,
+        output_type="pil",
+        step_func=progress_callback,
+    )
+
+    progress(1.0)
+
+    vis_images = [to_tensor(image) * 2 - 1 for image in results.images]
+    output_image = create_collage(vis_images)
+
+    if save_images:
+        # Create outputs directory if it doesn't exist
+        output_dir = SAVE_DIR
+        os.makedirs(output_dir, exist_ok=True)
+
+        # Generate unique filename with timestamp
+        timestamp = datetime.now().strftime("%Y_%m_%d-%H_%M_%S")
+
+        # Generate unique filename with timestamp
+        output_path = os.path.join(output_dir, f"{timestamp}_seed{seed_input}_{instruction[:20]}.png")
+        # Save the image
+        output_image.save(output_path)
+
+        # Save All Generated Images
+        if len(results.images) > 1:
+            for i, image in enumerate(results.images):
+                image_name, ext = os.path.splitext(output_path)
+                image.save(f"{image_name}_{i}{ext}")
+    return output_image
+
+
+def get_examples(base_dir="assets/examples/OmniGen2"):
+    example_keys = ["instruction", "width_input", "height_input", "image_input_1", "image_input_2", "image_input_3", "seed_input", "align_res", "output_image", "output_image_OmniGen2"]
+    examples = []
+    example_configs = glob(os.path.join(base_dir, "*", "config.json"))
+    for config_path in example_configs:
+        with open(config_path, "r", encoding="utf-8") as f:
+            config = json.load(f)
+        _example = [config.get(k, None) for k in example_keys]
+        examples.append(_example)
+    return examples
+
+
+with open("assets/logo.svg", "r", encoding="utf-8") as svg_file:
+    logo_content = svg_file.read()
+title = f"""
+<div style="display: flex; align-items: center; justify-content: center;">
+    <span style="transform: scale(0.7);margin-right: -5px;">{logo_content}</span>  
+    <span style="font-size: 1.8em;margin-left: -10px;font-weight: bold; font-family: Gill Sans;">UMO (based on OmniGen2) by UXO Team</span>
+</div>
+""".strip()
+
+badges_text = r"""
+<div style="text-align: center; display: flex; justify-content: center; gap: 5px;">
+<a href="https://github.com/bytedance/UMO"><img alt="Build" src="https://img.shields.io/github/stars/bytedance/UMO"></a> 
+<a href="https://bytedance.github.io/UMO/"><img alt="Build" src="https://img.shields.io/badge/Project%20Page-UMO-yellow"></a> 
+<a href="https://arxiv.org/abs/25xx.xxxxx"><img alt="Build" src="https://img.shields.io/badge/arXiv%20paper-UMO-b31b1b.svg"></a>
+<a href="https://huggingface.co/bytedance-research/UMO"><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/UMO-FLUX"><img src="https://img.shields.io/static/v1?label=%F0%9F%A4%97%20Hugging%20Face&message=demo&color=orange"></a>
+</div>
+""".strip()
+
+tips = """
+📌 ***UMO*** is a **U**nified **M**ulti-identity **O**ptimization framework to *boost the multi-ID fidelity and mitigate confusion* for image customization model, and the latest addition to the UXO family (<a href='https://github.com/bytedance/UMO' target='_blank'> UMO</a>, <a href='https://github.com/bytedance/USO' target='_blank'> USO</a> and <a href='https://github.com/bytedance/UNO' target='_blank'> UNO</a>).
+
+🎨 UMO in the demo is trained based on <a href='https://github.com/VectorSpaceLab/OmniGen2' target='_blank'> OmniGen2</a>.
+
+💡 We provide step-by-step instructions in our <a href='https://github.com/bytedance/UMO' target='_blank'> Github Repo</a>. Additionally, try the examples and comparison provided below the demo to quickly get familiar with UMO and spark your creativity!
+
+<details>
+<summary style="cursor: pointer; color: #d34c0e; font-weight: 500;"> ⚡️ Tips from the based OmniGen2</summary>
+
+- Image Quality: Use high-resolution images (**at least 512x512 recommended**).
+- Be Specific: Instead of "Add bird to desk", try "Add the bird from image 1 to the desk in image 2".
+- Use English: English prompts currently yield better results.
+- Increase image_guidance_scale for better consistency with the reference image:
+    - Image Editing: 1.3 - 2.0
+    - In-context Generation: 2.0 - 3.0
+- For in-context edit (edit based multiple images), we recommend using the following prompt format: "Edit the first image: add/replace (the [object] with) the [object] from the second image. [descripton for your target image]."
+- For example: "Edit the first image: add the man from the second image. The man is talking with a woman in the kitchen"
+""".strip()
+
+article = """
+```bibtex
+@article{cheng2025umo,
+  title={UMO: Scaling Multi-Identity Consistency for Image Customization via Matching Reward},
+  author={Cheng, Yufeng and Wu, Wenxu and Wu, Shaojin and Huang, Mengqi and Ding, Fei and He, Qian},
+  journal={arXiv preprint arXiv:25xx.xxxxx},
+  year={2025}
+}
+```
+""".strip()
+
+star = f"""
+If UMO is helpful, please help to ⭐ our <a href='https://github.com/bytedance/UMO' target='_blank'> Github Repo</a> or cite our paper. Thanks a lot!
+{article}
+"""
+
+
+def main(args):
+
+    # Gradio
+    with gr.Blocks() as demo:
+        gr.Markdown(title)
+        gr.Markdown(badges_text)
+        gr.Markdown(tips)
+
+        with gr.Row():
+            with gr.Column():
+                # text prompt
+                instruction = gr.Textbox(
+                    label='Enter your prompt',
+                    info='Use "first/second image" or “第一张图/第二张图” as reference.',
+                    placeholder="Type your prompt here...",
+                )
+
+                with gr.Row(equal_height=True):
+                    # input images
+                    image_input_1 = gr.Image(label="First Image", type="pil")
+                    image_input_2 = gr.Image(label="Second Image", type="pil")
+                    image_input_3 = gr.Image(label="Third Image", type="pil")
+
+                generate_button = gr.Button("Generate Image")
+
+                negative_prompt = gr.Textbox(
+                    label="Enter your negative prompt",
+                    placeholder="Type your negative prompt here...",
+                    value=NEGATIVE_PROMPT,
+                )
+
+                # slider
+                with gr.Row(equal_height=True):
+                    height_input = gr.Slider(
+                        label="Height", minimum=256, maximum=2048, value=1024, step=128
+                    )
+                    width_input = gr.Slider(
+                        label="Width", minimum=256, maximum=2048, value=1024, step=128
+                    )
+                
+                with gr.Accordion("Advanced Options", open=False):
+                    with gr.Row(equal_height=True):
+                        align_res = gr.Checkbox(
+                            label="Align Resolution",
+                            info="Align output's resolution with the first reference image. Only valid when there is only one reference image.",
+                            value=True
+                        )
+                    with gr.Row(equal_height=True):
+                        text_guidance_scale_input = gr.Slider(
+                            label="Text Guidance Scale",
+                            minimum=1.0,
+                            maximum=8.0,
+                            value=5.0,
+                            step=0.1,
+                        )
+
+                        image_guidance_scale_input = gr.Slider(
+                            label="Image Guidance Scale",
+                            minimum=1.0,
+                            maximum=3.0,
+                            value=2.0,
+                            step=0.1,
+                        )
+                    with gr.Row(equal_height=True):
+                        cfg_range_start = gr.Slider(
+                            label="CFG Range Start",
+                            minimum=0.0,
+                            maximum=1.0,
+                            value=0.0,
+                            step=0.1,
+                        )
+
+                        cfg_range_end = gr.Slider(
+                            label="CFG Range End",
+                            minimum=0.0,
+                            maximum=1.0,
+                            value=1.0,
+                            step=0.1,
+                        )
+                    
+                    def adjust_end_slider(start_val, end_val):
+                        return max(start_val, end_val)
+
+                    def adjust_start_slider(end_val, start_val):
+                        return min(end_val, start_val)
+                    
+                    cfg_range_start.input(
+                        fn=adjust_end_slider,
+                        inputs=[cfg_range_start, cfg_range_end],
+                        outputs=[cfg_range_end]
+                    )
+
+                    cfg_range_end.input(
+                        fn=adjust_start_slider,
+                        inputs=[cfg_range_end, cfg_range_start],
+                        outputs=[cfg_range_start]
+                    )
+
+                    with gr.Row(equal_height=True):
+                        scheduler_input = gr.Dropdown(
+                            label="Scheduler",
+                            choices=["euler", "dpmsolver++"],
+                            value="euler",
+                            info="The scheduler to use for the model.",
+                        )
+
+                        num_inference_steps = gr.Slider(
+                            label="Inference Steps", minimum=20, maximum=100, value=50, step=1
+                        )
+                    with gr.Row(equal_height=True):
+                        num_images_per_prompt = gr.Slider(
+                            label="Number of images per prompt",
+                            minimum=1,
+                            maximum=4,
+                            value=1,
+                            step=1,
+                        )
+
+                        seed_input = gr.Slider(
+                            label="Seed", minimum=-1, maximum=2147483647, value=-1, step=1
+                        )
+                    with gr.Row(equal_height=True):
+                        max_input_image_side_length = gr.Slider(
+                            label="max_input_image_side_length",
+                            minimum=256,
+                            maximum=2048,
+                            value=2048,
+                            step=256,
+                        )
+                        max_pixels = gr.Slider(
+                            label="max_pixels",
+                            minimum=256 * 256,
+                            maximum=1536 * 1536,
+                            value=1024 * 1024,
+                            step=256 * 256,
+                        )
+
+            with gr.Column():
+                with gr.Column():
+                    # output image
+                    output_image = gr.Image(label="Output Image")
+                    global save_images
+                    save_images = gr.Checkbox(label="Save generated images", value=True)
+                    with gr.Accordion("Examples Comparison with OmniGen2", open=False):
+                        output_image_omnigen2 = gr.Image(label="Generated Image (OmniGen2)")
+
+        gr.Markdown(star)
+        
+        global accelerator
+        global pipeline
+
+        bf16 = True
+        accelerator = Accelerator(mixed_precision="bf16" if bf16 else "no")
+        weight_dtype = torch.bfloat16 if bf16 else torch.float32
+
+        pipeline = load_pipeline(accelerator, weight_dtype, args)
+
+        # click
+        generate_button.click(
+            run,
+            inputs=[
+                instruction,
+                width_input,
+                height_input,
+                image_input_1,
+                image_input_2,
+                image_input_3,
+                scheduler_input,
+                num_inference_steps,
+                negative_prompt,
+                text_guidance_scale_input,
+                image_guidance_scale_input,
+                cfg_range_start,
+                cfg_range_end,
+                num_images_per_prompt,
+                max_input_image_side_length,
+                max_pixels,
+                seed_input,
+                align_res,
+            ],
+            outputs=output_image,
+        )
+
+        gr.Examples(
+            examples=get_examples("assets/examples/OmniGen2"),
+            inputs=[
+                instruction,
+                width_input,
+                height_input,
+                image_input_1,
+                image_input_2,
+                image_input_3,
+                seed_input,
+                align_res,
+                output_image,
+                output_image_omnigen2,
+            ],
+            label="We provide examples for academic research. The vast majority of images used in this demo are either generated or from open-source datasets. If you have any concerns, please contact us, and we will promptly remove any inappropriate content.",
+            examples_per_page=15
+        )
+
+    # launch
+    demo.launch(share=args.share, server_port=args.port, allowed_paths=[ROOT_DIR], server_name=args.server_name)
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--share", action="store_true", help="Share the Gradio app")
+    parser.add_argument(
+        "--port", type=int, default=7860, help="Port to use for the Gradio app"
+    )
+    parser.add_argument(
+        "--server_name", type=str, default=None
+    )
+    parser.add_argument(
+        "--model_path",
+        type=str,
+        default="OmniGen2/OmniGen2",
+        help="Path or HuggingFace name of the model to load."
+    )
+    parser.add_argument(
+        "--enable_model_cpu_offload",
+        action="store_true",
+        help="Enable model CPU offload."
+    )
+    parser.add_argument(
+        "--enable_sequential_cpu_offload",
+        action="store_true",
+        help="Enable sequential CPU offload."
+    )
+    parser.add_argument(
+        "--lora_path",
+        type=str,
+        default=None,
+        help="Path to the LoRA checkpoint to load."
+    )
+    args = parser.parse_args()
+    return args
+
+if __name__ == "__main__":
+    args = parse_args()
+    main(args)

assets/examples/OmniGen2/0/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/0/result.png",
+    "instruction": "The woman from the first image is walking on a red carpet.",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/0/ref.png",
+    "image_input_2": null,
+    "image_input_3": null,
+    "seed_input": 62905,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/0/result_omnigen2.png"
+}

assets/examples/OmniGen2/1/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/1/result.png",
+    "instruction": "Portrait of the woman from the first image.",
+    "width_input": 720,
+    "height_input": 1280,
+    "image_input_1": "assets/examples/OmniGen2/1/ref.webp",
+    "image_input_2": null,
+    "image_input_3": null,
+    "seed_input": 58519,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/1/result_omnigen2.png"
+}

assets/examples/OmniGen2/10/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/10/result.png",
+    "instruction": "Generate a photo of an anime-style figurine placed on a desk. The figurine model should be based on the character photo provided in the attachment, accurately replicating the full-body pose, facial expression, and clothing style of the character in the photo, ensuring the entire figurine is fully presented. The overall design should be exquisite and detailed, soft gradient colors and a delicate texture, leaning towards a Japanese anime style, rich in details, with a realistic quality and beautiful visual appeal.",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/10/ref.JPG",
+    "image_input_2": null,
+    "image_input_3": null,
+    "seed_input": 55301,
+    "align_res": true,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/10/result_omnigen2.png"
+}

assets/examples/OmniGen2/2/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/2/result.png",
+    "instruction": "Change the man from the first image to be wearing black suit.",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/2/ref.png",
+    "image_input_2": null,
+    "image_input_3": null,
+    "seed_input": 7028,
+    "align_res": true,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/2/result_omnigen2.png"
+}

assets/examples/OmniGen2/3/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/3/result.png",
+    "instruction": "Change the style of the first image to Retro comic style",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/3/ref.png",
+    "image_input_2": null,
+    "image_input_3": null,
+    "seed_input": 22067,
+    "align_res": true,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/3/result_omnigen2.png"
+}

assets/examples/OmniGen2/4/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/4/result.png",
+    "instruction": "The man from the first image and the woman from the second image shake hands and look straight ahead, smiling, with a blackboard written \"UMO\" on it.",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/4/ref_1.jpg",
+    "image_input_2": "assets/examples/OmniGen2/4/ref_2.png",
+    "image_input_3": null,
+    "seed_input": 34412,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/4/result_omnigen2.png"
+}

assets/examples/OmniGen2/5/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/5/result.png",
+    "instruction": "make a photo of all these 2 people",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/5/ref_1.jpg",
+    "image_input_2": "assets/examples/OmniGen2/5/ref_2.png",
+    "image_input_3": null,
+    "seed_input": 0,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/5/result_omnigen2.png"
+}

assets/examples/OmniGen2/6/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/6/result.png",
+    "instruction": "The two people shown in the images are sitting in a theater, watching the screen. One person points at the other person.",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/6/ref_1.webp",
+    "image_input_2": "assets/examples/OmniGen2/6/ref_2.webp",
+    "image_input_3": null,
+    "seed_input": 37984,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/6/result_omnigen2.png"
+}

assets/examples/OmniGen2/7/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/7/result.png",
+    "instruction": "The two people shown in the images are drinking wine, their eyes locked in an intense gaze at the camera, seated at an elegantly set table in a dimly lit room. Soft, golden light spills from a nearby chandelier, casting a warm glow over the polished wood surface, while the ambient sound of quiet conversation fills the air.",
+    "width_input": 1024,
+    "height_input": 1024,
+    "image_input_1": "assets/examples/OmniGen2/7/ref_1.webp",
+    "image_input_2": "assets/examples/OmniGen2/7/ref_2.webp",
+    "image_input_3": null,
+    "seed_input": 61593,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/7/result_omnigen2.png"
+}

assets/examples/OmniGen2/8/config.json

@@ -0,0 +1,12 @@
+{
+    "output_image": "assets/examples/OmniGen2/8/result.png",
+    "instruction": "The man from the first image is on the left. The old man with white hair and glasses in the second image is in the middle. The woman from the third image is on the right.",
+    "width_input": 1280,
+    "height_input": 720,
+    "image_input_1": "assets/examples/OmniGen2/8/ref_1.jpg",
+    "image_input_2": "assets/examples/OmniGen2/8/ref_2.jpg",
+    "image_input_3": "assets/examples/OmniGen2/8/ref_3.jpg",
+    "seed_input": 58101,
+    "align_res": false,
+    "output_image_OmniGen2": "assets/examples/OmniGen2/8/result_omnigen2.png"
+}