This might be the best synthetic performance I have got from a video mod

This might be the best synthetic performance I have got from a video model. We can generate 30-second shots and create near seamless extensions (with a bit of post blending) for far longer sequences, powered by @seedance 2.5. I set out to test this out by making the scene below all-in-one shot. I wrote the script, built the character references in Seedream and then broke the script down to two 30 second prompts. The prompts were really detailed with pauses, gestures and emotional shifts and the model delivered on all of them. I was really impressed with the performance and voices which were generated in the model. Previously, AI film scenes had to be composed of short cuts, now we have options on pacing, and nuance... we can let our scenes breathe. All this done with only around 3 rolls per take, we have left the slot machine age of AI video Gen behind and are now in the age of control. Next for me will be testing real actors performance as a reference to drive the generation. Seedance 2.5 API is about to launch on @BytePlusGlobal , and @lumina_ai_aiart will also roll out Seedance 2.5 soon.