{ "shot_index": 1, "raw_prompt": " is a young white female audio student with auburn hair in a low ponytail, a navy rain shell, and small over-ear field headphones. 's voice is a soft, grounded conversational voice with an unhurried cadence. 's mouth remains clearly visible and naturally synchronized with the spoken line. At normal speed, follows several paces behind and removes one headphone cup to compare the open air with her recorder's quiet monitor level. says in English, using a soft, grounded conversational voice with an unhurried cadence: \"I will keep the monitor low and listen directly between stops.\" The shot uses realistic cinematic imagery, physically plausible restrained motion, and a natural documentary-like tone. The audible soundscape is: Light footsteps, reed movement, and a faint equipment rustle.", "selector_query": " is a young white female audio student with auburn hair in a low ponytail, a navy rain shell, and small over-ear field headphones. 's voice is a soft, grounded conversational voice with an unhurried cadence. 's mouth remains clearly visible and naturally synchronized with the spoken line. At normal speed, follows several paces behind and removes one headphone cup to compare the open air with her recorder's quiet monitor level. says in English, using a soft, grounded conversational voice with an unhurried cadence: \"I will keep the monitor low and listen directly between stops.\" The shot uses realistic cinematic imagery, physically plausible restrained motion, and a natural documentary-like tone. The audible soundscape is: Light footsteps, reed movement, and a faint equipment rustle. Identity continuity requirements: Keep 's facial identity, age, auburn low ponytail, facial proportions, skin tone, navy rain shell, and small over-ear field headphones exactly consistent with her identity reference; do not redesign or substitute her.", "generation_prompt": " is a young white female audio student with auburn hair in a low ponytail, a navy rain shell, and small over-ear field headphones. 's voice is a soft, grounded conversational voice with an unhurried cadence. 's mouth remains clearly visible and naturally synchronized with the spoken line. At normal speed, follows several paces behind and removes one headphone cup to compare the open air with her recorder's quiet monitor level. says in English, using a soft, grounded conversational voice with an unhurried cadence: \"I will keep the monitor low and listen directly between stops.\" The shot uses realistic cinematic imagery, physically plausible restrained motion, and a natural documentary-like tone. The audible soundscape is: Light footsteps, reed movement, and a faint equipment rustle. Identity continuity requirements: Keep 's facial identity, age, auburn low ponytail, facial proportions, skin tone, navy rain shell, and small over-ear field headphones exactly consistent with her identity reference; do not redesign or substitute her.", "pictures": [ { "picture": 1, "path": "picture_01.png", "source_shot": null, "frame_index": null, "identity_anchor": "", "source": "/datadrive1/v-zihanyang/evals/unityshotsbench/ltx23_joyai_echo_subject_labels_20261111/refine_assets/worker_04_tier3/subject_3_anchor_shot01_frame0060.png" }, { "picture": 2, "path": "picture_02.png", "source_shot": 0, "frame_index": 78, "identity_anchor": null, "source": "generated" }, { "picture": 3, "path": "picture_03.png", "source_shot": 0, "frame_index": 120, "identity_anchor": null, "source": "generated" } ], "audio_references": [], "generation": { "model": "/datadrive1/v-zihanyang/Audio-Video-Benchmarks/LTX-2/models/ltx-2.3-22b-dev.safetensors", "lora": "/datadrive2/v-zihanyang/restored_training_local/ltx23/rl_checkpoint61_peft/lora_bf16.safetensors", "width": 1280, "height": 736, "num_frames": 121, "fps": 24, "num_inference_steps": 30, "seed": 43, "video_cfg_scale": 4.0, "audio_cfg_scale": 7.0, "video_stg_scale": 1.0, "audio_stg_scale": 1.0, "stg_blocks": [ 29 ], "guidance_rescale": 0.7, "video_modality_guidance_scale": 1.0, "audio_modality_guidance_scale": 1.0, "token_order": "prepend" } }