BoxDiff 🎨 (ICCV 2023)

BoxDiff: Text-to-Image Synthesis with Training-Free Box-Constrained Diffusion

Jinheng Xie¹  Yuexiang Li²  Yawen Huang²  Haozhe Liu^2,3  Wentian Zhang² Yefeng Zheng²  Mike Zheng Shou¹

¹ National University of Singapore  ² Tencent Jarvis Lab  ³ KAUST

Integration in diffusers

Thanks to @zjysteven for his efforts. Below shows an example with stable-diffusion-2-1-base.

import torch
from PIL import Image, ImageDraw
from copy import deepcopy

from examples.community.pipeline_stable_diffusion_boxdiff import StableDiffusionBoxDiffPipeline

def draw_box_with_text(img, boxes, names):
    colors = ["red", "olive", "blue", "green", "orange", "brown", "cyan", "purple"]
    img_new = deepcopy(img)
    draw = ImageDraw.Draw(img_new)

    W, H = img.size
    for bid, box in enumerate(boxes):
        draw.rectangle([box[0] * W, box[1] * H, box[2] * W, box[3] * H], outline=colors[bid % len(colors)], width=4)
        draw.text((box[0] * W, box[1] * H), names[bid], fill=colors[bid % len(colors)])
    return img_new

pipe = StableDiffusionBoxDiffPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-1-base",
    torch_dtype=torch.float16,
)
pipe.to("cuda")

# example 1
prompt = "as the aurora lights up the sky, a herd of reindeer leisurely wanders on the grassy meadow, admiring the breathtaking view, a serene lake quietly reflects the magnificent display, and in the distance, a snow-capped mountain stands majestically, fantasy, 8k, highly detailed"
phrases = [
    "aurora",
    "reindeer",
    "meadow",
    "lake",
    "mountain"
]
boxes = [[1,3,512,202], [75,344,421,495], [1,327,508,507], [2,217,507,341], [1,135,509,242]]

# example 2
# prompt = "A rabbit wearing sunglasses looks very proud"
# phrases = ["rabbit", "sunglasses"]
# boxes = [[67,87,366,512], [66,130,364,262]]

boxes = [[x / 512 for x in box] for box in boxes]

images = pipe(
    prompt,
    boxdiff_phrases=phrases,
    boxdiff_boxes=boxes,
    boxdiff_kwargs={
        "attention_res": 16,
        "normalize_eot": True
    },
    num_inference_steps=50,
    guidance_scale=7.5,
    generator=torch.manual_seed(42),
    safety_checker=None
).images

draw_box_with_text(images[0], boxes, phrases).save("output.png")

Setup

Note that we only test the code using PyTorch==1.12.0. You can build the environment via pip as follow:

pip3 install -r requirements.txt

To apply BoxDiff on GLIGEN pipeline, please install diffusers as follow:

git clone git@github.com:gligen/diffusers.git
pip3 install -e .

Usage

To add spatial control on the Stable Diffusion model, you can simply use run_sd_boxdiff.py. For example:

CUDA_VISIBLE_DEVICES=0 python3 run_sd_boxdiff.py --prompt "as the aurora lights up the sky, a herd of reindeer leisurely wanders on the grassy meadow, admiring the breathtaking view, a serene lake quietly reflects the magnificent display, and in the distance, a snow-capped mountain stands majestically, fantasy, 8k, highly detailed" --P 0.2 --L 1 --seeds [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,21,22,23,24,25,26,27,28,29,30] --token_indices [3,12,21,30,46] --bbox [[1,3,512,202],[75,344,421,495],[1,327,508,507],[2,217,507,341],[1,135,509,242]] --refine False

or another example:

CUDA_VISIBLE_DEVICES=0 python3 run_sd_boxdiff.py --prompt "A rabbit wearing sunglasses looks very proud"  --P 0.2 --L 1 --seeds [1,2,3,4,5,6,7,8,9] --token_indices [2,4] --bbox [[67,87,366,512],[66,130,364,262]]

Note that you can specify the token indices as the indices of words you want control in the text prompt and one token index has one corresponding conditoning box. P and L are hyper-parameters for the proposed constraints.

When --bbox is not specified, there is a interface to draw bounding boxes as conditions.

CUDA_VISIBLE_DEVICES=0 python3 run_sd_boxdiff.py --prompt "A rabbit wearing sunglasses looks very proud"  --P 0.2 --L 1 --seeds [1,2,3,4,5,6,7,8,9] --token_indices [2,4]

To add spatial control on the GLIGEN model, you can simply use run_gligen_boxdiff.py. For example:

CUDA_VISIBLE_DEVICES=0 python3 run_gligen_boxdiff.py --prompt "A rabbit wearing sunglasses looks very proud" --gligen_phrases ["a rabbit","sunglasses"] --P 0.2 --L 1 --seeds [1,2,3,4,5,6,7,8,9] --token_indices [2,4] --bbox [[67,87,366,512],[66,130,364,262]] --refine False

The direcory structure of synthetic results are as follows:

outputs/
|-- text prompt/
|   |-- 0.png 
|   |-- 0_canvas.png 
|   |-- 1.png
|   |-- 1_canvas.png 
|   |-- ...

Customize Your Layout

VisorGPT can customize layouts as spatial conditions for image synthesis using BoxDiff.

Citation

@InProceedings{Xie_2023_ICCV,
    author    = {Xie, Jinheng and Li, Yuexiang and Huang, Yawen and Liu, Haozhe and Zhang, Wentian and Zheng, Yefeng and Shou, Mike Zheng},
    title     = {BoxDiff: Text-to-Image Synthesis with Training-Free Box-Constrained Diffusion},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    year      = {2023},
    pages     = {7452-7461}
}

Acknowledgment - the code is highly based on the repository of diffusers, google, and yuval-alaluf.