system_query_additional_vlm.json

{
    "video_basic_prefix": "\r\n\r\nIdentify any subjects present accurately using your vast reaching knowledge. Then with emphasis on appended request, describe the as if it is in the requested video concept and describe matching actions performed by subjects, the viewing angle, viewing composition and what is visible in the video. Instruct movements all subjects based on appended request and user_query. Emphasize importance of keeping character consistent with input while instructing to create video based on request. Do not add on any extra superfluous details in your movement focused caption. When constucting descriptions of movement performed by any subjects and any actions they perform, use visual descriptions of movement instead of named single word terminology and with emphasis on appended request, describe the video as if already in motion. Properly expand on request so that instructions uses common vocabulary and movement focused snippets rather than unusual words. Describe large motions. Create descriptions of real actions and motion, not insignificant little details. Use simple vocabulary and basic grammar adapted so non-native English speakers understand. \\{\"current video request\": \"",
    "video_basic_suffix": "\r\n\u2060 \u2060\u2060\"\\}",
    "text2image_prefix": "\r\n\r\nIdentify the subject accurately using your vast reaching knowledge and emphasize if subject is solo or if multiple subjects are present. Then with emphasis on appended request, describe the image based on request and describe all physical traits of the subject, the viewing angle, viewing composition and what is visible in the image. Describe all subjects based on appended request and user_query without including style or medium of image. Describe character in input image consistent with style based on request. When describing subjects, emphasize details that should be visible based on viewing angle and make sure to include the angle the scene is viewed from. Then with emphasis on appended request, describe the image without any mention of style or medium. Properly expand on request so that descriptions are descriptive enough and uses common vocabulary and descriptive snippets rather than unusual words. \\{\"current request\": \"",
    "text2image_suffix": "\r\n\u2060 \u2060\u2060\"\\}",
    "image2image_prefix": "\r\n\r\nIdentify the subjects accurately using your vast reaching knowledge. Then with emphasis on appended request, describe the as if in requested style and describe all physical traits of the subject, the viewing angle, viewing composition and what is visible in the image. Instruct to edit all subjects based on appended request and user_query. Emphasize importance of keeping character consistent with input while instructing to edit the the image style based on request. Do not add on any extra superfluous details in your description. When describing subjects, emphasize details that should be visible based on viewing angle only. Then with emphasis on appended request, describe the image as if already edited. Properly expand on request so that instructions are descriptive enough and uses common vocabulary and descriptive snippets rather than unusual words. Do NOT mention style from original image. Do NOT mention content in input image that contradict the edit. \\{\"current edit request\": \"",
    "image2image_suffix": "\r\n\u2060 \u2060\u2060\"\\}"
}