{
    "model": {
        "name": "MiniMax H3 Ref2VA PTD RefAsymOPD 8-Step 33B",
        "architecture": "minimax_h3_ref2va",
        "description": "Accelerated H3 Ref2VA for generating video with synchronized sound from a prompt and reference images, using 8 steps instead of the standard 40. RefAsymOPD targets sharper details, more believable backgrounds and greater realism than plain PTD at the same step count. Choose it for faster iteration with improved quality at low step counts; standard H3 at 40 steps remains the quality reference.",
        "URLs": "minimax_h3_ref2va",
        "loras": [
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop01_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop02_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop03_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop04_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop05_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop06_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop07_bf16.safetensors|ptd_refasymopd",
            "https://huggingface.co/DeepBeepMeep/MiniMax-H3/resolve/main/loras/ptd_refasymopd/minimax_h3_ref2va_ptd_refasymopd_8step_hop08_bf16.safetensors|ptd_refasymopd"
        ],
        "loras_multipliers": [
            "1,0,0,0,0,0,0,0",
            "0,1,0,0,0,0,0,0",
            "0,0,1,0,0,0,0,0",
            "0,0,0,1,0,0,0,0",
            "0,0,0,0,1,0,0,0",
            "0,0,0,0,0,1,0,0",
            "0,0,0,0,0,0,1,0",
            "0,0,0,0,0,0,0,1"
        ],
        "pdd": true,
        "ptd": true
    },
    "num_inference_steps": 8,
    "guidance_phases": 1,
    "guidance_scale": 1,
    "flow_shift": 12,
    "sample_solver": "euler",
    "video_length": 124,
    "video_prompt_type": "I",
    "audio_prompt_type": "",
    "skip_steps_cache_type": "",
    "prompt_enhancer": "",
    "prompt": "subject_definitions:\n<Subject 1> is the person in <Picture 1>, preserving their exact identity, facial features, skin tone, hairstyle, body proportions, clothing, footwear, and distinctive accessories.\nsummary:\n[reference generation] Place <Subject 1> on a deserted midnight railway platform where a paper bird awakens the station clock during one five-second shot.\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - their identity, face, hair, proportions, clothing, footwear, and visible accessories remain recognizable and unchanged; only the location, lighting, action, and paper-bird prop are new.\ndetailed_description:\nThe target video is a five-second cinematic magical-realist single take on an abandoned railway platform before dawn. [Shot 1] <Subject 1> stands beneath a monumental stopped clock, framed in a steady medium close-up that keeps their recognizable face and exact appearance from <Picture 1> clearly visible. Cold blue light outlines the iron platform while a warm lamp illuminates their natural skin texture, hair, clothing, and accessories without changing them. A tiny folded paper bird rests on their raised palm. As its paper wings open, the camera gently pushes closer and <Subject 1> watches it with quiet resolve, saying clearly (S1) <d>[English] Some journeys begin when the map runs out.</d> During the final words, the bird lifts a few inches, lands on the clock's minute hand, and the mechanism advances with one solid click. <Subject 1> follows it with their eyes and gives a small hopeful smile. The spoken line and simple transformation complete naturally within the five-second shot; identity, lip motion, gaze, hand position, posture, and paper movement remain stable and physically coherent.\noverall_soundscape: Spacious nighttime station ambience, faint wind through ironwork, a soft paper-wing flutter crossing the stereo field, the synchronized spoken line, one heavy clockwork click, and a distant rail hum.\nnon_diegetic_music: A minimal celesta phrase of three notes, resolving softly as the clock moves."
}