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.

docs: refresh Extend Q8 release notes

mrbizarro3421a67911d9
diff --git a/README.md b/README.md
index fa2cddc..76333ee 100644
--- a/README.md
+++ b/README.md
@@ -147,7 +147,7 @@ and step count change.
| **Quick** | Q4 distilled | 640×480 · ~2 min | Fast sanity checks. 4:3 native output. |
| **Balanced** | Q4 distilled + Lanczos export | 1024×576 → 1280×720 · ~4–5 min | Default. Standard 16:9 delivery without cropping; good first bet for faster HD-looking clips. |
| **Standard** | Q4 distilled | 1280×704 · ~8 min | The daily driver. Q4 weights (~20 GB on disk). Boost/Turbo Speed shaves ~23 % / ~34 %. |
-| **High** | Q8 two-stage + TeaCache | 1280×704 · ~12 min | Sharper detail, fewer artifacts on faces and text. Optional Q8 download (~37 GB extra). Required for FFLF. |
+| **High** | Q8 two-stage + TeaCache | 1280×704 · ~12 min | Sharper detail, fewer artifacts on faces and text. Optional Q8 download (~37 GB extra). Required for FFLF and Extend. |
The **Export** control can also run a lightweight ffmpeg Lanczos pass after
generation:
@@ -404,17 +404,17 @@ ffmpeg -i in.mp4 \
## Performance reference
-Wall-clock times on an **M4 Mac Studio, 64 GB**:
+Wall-clock times on an **M4-class Mac, 64 GB**:
| Mode | Resolution | Frames | Steps | ~Time |
|---|---|---|---|---|
| T2V Quick | 640×480 | 121 (5s) | 8 | ~2 min |
-| T2V Standard | 1280×704 | 121 (5s) | 8 | ~7 min |
-| I2V Standard | 1280×704 | 121 (5s) | 8 | ~7 min |
-| Extend (Q4 dev, cfg=1.0) | 768×416 | +6 latents (~2s) | 12 | ~12 min |
-| Extend (Q4 dev, cfg=3.0 "Quality") | 1280×704 | +6 latents | 12 | ~30 min |
+| T2V Balanced | 1024×576 → 1280×720 | 121 (5s) | 8 | ~5 min |
+| T2V Standard | 1280×704 | 121 (5s) | 8 | ~8 min |
+| I2V Standard | 1280×704 | 121 (5s) | 8 | ~9 min |
| High (Q8 two-stage) | 1280×704 | 121 | s1=15 + s2=3 | ~12 min |
-| FFLF (Q8) | 768×416 | 121 | s1=15 + s2=3 | ~5 min |
+| FFLF (Q8) | 768×416 | 121 | s1=15 + s2=3 | ~6 min |
+| Extend (Q8 dev, cfg=1.0) | 768×416 | +3s | 12 | ~16 min |
Experimental Speed smoke test on the same 64 GB machine, I2V Standard,
768×512, 121 frames, 8 steps:
diff --git a/mlx_ltx_panel.py b/mlx_ltx_panel.py
index 79bf959..cfc2785 100644
--- a/mlx_ltx_panel.py
+++ b/mlx_ltx_panel.py
@@ -1792,8 +1792,8 @@ def open_pinokio() -> None:
# T2V/I2V Q8 HQ at 1280×704: ~50 GB peak, fits 64 GB cleanly
# FFLF (Q8 dev) at 1280×704: > 64 GB peak, OOMs into swap
# FFLF (Q8 dev) at 768×416: ~48 GB peak, fits 64 GB
-# Extend (Q4 dev) at 1280×704: ~50 GB peak + 12 GB swap (240s/step)
-# Extend (Q4 dev) at 768×416: ~46 GB peak, no swap (54s/step)
+# Extend (Q8 dev) at 1280×704: > 64 GB peak, OOMs into swap
+# Extend (Q8 dev) at 768×416: ~48 GB peak, fits 64 GB
#
# `t2v_max_dim` / `i2v_max_dim` clamp T2V/I2V resolution. `keyframe_max_dim`
# and `extend_max_dim` clamp those modes' resolutions specifically (since