A model that try to replicate, let's say walking in the street (and seen others walk in the street) from First person PoV, I am still struggling learning the good use of a workflow so my example videos are not optimal right now, I just uploaded the training data prompt .tct files so you can see what the model was trained on (note these are not Good Prompt, they are descriptive for the training prompt, you can definately improve them to genere better output, but if you want a quick test you can still use them as is)Example prompt : str3etLIF3 video scene featuring a group of young people walking on the sidewalk in front of a night club, outdoor at night in a (well lit:0.7) city of Los Angeles, two women and two men are the main focus, they are walking toward the camera, one of the woman is a beautiful blonde wearing a fashion beige minidress form fitting outfit, she walk, and she is smoking vaping e-cigare, one man is holding his cellphone as he talk to someone else, one of the men is wearing a black pants, and the other main man is wearing a light blue jeans, (sports cars can be seen passing on the road left side:1.15) of the frame, dynamic, 4k ultra quality, color graded mastery. (detailed faces, detailed body shapes and body movements:1.25)Or just start your prompt by str3etLIF3 and be creative, knowing that the model was trained on first person street walking, street 'chilling' footages, mostly of people walking on the sidewalk, and in particular pretty girls in the night life of big cities (no NSFW footage used, just some provocative outfit for some but that's all) there is also men in the training video, just less than women, so model probably will generate more females unless prompted accordinglythis is epoch 40, the last, but I have a sneaky feeling epoch 30 is actually a touch better, i'll test and see, but this one for now.try also if you want, combining it with my previouus lora Move_Enhancer at 0.6 or so (for Move_Enhancer) it may give decent result, or funny resultI will try to understnad this workflo (or change workflow if I don't figure it out) and generate more accaptable demo videos and their prompts during the day/night, one user already made stunning videos with the lora so it's possible :pKnown flaws : Tendency to give handbags to everyone (men included) this is due to many of the women in the training data carrying handbags, it kinda generalized people with handbags, you can probably counter that by prompting something like : No handbags or No one is carrying handbag, or Men don't have handbag (at the end of the prompt for example, just as a hint for the model to be guided) A slight tendency to give people short arms, it's subtle, it's not completly our of place, but they tend to have shorter arms than usual (visible mostly when their arms are along their body relaxed) you can try prompting : natural long arms, or elegant long arms, or accuratebody proportion in the prompt (again near the end, just as a final touch)

Street life, walking/chilling in the street

this is epoch 40 (the last) After proper testing I may go back to epoch 30 or eventually 20 (epoch 10 was already tested in alpha during training, thanks to the users who helped with that, and epoch 10 is probably not the best of the lot)

Files with SHA256: 655d354d75c78dbd0c9ec5ff1ae1682b06fb16a15ab867ca292c2707bce66ec3