Oct 2 01:44 Model card: remove the model-downloads chart Oct 1 23:20 VTM-1.5.1: 1-step SFT distill (CFG baked in) + Ultra fast decoder + 60 fps demo Replaces VTM-1.5.1.pt with the 1-step distill of VTM-1.5.1-SFT (30-step teacher, CFG baked in; run at 1 step, CFG 1.0). Adds decoder/vtm-fast-decoder.pt (Ultra stream decoder) and a 60 fps demo recorded from the VTM Spark desk. Previous VTM-1.5.1.pt remains in history. 19:44 Card: get VTM Spark from vtmstudio.dev/download 19:41 VTM-1.5.1.pt = the release checkpoint (10-step teacher, 1 or 2 steps); card: how it works, download chart, commercial-use column, license: other Sep 30 05:19 Point the card at the renamed repo sinBoo1/VTM-Spark 05:14 Update model card: VTM-1.5.1, VTM Spark, demo, files and licences Sep 26 06:40 Add trackers/animeseg_hair3.pt 06:23 Upload VTM-1.5.1.pt latest model it adds SFT and AD at 7k steps and 6k steps 06:16 Delete VTM-1.5.1.pt 06:15 Delete VTM-ELF.pt Sep 4 20:32 Upload VTM-1.5.1.pt Aug 7 17:10 Add openseeface/priorbox_640x640.json 17:10 Add openseeface/mnv3_gaze32_split_opt.onnx 17:10 Add openseeface/mnv3_detection_opt.onnx 17:10 Add openseeface/retinaface_640x640_opt.onnx 17:10 Add openseeface/lm_model3_opt.onnx 17:10 Add trackers/pose_landmarker_lite.task 17:10 Add trackers/dwpose_v2.pt 17:10 Add trackers/iris_pose.pt 16:52 Upload VTM-ELF.pt 16:52 Delete VTM-elf.pt-000060000 Aug 2 16:48 Delete method illustration 16:48 Remove method illustration and caption 16:40 Update method figure: detailed face keypoints (brows, 3-pt eyes, mouth) 16:27 Simplify card: torso limits, shorter architecture, new illustration 16:19 Revise model card: architecture, 7 FPS note, method figure 16:13 Expand model card: VTuber pose-transfer beta docs 15:49 Update README.md 15:31 Upload VTM-elf.pt-000060000 base model 15:29 initial commit 
