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.
(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
image
Prompt: beautiful woman, detailed face, sunshine from window
Lighting Preference: Left
image
beautiful woman, detailed face, neon, Wong Kar-wai, warm
Prompt: dog, magic lit, sci-fi RGB glowing, studio lighting
Lighting Preference: Bottom
image
Prompt: mysteriou human, warm atmosphere, warm atmosphere, at home, bedroom
Lighting Preference: Right
image
Background-Conditioned Model
The background conditioned model does not require careful prompting. One can just use simple prompts like "handsome man, cinematic lighting".
image
image
image
image
A more structured visualization:
r1
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:
cons
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.
r2
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)
image
image
Below are bigger images (feel free to try yourself to get more results!)
image
image
For reference, geowizard (geowizard is a really great work!):
image
And, switchlight (switchlight is another great work!):
image
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
@Misc{iclight,
author = {Lvmin Zhang and Anyi Rao and Maneesh Agrawala},
title = {IC-Light GitHub Page},
year = {2024},
}
