I would love for the magic camera to respect the image composition that was shot and not just reimagine it. I think it would give more ownership of the images that are generated and could help you really re-imagine your world.
This can be done with things like Controlnet. It would make me look at the R1 as more like a real camera instead of a backwards prompt creator.
I completely agree with you! Having the magic camera respect the original image composition would be a game-changer. It would allow us to maintain the integrity of our shots while still benefiting from the creative enhancements that AI can provide. This approach, especially with tools like ControlNet, would definitely give us more ownership over the final images and make the experience feel more like using a real camera rather than just relying on prompts. It’s about blending our vision with AI’s capabilities, not letting the AI take over entirely. This could truly elevate the way we re-imagine our world through photography.
PS: Would be also a massive future for the Magic Cam to be able to create Constant Character. So for example after you take 10 selfies, the Magic Cam will create a version of yourself at least for the face, and then be able to detect when you are in the pictures and then use that character to add it to the original image composition.
Inpainting and outpainting would be interesting aswell
I support this. There were many times I wanted the Magic Camera to respect the original image. Is there a suggestion for this already in this forum?
I’ve read here that someone saved a very comprehensive prompt as a note and then used that note for the Magic Camera to provide pics with a specific set of characteristics. Can something like that be used for this?
just make it look a little how it was, in the end you want like a person human etc to recall , remember that moment. thats all about photography
What is Control Net? Are you referring to Comfy’s plugin?
Controlnet are multiple models trained to steer an image model to recreate the composition of an image. You must have heard of image to image but this is way more robust. You can use a controlnet in ComfyUI or another interface.