Skip to content

Free, open-source AI video generator for Mac. Run LTX-2.5 and MiniMax H3 (Hailuo) locally on Apple Silicon: text-to-video and image-to-video with sound, lip-sync, character LoRA training, songs from lyrics (YuE2) and image editing (Qwen-Image-Edit). One-click install via Pinokio. Offline: no cloud, no API key, no NVIDIA GPU.

fix(hq): TeaCache default 2.0 → 1.0 (calibrated sweet spot)

mrbizarro68954c22693b
diff --git a/mlx_ltx_panel.py b/mlx_ltx_panel.py
index 512533f..516eca5 100644
--- a/mlx_ltx_panel.py
+++ b/mlx_ltx_panel.py
@@ -4484,7 +4484,15 @@ def make_job(form: dict[str, list[str]] | dict[str, str], *,
# the advanced fields.
"stage1_steps": max(1, int(f("stage1_steps", "10") or 10)),
"stage2_steps": max(0, int(f("stage2_steps", "3") or 3)),
- "teacache_thresh": float(f("teacache_thresh", "2.0") or 2.0),
+ # HQ TeaCache default — 1.0 is the calibrated sweet spot per
+ # ti2vid_two_stages_hq.LTX2_HQ_TEACACHE_THRESH ("~52% skip,
+ # ~2× speedup, sweet spot"). 2.0 was the panel's previous
+ # default and sat WAY past the calibration cliff at ~1.0,
+ # causing near-100% skip rate — late denoise steps reused
+ # ancient cached residuals and produced visible ghost-frame
+ # fusion (two faces, motion smear). Mr Bizarro caught this
+ # in a real render on 2026-05-18.
+ "teacache_thresh": float(f("teacache_thresh", "1.0") or 1.0),
"cfg_scale": float(f("cfg_scale", "3.0") or 3.0),
# Bongmath inner-loop cap for HQ res_2s sampler. Default 100
# (matches upstream). Lower values save latent algebra time
@@ -6155,7 +6163,10 @@ def run_job_inner(job: dict) -> None:
# by mistake (copy from standard params); fixed here.
"stg_scale": 0.0,
"enable_teacache": True,
- "teacache_thresh": float(p.get("teacache_thresh", 2.0)),
+ # HQ TeaCache default — see make_job comment. 1.0 is the
+ # calibrated sweet spot; 2.0 (prior default) sat past the
+ # skip cliff and produced ghost-frame artifacts.
+ "teacache_thresh": float(p.get("teacache_thresh", 1.0)),
"bongmath_max_iter": int(p.get("bongmath_max_iter", 100)),
"video_skip_step": int(p.get("video_skip_step", 0)),
"audio_skip_step": int(p.get("audio_skip_step", 0)),
@@ -8979,7 +8990,9 @@ class Handler(BaseHTTPRequestHandler):
"temporal_mode": ["native"],
"stage1_steps": [_val("stage1_steps", "10")],
"stage2_steps": [_val("stage2_steps", "3")],
- "teacache_thresh": [_val("teacache_thresh", "1.8")],
+ # HQ TeaCache default — was 1.8 (past calibrated cliff
+ # at ~1.0). Matches the /run + make_job default now.
+ "teacache_thresh": [_val("teacache_thresh", "1.0")],
"cfg_scale": [_val("cfg_scale", "3.0")],
"bongmath_max_iter": [_val("bongmath_max_iter", "100")],
"video_skip_step": [_val("video_skip_step", "1")],