IF by DeepFloyd Lab at StabilityAI

We introduce DeepFloyd IF, a novel state-of-the-art open-source text-to-image model with a high degree of photorealism and language understanding. DeepFloyd IF is a modular composed of a frozen text encoder and three cascaded pixel diffusion modules: a base model that generates 64x64 px image based on text prompt and two super-resolution models, each designed to generate images of increasing resolution: 256x256 px and 1024x1024 px. All stages of the model utilize a frozen text encoder based on the T5 transformer to extract text embeddings, which are then fed into a UNet architecture enhanced with cross-attention and attention pooling. The result is a highly efficient model that outperforms current state-of-the-art models, achieving a zero-shot FID score of 6.66 on the COCO dataset. Our work underscores the potential of larger UNet architectures in the first stage of cascaded diffusion models and depicts a promising future for text-to-image synthesis.

Inspired by Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding

Minimum requirements to use all IF models:

16GB vRAM for IF-I-XL (4.3B text to 64x64 base module) & IF-II-L (1.2B to 256x256 upscaler module)
24GB vRAM for IF-I-XL (4.3B text to 64x64 base module) & IF-II-L (1.2B to 256x256 upscaler module) & Stable x4 (to 1024x1024 upscaler)
xformers and set env variable FORCE_MEM_EFFICIENT_ATTN=1

Quick Start

pip install deepfloyd_if==1.0.2rc0
pip install xformers==0.0.16
pip install git+https://github.com/openai/CLIP.git --no-deps

Local notebooks

The Dream, Style Transfer, Super Resolution or Inpainting modes are avaliable in a Jupyter Notebook here.

Integration with 🤗 Diffusers

IF is also integrated with the 🤗 Hugging Face Diffusers library.

Diffusers runs each stage individually allowing the user to customize the image generation process as well as allowing to inspect intermediate results easily.

Example

Before you can use IF, you need to accept its usage conditions. To do so:

Make sure to have a Hugging Face account and be loggin in
Accept the license on the model card of DeepFloyd/IF-I-XL-v1.0
Make sure to login locally. Install huggingface_hub

pip install huggingface_hub --upgrade

run the login function in a Python shell

from huggingface_hub import login

login()

and enter your Hugging Face Hub access token.

Next we install diffusers and dependencies:

pip install diffusers accelerate transformers safetensors

And we can now run the model locally.

By default diffusers makes use of model cpu offloading to run the whole IF pipeline with as little as 14 GB of VRAM.

If you are using torch>=2.0.0, make sure to delete all enable_xformers_memory_efficient_attention() functions.

from diffusers import DiffusionPipeline
from diffusers.utils import pt_to_pil
import torch

# stage 1
stage_1 = DiffusionPipeline.from_pretrained("DeepFloyd/IF-I-XL-v1.0", variant="fp16", torch_dtype=torch.float16)
stage_1.enable_xformers_memory_efficient_attention()  # remove line if torch.__version__ >= 2.0.0
stage_1.enable_model_cpu_offload()

# stage 2
stage_2 = DiffusionPipeline.from_pretrained(
    "DeepFloyd/IF-II-L-v1.0", text_encoder=None, variant="fp16", torch_dtype=torch.float16
)
stage_2.enable_xformers_memory_efficient_attention()  # remove line if torch.__version__ >= 2.0.0
stage_2.enable_model_cpu_offload()

# stage 3
safety_modules = {"feature_extractor": stage_1.feature_extractor, "safety_checker": stage_1.safety_checker, "watermarker": stage_1.watermarker}
stage_3 = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-x4-upscaler", **safety_modules, torch_dtype=torch.float16)
stage_3.enable_xformers_memory_efficient_attention()  # remove line if torch.__version__ >= 2.0.0
stage_3.enable_model_cpu_offload()

prompt = 'a photo of a kangaroo wearing an orange hoodie and blue sunglasses standing in front of the eiffel tower holding a sign that says "very deep learning"'

# text embeds
prompt_embeds, negative_embeds = stage_1.encode_prompt(prompt)

generator = torch.manual_seed(0)

# stage 1
image = stage_1(prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_embeds, generator=generator, output_type="pt").images
pt_to_pil(image)[0].save("./if_stage_I.png")

# stage 2
image = stage_2(
    image=image, prompt_embeds=prompt_embeds, negative_prompt_embeds=negative_embeds, generator=generator, output_type="pt"
).images
pt_to_pil(image)[0].save("./if_stage_II.png")

# stage 3
image = stage_3(prompt=prompt, image=image, generator=generator, noise_level=100).images
image[0].save("./if_stage_III.png")

There are multiple ways to speed up the inference time and lower the memory consumption even more with diffusers. To do so, please have a look at the Diffusers docs:

🚀 Optimizing for inference time
⚙️ Optimizing for low memory during inference

For more in-detail information about how to use IF, please have a look at the IF blog post and the documentation 📖.

Diffusers dreambooth scripts also supports fine-tuning 🎨 IF. With parameter efficient finetuning, you can add new concepts to IF with a single GPU and ~28 GB VRAM.

Run the code locally

Loading the models into VRAM

from deepfloyd_if.modules import IFStageI, IFStageII, StableStageIII
from deepfloyd_if.modules.t5 import T5Embedder

device = 'cuda:0'
if_I = IFStageI('IF-I-XL-v1.0', device=device)
if_II = IFStageII('IF-II-L-v1.0', device=device)
if_III = StableStageIII('stable-diffusion-x4-upscaler', device=device)
t5 = T5Embedder(device="cpu")

I. Dream

Dream is the text-to-image mode of the IF model

from deepfloyd_if.pipelines import dream

prompt = 'ultra close-up color photo portrait of rainbow owl with deer horns in the woods'
count = 4

result = dream(
    t5=t5, if_I=if_I, if_II=if_II, if_III=if_III,
    prompt=[prompt]*count,
    seed=42,
    if_I_kwargs={
        "guidance_scale": 7.0,
        "sample_timestep_respacing": "smart100",
    },
    if_II_kwargs={
        "guidance_scale": 4.0,
        "sample_timestep_respacing": "smart50",
    },
    if_III_kwargs={
        "guidance_scale": 9.0,
        "noise_level": 20,
        "sample_timestep_respacing": "75",
    },
)

if_III.show(result['III'], size=14)

II. Zero-shot Image-to-Image Translation

In Style Transfer mode, the output of your prompt comes out at the style of the support_pil_img

from deepfloyd_if.pipelines import style_transfer

result = style_transfer(
    t5=t5, if_I=if_I, if_II=if_II,
    support_pil_img=raw_pil_image,
    style_prompt=[
        'in style of professional origami',
        'in style of oil art, Tate modern',
        'in style of plastic building bricks',
        'in style of classic anime from 1990',
    ],
    seed=42,
    if_I_kwargs={
        "guidance_scale": 10.0,
        "sample_timestep_respacing": "10,10,10,10,10,10,10,10,0,0",
        'support_noise_less_qsample_steps': 5,
    },
    if_II_kwargs={
        "guidance_scale": 4.0,
        "sample_timestep_respacing": 'smart50',
        "support_noise_less_qsample_steps": 5,
    },
)
if_I.show(result['II'], 1, 20)

Alternative Text

III. Super Resolution

For super-resolution, users can run IF-II and IF-III or 'Stable x4' on an image that was not necessarely generated by IF (two cascades):

from deepfloyd_if.pipelines import super_resolution

middle_res = super_resolution(
    t5,
    if_III=if_II,
    prompt=['woman w

IF

Install / Use

README