 Hawk Lee
|
eb96aaa966
|
style: Reduce blink hold duration to 1-3 frames for sharper blink
|
2026-01-21 17:47:34 +08:00 |
|
|
Hawk Lee
|
0f75bdde56
|
feat: Implement Blink Hold logic to extend closed-eye duration for more natural realism
|
2026-01-21 17:24:30 +08:00 |
|
|
Hawk Lee
|
ae30694c74
|
style: Refine blink parameters (lower double blink prob, increase gap) for smoother look
|
2026-01-21 17:21:25 +08:00 |
|
|
Hawk Lee
|
a6faa43b3a
|
feat: Implement Dynamic Blink Rate based on VAD (Fast during speech, Slow during silence)
|
2026-01-21 17:16:55 +08:00 |
|
|
Hawk Lee
|
5aaf9d8bb0
|
feat: Tune Normal Blink rate to 50-85 frames (~23bpm) to better match human speaking frequency
|
2026-01-21 17:14:06 +08:00 |
|
|
Hawk Lee
|
75ef9cb90c
|
docs: Update Ditto Sampler docs with Seed, Double Blink, and Exact Onset Alignment details
|
2026-01-21 17:09:53 +08:00 |
|
|
Hawk Lee
|
a751dd3ac6
|
feat: Implement Double Blink logic with randomized interval for natural eye motion
|
2026-01-21 17:08:31 +08:00 |
|
|
Hawk Lee
|
773477d768
|
fix: Initialize self.seed in ComfyStreamSDK.setup explicitly to avoid AttributeError
|
2026-01-21 16:55:30 +08:00 |
|
|
Hawk Lee
|
7dfd9197cb
|
fix: Add seed input to AIIA_DittoSampler node definition to resolve NameError
|
2026-01-21 16:51:39 +08:00 |
|
|
Hawk Lee
|
5320d5048b
|
fix: Implement RNG Seed Reset on Speech Onset to ensure diffusion noise consistency and lip sync accuracy
|
2026-01-21 16:44:50 +08:00 |
|
|
Hawk Lee
|
bf2958b021
|
fix: Initialize do_reset default value to prevent UnboundLocalError
|
2026-01-21 16:19:36 +08:00 |
|
|
Hawk Lee
|
aeb28f246d
|
fix: Implement Exact Onset Alignment via Variable Step Length to perfectly mimic Clean Start conditions and eliminate jitter
|
2026-01-21 16:03:07 +08:00 |
|
|
Hawk Lee
|
a607adaca0
|
fix: Reset clip_idx to 0 on onset reset to enforce exact First Segment conditions
|
2026-01-21 15:29:26 +08:00 |
|
|
Hawk Lee
|
407ae60571
|
fix: Implement Duration-Based Onset Reset (min 15 frames) to avoid over-resetting on breaths
|
2026-01-21 14:49:07 +08:00 |
|
|
Hawk Lee
|
50eefd636f
|
fix: Robust Sub-Chunk Speech Onset Detection to fix reset failure in short silence gaps
|
2026-01-21 14:40:24 +08:00 |
|
|
Hawk Lee
|
6b65b228e3
|
fix: Disable inner 'dit' tqdm to prevent console log conflict
|
2026-01-21 14:30:03 +08:00 |
|
|
Hawk Lee
|
307e027807
|
fix: Implement Speech Onset Hard Reset to eliminate drift at speech start
|
2026-01-21 14:13:47 +08:00 |
|
|
Hawk Lee
|
c90a0948f9
|
fix: Switch from Hard Reset to Soft Reset (50%) to dampen drift without oscillation
|
2026-01-21 14:01:34 +08:00 |
|
|
Hawk Lee
|
8dfdeadc47
|
fix: Pass total_frames to SDK setup to enable Ditto console progress bar
|
2026-01-21 13:58:20 +08:00 |
|
|
Hawk Lee
|
80923f2e29
|
fix: Initialize vad_timeline in ComfyStreamSDK.setup to ensure attribute existence
|
2026-01-21 13:51:42 +08:00 |
|
|
Hawk Lee
|
a1f511f9d3
|
fix: Use getattr for vad_timeline to prevent AttributeError on cached SDK instances
|
2026-01-21 13:49:19 +08:00 |
|
|
Hawk Lee
|
5a45e66907
|
fix: Implement Hard State Reset for Auto-Correction of LMDM Drift during silence
|
2026-01-21 13:44:13 +08:00 |
|
|
Hawk Lee
|
f7bced771a
|
feat: Add tqdm console progress bar to Video Combine node frame saving
|
2026-01-21 13:21:47 +08:00 |
|
|
Hawk Lee
|
076100e3db
|
fix: Direct tqdm output to stdout for console visibility
|
2026-01-21 13:20:04 +08:00 |
|
|
Hawk Lee
|
36fa0475ba
|
feat: Replace manual print logging with tqdm progress bar in console
|
2026-01-21 13:18:01 +08:00 |
|
|
Hawk Lee
|
7698be1bfb
|
fix: Check type of ditto_config before unpacking to handle both dict and object
|
2026-01-21 13:14:29 +08:00 |
|
|
Hawk Lee
|
f8e71620e6
|
docs: Clarify crop_scale parameter to explain Face Processing ROI vs Content Crop
|
2026-01-21 13:13:38 +08:00 |
|
|
Hawk Lee
|
16592b3256
|
fix: Define ditto_config by unpickling pipe data to fix NameError
|
2026-01-21 13:09:36 +08:00 |
|
|
Hawk Lee
|
8e4306040f
|
fix: Replace undefined ditto_config with ditto_model in setup call
|
2026-01-21 13:04:14 +08:00 |
|
|
Hawk Lee
|
819547da37
|
docs: Update README to reflect Relative VAD Threshold and Drift Correction features
|
2026-01-21 13:02:49 +08:00 |
|
|
Hawk Lee
|
00c6ce76e2
|
fix: Restore correct arguments for master_sdk.setup call to fix NameError
|
2026-01-21 13:00:22 +08:00 |
|
|
Hawk Lee
|
cdd1581a0d
|
feat: Implement VAD-Triggered LMDM State Reset to clear accumulated drift during silences
|
2026-01-21 12:56:15 +08:00 |
|
|
Hawk Lee
|
42e78d72ce
|
fix: Audio2Motion smoothing applied to wrong range (only seam), causing jittery motion on long clips
|
2026-01-21 12:37:45 +08:00 |
|
|
Hawk Lee
|
3d79447cff
|
feat: Implement Relative RMS Thresholding for VAD to handle volume variations
|
2026-01-21 12:30:14 +08:00 |
|
|
Hawk Lee
|
5f48b18d51
|
feat: Implement VAD Gap Filling and Spike Removal to stabilize mouth motion
|
2026-01-21 12:26:17 +08:00 |
|
|
Hawk Lee
|
9b775acbb8
|
feat: Switch VAD to Linear Ramping (Smoother Attack/Release)
|
2026-01-21 12:15:50 +08:00 |
|
|
Hawk Lee
|
e6681a6220
|
feat: Implement Micro-Motion (head sway) during silence
|
2026-01-21 12:07:02 +08:00 |
|
|
Hawk Lee
|
df204e2776
|
feat: Tune VAD Release coeff to 0.80 for faster natural closing
|
2026-01-21 12:03:48 +08:00 |
|
|
Hawk Lee
|
77248c539a
|
docs: Explicitly mention Instant Attack behavior in README
|
2026-01-21 11:28:22 +08:00 |
|
|
Hawk Lee
|
84148a7c1a
|
feat: Instant VAD Attack and VAD stats logging, lower threshold to 0.005
|
2026-01-21 11:25:31 +08:00 |
|
|
Hawk Lee
|
654fcc5b43
|
feat: Rename auto_ref_frame to relax_on_silence and implement VAD Attack/Release
|
2026-01-21 09:59:21 +08:00 |
|
|
Hawk Lee
|
881bae2742
|
feat: Refine VAD to preserve blinks/head motion and force INFO logging
|
2026-01-21 09:47:45 +08:00 |
|
|
Hawk Lee
|
d785e9378d
|
feat: Add smoothing to VAD transitions for natural silence handling
|
2026-01-21 09:36:00 +08:00 |
|
|
Hawk Lee
|
c7c0d722c1
|
fix: Robust logging protection against ABSL hijack
|
2026-01-21 09:33:06 +08:00 |
|
|
Hawk Lee
|
169fdbad69
|
docs: Update Ditto params description in README
|
2026-01-21 09:29:36 +08:00 |
|
|
Hawk Lee
|
cd0a8d23b5
|
feat: Add VAD-based silence handling and Blink rate control for Ditto
|
2026-01-21 01:46:04 +08:00 |
|
|
Hawk Lee
|
7dddbbb0e0
|
fix: Wrap Ditto execution with logging restoration to prevent hijack
|
2026-01-21 01:35:40 +08:00 |
|
|
Hawk Lee
|
096608ac4a
|
fix: Add explicit console logging to Ditto sampler
|
2026-01-21 01:29:12 +08:00 |
|
|
Hawk Lee
|
ed65157bd0
|
chore: Group FLOAT with Talking Head models in metadata
|
2026-01-21 01:16:25 +08:00 |
|
|
Hawk Lee
|
b2453bb325
|
chore: Update pyproject.toml description and keywords for Ditto/FLOAT
|
2026-01-21 01:15:14 +08:00 |
|