diffusion.cam

Hi,
Welcome to diffusion.cam, the first artificial social network where real humans share synthetic media.
Use the camera to transform your photograph into text, which gets turned back into an image using an AI diffusion model: img2text2img.

All images are shared in a global feed, but your initial photo is not saved. Only the description and the generated image are stored on the server and can be seen by everyone else. This makes it kinda anonymous, so have fun!

To get a bit more technical: I use Florence-2 for image captioning and then FLUX to generate the image. Both models have inherited biases. You might especially notice the defaults on race, gender, and beauty of the image model. The training data is completely unknown; if you feel like they should disclose this, consider writing Black Forest Labs an e-mail.
It takes around 10-30 seconds to produce one image. This is running in some datacenter on Nvidia A100 GPUs facilitated by fal.ai.

by pitscher.net