README.md

---
base_model: black-forest-labs/FLUX.1-dev
datasets: TIGER-Lab/OmniEdit-Filtered-1.2M
library_name: diffusers
license: other
inference: true
tags:
- flux
- flux-diffusers
- text-to-image
- diffusers
- control
- diffusers-training
widget:
- text: Give this the look of a traditional Japanese woodblock print.
  output:
    url: >-
      https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_car.jpg
- text: transform the setting to a winter scene
  output:
    url: >-
      https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_green_creature.jpg
- text: turn the color of mushroom to gray
  output:
    url: >-
      https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_mushroom.jpg
- text: Change it to look like it's in the style of an impasto painting.
  output:
    url: >-
      https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_norte_dam.jpg
---

# Flux Edit

<Gallery />

These are the control weights trained on [black-forest-labs/FLUX.1-dev](htpss://hf.co/black-forest-labs/FLUX.1-dev) 
and [TIGER-Lab/OmniEdit-Filtered-1.2M](https://huggingface.co./datasets/TIGER-Lab/OmniEdit-Filtered-1.2M) for image editing. We use the
[Flux Control framework](https://blackforestlabs.ai/flux-1-tools/) for fine-tuning.


## License

Please adhere to the licensing terms as described [here](https://huggingface.co./black-forest-labs/FLUX.1-dev/blob/main/LICENSE.md)


## Intended uses & limitations

### Inference

```py
from diffusers import FluxControlPipeline, FluxTransformer2DModel
from diffusers.utils import load_image
import torch 

path = "sayakpaul/FLUX.1-dev-edit-v0" 
edit_transformer = FluxTransformer2DModel.from_pretrained(path, torch_dtype=torch.bfloat16)
pipeline = FluxControlPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev", transformer=edit_transformer, torch_dtype=torch.bfloat16
).to("cuda")

url = "https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/assets/mushroom.jpg"
image = load_image(url) # resize as needed.
print(image.size)

prompt = "turn the color of mushroom to gray"
image = pipeline(
    control_image=image,
    prompt=prompt,
    guidance_scale=30., # change this as needed.
    num_inference_steps=50, # change this as needed.
    max_sequence_length=512,
    height=image.height,
    width=image.width,
    generator=torch.manual_seed(0)
).images[0]
image.save("edited_image.png")
```

### Speeding inference with a turbo LoRA

We can speed up the inference by reducing the `num_inference_steps` to produce a nice image by using turbo LoRA like [`ByteDance/Hyper-SD`](https://hf.co/ByteDance/Hyper-SD).

Make sure to install `peft` before running the code below: `pip install -U peft`.

<details>
<summary>Code</summary>

```py
from diffusers import FluxControlPipeline, FluxTransformer2DModel
from diffusers.utils import load_image
from huggingface_hub import hf_hub_download
import torch

path = "sayakpaul/FLUX.1-dev-edit-v0"
edit_transformer = FluxTransformer2DModel.from_pretrained(path, torch_dtype=torch.bfloat16)
pipeline = FluxControlPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-dev", transformer=edit_transformer, torch_dtype=torch.bfloat16
).to("cuda")

# load the turbo LoRA
pipeline.load_lora_weights(
    hf_hub_download("ByteDance/Hyper-SD", "Hyper-FLUX.1-dev-8steps-lora.safetensors"), adapter_name="hyper-sd"
)
pipeline.set_adapters(["hyper-sd"], adapter_weights=[0.125])


url = "https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/assets/mushroom.jpg"
image = load_image(url) # resize as needed.
print(image.size)

prompt = "turn the color of mushroom to gray"
image = pipeline(
    control_image=image,
    prompt=prompt,
    guidance_scale=30., # change this as needed.
    num_inference_steps=8, # change this as needed.
    max_sequence_length=512,
    height=image.height,
    width=image.width,
    generator=torch.manual_seed(0)
).images[0]
image.save("edited_image.png")
```

</details>
<br>
<details>
<summary>Comparison</summary>

<table align="center">
  <tr>
    <th>50 steps</th>
    <th>8 steps</th>
  </tr>
  <tr>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_car.jpg" alt="50 steps 1" width="150"></td>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_8steps_car.jpg" alt="8 steps 1" width="150"></td>
  </tr>
  <tr>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_norte_dam.jpg" alt="50 steps 2" width="150"></td>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_8steps_norte_dam.jpg" alt="8 steps 2" width="150"></td>
  </tr>
  <tr>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_mushroom.jpg" alt="50 steps 3" width="150"></td>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_8steps_mushroom.jpg" alt="8 steps 3" width="150"></td>
  </tr>
  <tr>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_green_creature.jpg" alt="50 steps 4" width="150"></td>
    <td align="center"><img src="https://huggingface.co./datasets/sayakpaul/sample-datasets/resolve/main/flux-edit-artifacts/edited_8steps_green_creature.jpg" alt="8 steps 4" width="150"></td>
  </tr>
</table>


</details>

You can also choose to perform quantization if the memory requirements cannot be satisfied further w.r.t your hardware. Refer to the [Diffusers documentation](https://huggingface.co./docs/diffusers/main/en/quantization/overview) to learn more.

`guidance_scale` also impacts the results:

<table align="center">
  <tr>
    <th>Prompt</th>
    <th>Collage (gs: 10)</th>
    <th>Collage (gs: 20)</th>
    <th>Collage (gs: 30)</th>
    <th>Collage (gs: 40)</th>
  </tr>
  <tr>
    <td align="center">
      <em>Give this the look of a traditional Japanese woodblock print.</em>
    </td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_0.png" alt="Edited Image gs 10"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_1.png" alt="Edited Image gs 20"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_2.png" alt="Edited Image gs 30"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_3.png" alt="Edited Image gs 40"></td>
  </tr>
  <tr>
    <td align="center">
      <em>transform the setting to a winter scene</em>
    </td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_4.png" alt="Edited Image gs 10"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_5.png" alt="Edited Image gs 20"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_5.png" alt="Edited Image gs 30"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_6.png" alt="Edited Image gs 40"></td>
  </tr>
  <tr>
    <td align="center">
      <em>turn the color of mushroom to gray</em>
    </td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_12.png" alt="Edited Image gs 10"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_13.png" alt="Edited Image gs 20"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_14.png" alt="Edited Image gs 30"></td>
    <td align="center"><img src="https://huggingface.co./sayakpaul/FLUX.1-dev-edit-v0/resolve/main/images_15.png" alt="Edited Image gs 40"></td>
  </tr>
</table>


### Limitations and bias

Expect the model to perform underwhelmingly as we don't know the exact training details of Flux Control.

## Training details

Fine-tuning codebase is [here](https://github.com/sayakpaul/flux-image-editing). Training hyperparameters:

* Per GPU batch size: 4
* Gradient accumulation steps: 4
* Guidance scale: 30
* BF16 mixed-precision
* AdamW optimizer (8bit from `bitsandbytes`)
* Constant learning rate of 5e-5
* Weight decay of 1e-6
* 20000 training steps


Training was conducted using a node of 8xH100s.

We used a simplified flow mechanism to perform the linear interpolation. In pseudo-code, that looks like:

```py
sigmas = torch.rand(batch_size)
timesteps = (sigmas * noise_scheduler.config.num_train_timesteps).long()
...

noisy_model_input = (1.0 - sigmas) * pixel_latents + sigmas * noise
```

where `pixel_latents` is computed from the source images and `noise` is drawn from a Gaussian distribution. For more details, [check out
the repository](https://github.com/sayakpaul/flux-image-editing/blob/b041f62df8f959dc3b2f324d2bfdcdf3a6388598/train.py#L403).