Tool for location and interior design consistency
Hi. I'm very interested by your work. I am looking for a solution to location consistency in a personal indy short film project. The problem is that in cinematography we need to make consistent shots/reverse shots for dialogs. In that case, unless using unfocused backgrounds, most of the video models will just hallucinate different details in EVERY shot that points out of the background reference images or to the 'back' of the IA 'scene'. Mick Mumpitz is trying hard to solve that problem in many of his tools, including the ultimate one that seems to be very powerful, generating a whole 3D splat cohesive scene. Unfortunately it's heavy, and there is no link that allows you to link the IA 'camera' with the correct 3D background. In that case we would need some kind of dynamic I2V process with the first frame changing on each frame. This is out of my knowledge. Back to your work. For an indoor scene I have generated with Qwen Image multiple views of a room. I tried to train a Lora for it or to 'reconstruct' a 3D scene with multiple reference sheets in Minimax. But there are different details on each shot, so the results are bad. With a camera rotation sequence there are objects or details that are popping or desappearing. It seems that you could generate better results, as Minimax seems to generate more cohesive results than Qwen. Is that right? Did you try to use your sheets as reference images? Did you check the work of LumousAI that seems to be close to your solution ? Do you think your solution is better ? https://www.patreon.com/u75172830/posts/create-full-in-166728952