IC-Light
IC-Light is a project to manipulate the illumination of images.
The name “IC-Light” stands for “Imposing Consistent Light” (we will briefly describe this at the end of this page).
Currently, we release two types of models: text-conditioned relighting model and background-conditioned model. Both types take foreground images as inputs.
Note that “iclightai dot com” is a scam website. They have no relationship with us. Do not give scam websites money! This GitHub repo is the only official IC-Light.
News
Alternative model for stronger illumination modifications.
Some news about flux is here. (A fix update is added at Nov 25, more demos will be uploaded soon.)
Get Started
Below script will run the text-conditioned relighting model:
git clone https://github.com/lllyasviel/IC-Light.git
cd IC-Light
conda create -n iclight python=3.10
conda activate iclight
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt
python gradio_demo.py
Or, to use background-conditioned demo:
python gradio_demo_bg.py
Model downloading is automatic.
Note that the “gradio_demo.py” has an official huggingFace Space here.
Screenshot
Text-Conditioned Model
(Note that the “Lighting Preference” are just initial latents - eg., if the Lighting Preference is “Left” then initial latent is left white right black.)
Prompt: beautiful woman, detailed face, warm atmosphere, at home, bedroom
Lighting Preference: Left
Prompt: beautiful woman, detailed face, sunshine from window
Lighting Preference: Left
beautiful woman, detailed face, neon, Wong Kar-wai, warm
Lighting Preference: Left
Prompt: beautiful woman, detailed face, sunshine, outdoor, warm atmosphere
Lighting Preference: Right
Prompt: beautiful woman, detailed face, sunshine, outdoor, warm atmosphere
Lighting Preference: Left
Prompt: beautiful woman, detailed face, sunshine from window
Lighting Preference: Right
Prompt: beautiful woman, detailed face, shadow from window
Lighting Preference: Left
Prompt: beautiful woman, detailed face, sunset over sea
Lighting Preference: Right
Prompt: handsome boy, detailed face, neon light, city
Lighting Preference: Left
Prompt: beautiful woman, detailed face, light and shadow
Lighting Preference: Left
(beautiful woman, detailed face, soft studio lighting)
Prompt: Buddha, detailed face, sci-fi RGB glowing, cyberpunk
Lighting Preference: Left
Prompt: Buddha, detailed face, natural lighting
Lighting Preference: Left
Prompt: toy, detailed face, shadow from window
Lighting Preference: Bottom
Prompt: toy, detailed face, sunset over sea
Lighting Preference: Right
Prompt: dog, magic lit, sci-fi RGB glowing, studio lighting
Lighting Preference: Bottom
Prompt: mysteriou human, warm atmosphere, warm atmosphere, at home, bedroom
Lighting Preference: Right
Background-Conditioned Model
The background conditioned model does not require careful prompting. One can just use simple prompts like “handsome man, cinematic lighting”.
A more structured visualization:
Imposing Consistent Light
In HDR space, illumination has a property that all light transports are independent.
As a result, the blending of appearances of different light sources is equivalent to the appearance with mixed light sources:
Using the above light stage as an example, the two images from the “appearance mixture” and “light source mixture” are consistent (mathematically equivalent in HDR space, ideally).
We imposed such consistency (using MLPs in latent space) when training the relighting models.
As a result, the model is able to produce highly consistent relight - so consistent that different relightings can even be merged as normal maps! Despite the fact that the models are latent diffusion.
From left to right are inputs, model outputs relighting, devided shadow image, and merged normal maps. Note that the model is not trained with any normal map data. This normal estimation comes from the consistency of relighting.
You can reproduce this experiment using this button (it is 4x slower because it relight image 4 times)
Below are bigger images (feel free to try yourself to get more results!)
For reference, geowizard (geowizard is a really great work!):
And, switchlight (switchlight is another great work!):
Model Notes
-
iclight_sd15_fc.safetensors - The default relighting model, conditioned on text and foreground. You can use initial latent to influence the relighting.
-
iclight_sd15_fcon.safetensors - Same as “iclight_sd15_fc.safetensors” but trained with offset noise. Note that the default “iclight_sd15_fc.safetensors” outperform this model slightly in a user study. And this is the reason why the default model is the model without offset noise.
-
iclight_sd15_fbc.safetensors - Relighting model conditioned with text, foreground, and background.
Also, note that the original BRIA RMBG 1.4 is for non-commercial use. If you use IC-Light in commercial projects, replace it with other background replacer like BiRefNet.
Cite
@inproceedings{
zhang2025scaling,
title={Scaling In-the-Wild Training for Diffusion-based Illumination Harmonization and Editing by Imposing Consistent Light Transport},
author={Lvmin Zhang and Anyi Rao and Maneesh Agrawala},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=u1cQYxRI1H}
}
Also read …
Total Relighting: Learning to Relight Portraits for Background Replacement
Relightful Harmonization: Lighting-aware Portrait Background Replacement
SwitchLight: Co-design of Physics-driven Architecture and Pre-training Framework for Human Portrait Relighting
评论