Image-Text-to-Video
Diffusers
Safetensors
orbitquant
comfyui
w4
w4a4
native-w4a4-transformer-runtime
text-to-video
audio-video-generation
8-bit precision
Instructions to use WaveCut/MiniMax-H3-OrbitQuant-W4A4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use WaveCut/MiniMax-H3-OrbitQuant-W4A4 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("WaveCut/MiniMax-H3-OrbitQuant-W4A4", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
File size: 5,591 Bytes
fa2d87b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 | [
{
"id": "source-bf16-control",
"title": "Source BF16 control — dynamic zoom and dialogue",
"role": "source control",
"task": "t2va",
"seed": 74001,
"variant": "controlled_bf16_ablation",
"generation_seconds": 451.5754231447354,
"decode_seconds": 19.563463208731264,
"load_seconds": 1.5101884552277625,
"cuda_generation_peak_bytes": 68463336960,
"cuda_decode_peak_bytes": 12139359744,
"offload": {
"mode": "components_manager_auto_cpu_offload",
"device": "cuda",
"memory_reserve_margin": "64GB"
},
"quality_gate": {
"status": "pass",
"model_evaluations": 49,
"reviewed_frame_count": 26
},
"manual_review": {
"status": "pass",
"review_type": "manual_visual_temporal",
"reviewed_frame_indices": [
0,
5,
10,
15,
20,
25,
30,
35,
40,
45,
50,
55,
60,
65,
70,
75,
80,
85,
90,
95,
100,
105,
110,
115,
120,
123
],
"prompt_subject_recognizable": true,
"coherent_motion": true,
"repeating_tile_artifacts": false,
"motion_ghosting": false,
"texture_breakup": false,
"face_integrity": true,
"full_resolution_frame_reviewed": true,
"notes": "Dense every-fifth-frame and 6 FPS tail review preserves one face, aligned eyes and lips, and persistent wet-skin texture. Fast running frames contain ordinary motion blur that resolves in adjacent frames. This review is visual only; generated audio still requires a listening gate."
},
"h264_bytes": 635349,
"hevc_bytes": 375788
},
{
"id": "w4-text-control",
"title": "OrbitQuant W4A4 text encoder, BF16 denoiser",
"role": "component isolation control",
"task": "t2va",
"seed": 74001,
"variant": "controlled_bf16_ablation",
"generation_seconds": 381.5656377780251,
"decode_seconds": 13.154935840982944,
"load_seconds": 2.5672516343183815,
"cuda_generation_peak_bytes": 68464447488,
"cuda_decode_peak_bytes": 12139359744,
"offload": {
"mode": "components_manager_auto_cpu_offload",
"device": "cuda",
"memory_reserve_margin": "64GB"
},
"quality_gate": {
"status": "pass",
"model_evaluations": 49,
"reviewed_frame_count": 26
},
"manual_review": {
"status": "pass",
"review_type": "manual_visual_temporal",
"reviewed_frame_indices": [
0,
5,
10,
15,
20,
25,
30,
35,
40,
45,
50,
55,
60,
65,
70,
75,
80,
85,
90,
95,
100,
105,
110,
115,
120,
123
],
"prompt_subject_recognizable": true,
"coherent_motion": true,
"repeating_tile_artifacts": false,
"motion_ghosting": false,
"texture_breakup": false,
"face_integrity": true,
"full_resolution_frame_reviewed": true,
"audio_listening_status": "pass",
"audio_reviewed_by": "user",
"notes": "Dense every-fifth-frame and 6 FPS tail review preserves one face, aligned eyes and lips, and persistent wet-skin texture. Full-resolution frames 80 and 123 were inspected directly. The user reviewed the embedded video with audio and judged it ideal."
},
"h264_bytes": 753550,
"hevc_bytes": 449021
},
{
"id": "w4a4-native-t2va",
"title": "Final native W4A4 text + W4A4 T2VA transformer",
"role": "final native W4A4 T2VA recipe",
"task": "t2va",
"seed": 42,
"variant": "orbitquant_w4a4",
"generation_seconds": 114.6722432100214,
"decode_seconds": 4.709521219134331,
"load_seconds": 1.4625724367797375,
"cuda_generation_peak_bytes": 46359644160,
"cuda_decode_peak_bytes": 12139359744,
"offload": {
"mode": "manual_stage_offload",
"conditioner": "cuda_then_cpu",
"transformer": "cpu_then_cuda",
"vae": "cpu",
"audio_vae": "cpu"
},
"quality_gate": {
"status": "pass",
"model_evaluations": 49,
"reviewed_frame_count": 19
},
"manual_review": {
"status": "pass",
"review_type": "manual_visual_temporal_and_audio_signal",
"reviewed_frame_indices": [
0,
24,
48,
60,
72,
80,
84,
88,
90,
92,
96,
100,
104,
108,
110,
112,
116,
120,
123
],
"prompt_subject_recognizable": true,
"coherent_motion": true,
"repeating_tile_artifacts": false,
"motion_ghosting": false,
"texture_breakup": false,
"face_integrity": true,
"full_resolution_frame_reviewed": true,
"audio_signal_integrity": true,
"audio_user_listening_status": "pending",
"notes": "Full-timeline and dense late-motion grids plus full-resolution frames 60, 90, 110, and 123 preserve one coherent face, aligned eyes, connected lip motion, teeth, rain, and skin texture without the earlier grid, doubled contour, ghosting, or breakup. The HEVC/AAC file decodes to 124 frames at 608x480. Audio has no NaN/Inf samples or clipping, peaks at -0.27 dBFS, and has no persistent high-frequency narrow-band tone; energy above 8 kHz is 1.42%, below the accepted W4A16 control's 2.99%."
},
"h264_bytes": 815089,
"hevc_bytes": 474343
}
]
|