First, please take the time to watch the video, you wont regret it. This is a very large project and these cameras have many capabilities and specific settings which I go over in the video.
I am proud to introduce to you an innovative workflow solution for all of your prompting needs! Latent Layer Cameras are a set of 30 Hypernetwork. These are virtual cameras that have been embedded in to the latent space which allow you to choose the camera angle and camera shot for your subject. These cameras put you in to the directors seat so that you always get the shot you envisioned in your mind. They are also compatible with any 1.5 model, give them a try with your favorite model! Besides being very flexible, the latent cameras also tend to be very cohesive and produce very good results through natural language prompting. That means there is no need for a word salad in your positive or negative prompts. In fact you can use the Hypernetworks without any negative prompts at all and still get very good results. As long as you follow very basic rules outlined in the video guide, on average you will get very good results. Thank you!
People have been asking me to release only the cameras without the Prometheus model, so here we are. If you still want the Prometheus chekpoint you can find it here. https://civarchive.com/models/113034/prometheus
Camera shot and angle cheat sheet. https://postimg.cc/gallery/V645cMS
Huggingface link. https://huggingface.co/doctorderp/prometheus.safetensors/resolve/main/prometheus.zip
Please note the Hypernetworks work with all models...that are based on the 1.5 architecture... sorry if I misled in the video. Sorry no 2.0 or 2.1 support.
Don't forget to enable your dynamic prompts by checking the enable box, otherwise you can load the Hypernetworks manually, just know that the strength of the Hypernetworks should be at 0.55 by default, otherwise you will not get good results.
[Tip Jar](https://www.paypal.com/donate/?business=9B4V7D6BBUNPL&no_recurring=0¤cy_code=USD)
Discord if you have any questions = dr.derp#7202 or just dr.derp
Description
FAQ
Comments (19)
lmao you really did it
SDXL ? Please !
Working out the kinks on the SDXL version, LORA doesn't want to behave as well as the hypernetworks. Stay tuned.
this is brilliant.
Absolutely 5 stars! Great HyperNetworks! Now I can control the camera whatever I want. Thanks for making this.
I've combined all you wildcards into one yaml file for easier useage. I uploaded in here and hope you don't mind馃構. Thanks again for this powerful hypernetwork.
Looks great!
Can you do a 45-degree high angle camera? it will be very helpful for building interior like CCTV camera but without lens distortion
Don't have those angles trained, but they do interpolate sometimes. Try adding the prefix "high angle overhead view" before the camera name and it works sometimes. So for example "high angle overhead view. __b2_ _ a woman standing inside the kitchen wearing an apron.
this is a gamechanger
SDXL ? Please !
Awesome job for awesome idea!
Whats a usage of _np_ (2/3/i/i2) and xa/xb/xc/, p, allcams etc txt in wild cards?
Will you plan to do a short textual discription?
Whenever I use this, my results are totally broken. :(
Why no sausage? haha
Can you make these safetensors?
hypernetworks can not be made in to safetensors
@drderp聽Hey thanks for the info!
I'm getting poor results, all the women are young, the prompt is not working. Only the camera setting is working correctly...
I recommend you watch the video for the full use case description. Most of the things are covered there. Laten layer cameras are hypernetworks that work in conjunction with another model of your own choosing. If the results are not good its because the model you are using is no good. Test the latent cameras on my recommended Prometheus model. If you are still getting results that don't match what I show in my video, you set up the cameras wrong somewhere.