Multiple Camera Angle Edit
Can you also train the model for such feature? fal/Qwen-Image-Edit-2511-Multiple-Angles-LoRA
Hey, yes I am a big fan of that Lora also. The big problem as always is training data. I personally am working on creating a dataset where you can specify the exact pitch and yaw or dolly movements. But can't yet promise that it'll be implemented in the next update already :)
That is great you are trying something like pitch and yaw. Why not use that Qwen lora to generate dataset and train it as From Angle To Angle called "x" ? So Ai understands it as "Ohh I have to change from this angle to that angle and that is called this POV". So if we just type a pov it will understand what we mean. I don't know how such concepts are trained but it would be great.
@Conradlocke The training set will be best made with real photos and 3d renders from blender/unreal. Might be interesting to get the training set from the qwen lora too
would be really interesting to use synth dataset and put exact angles into captions.. would be curious if that could control it that well.. like 85dg, 5 meters from focus.. you could easily generate such data pairs, but question is render quality :))and also quantity you need? But as I understand you are able to learn lora on rather small datasets which is nuts for me