add live churn and MRR by counting cancellations immediately (not on cycle end) so we can directly tweak cancel page to reduce churn % #photoai

tell Claude to get its fucking SHIT together #photoai

fix language voice selector #photoai

add background audio to video2video lipsync #photoai

try Wan2.1 training txt2img but it looks a bit overtrained #photoai

get both video2video and img2video work with lipsync with 2 different models so you can get the benefits of high movement video models like Kling 2.1 with lipsync (video2video) and regular lipsync only model (img2video) #photoai

add voice to WAN2.1 LoRa video + interpolation + boomerang loop for 5->10 sec #photoai

interpolate WAN2.1 LoRa video and add 1 extra frame #photoai

try WAN2.1 LoRa trainer inference and it's great but low FPS so we need an interpolator #photoai

change steps to be same for all trainings because more steps seems to create the vertical scanlines problem on Flux turbo trainer #photoai

try WAN2.1 LoRa trainer for training on people to generate videos #photoai

in case user [ Make talking video ] on existing video just use lipsync model so that they can generate cool videos with moving cameras etc. and then still submit it as lipsync video with generated voice (video2video) (different from submitting just a photo, then our other model will generate a whole video for it along with our voice audio, img2video) #photoai

add tags design to #photoai voice selector

try Veo3 image-to-video (just came out) but it's limited to 16:9, so try rotate input image and then rotate it back and it doesn't really work because the model thinks the person is lying down and it starts putting its head up lol #photoai

make the script different based on AI model's gender #photoai

add talking video script generator button #photoai

make voice selector filters #photoai

add other languages for voices #photoai

launch new talking video feature with multiple voices #photoai x.com/levelsio/status/1950986…

make first new talking video with my own voice #photoai