Conversation

Emad
@EMostaque
We will soon be announcing our cutting-edge Deep Fake detection initiative using the new OpenCLIP models and more. As the below great work shows it’s not only image creation but image description and more the models we release enable.
Quote Tweet
Sharif Shameem
@sharifshameem
Just added reverse image search to Lexica. You can upload a photo and it’ll return the most similar Stable Diffusion images and their prompts. This makes it very to turn a real world concept into a Stable Diffusion prompt. lexica.art
Embedded video
0:31
79.9K views
8
477
Emad
@EMostaque
This technology is inevitable and bad actors already have it in there hands which is one reason we looked for this to go out wide with our partners and build a developer community that could improve upon the classifiers and other elements we bundled. Everyone should be able to
1
53
Emad
@EMostaque
Control what they want to see and also detect what is generated and not as this inevitably goes to high resolution amazing composition within a few short years. Hiding away datasets and models does not enable us to mount a proper defence against this and educate society
2
60
Emad
@EMostaque
As the generative and classifier models may understand different things as discussed here
Quote Tweet
Emad
@EMostaque
The reason Greg Rutkowski and many other artists “work” in #StableDiffusion is not actually due to the LAION dataset but the @OpenAI CLIP model (L14) used to teach it language We don’t know dataset it’s based on (closed) so even if an artist removed from LAION still “knows” style twitter.com/KyrickYoung/st…
1
26
Emad
@EMostaque
Replying to
@EMostaque
In the coming months not only will we be releasing more diverse and optimised datasets and models but we will be actively funding development in technologies such as this as well as creator control and other elements alongside education in understanding these models.
2
70
Emad
@EMostaque
These are powerful tools that everyone will use in the coming years and are essential infrastructure for the future. As such the benchmark models should be open infrastructure and we should all have the tools we need individually and as a society to make best use of these.
1
66