Stability AI is excited to announce the launch of Stable Diffusion Reimagine! We invite users to experiment with images and ‘reimagine’ their designs through Stable Diffusion.
Stable Diffusion Reimagine is a new Clipdrop tool that allows users to generate multiple variations of a single image without limits. No need for complex prompts: Users can simply upload an image into the algorithm to create as many variations as they want.
In the examples below, the top left images are the original files fed into the tool, while the others are ‘reimagined’ creations inspired by the original.
Your bedroom can be transformed with the click of a button:
You can also play around with fashion looks, and so much more:
Clipdrop also features an upscaler, allowing a user to upload a small image and generate one with at least double the level of detail.
Usage and Limitations
Stable Diffusion Reimagine does not recreate images driven by original input. Instead, Stable Diffusion Reimagine creates new images inspired by originals.
This technology has known limitations: It can inspire amazing results based on some images and produce less impressive results for others.
We have installed a filter into the model to block inappropriate requests, but there is a chance that the filter will succumb to false negatives or false positives on occasion.
The model may also produce abnormal results or exhibit biased behavior at times. We are eager to collect user feedback to aid in our ongoing work to improve this system and mitigate against these biases.
Stable Diffusion Reimagine is based on a new algorithm created by stability.ai. The classic text-to-image Stable Diffusion model is trained to be conditioned on text inputs.
This version replaces the original text encoder with an image encoder. Instead of generating images based on text input, images are generated from an image. Some noise is added to generate variation after the encoder is put through the algorithm.
This approach produces similar looking images with different details and compositions. Unlike the image-to-image algorithm, the source image is first fully encoded. This means the generator does not use a single pixel sourced from the original image.