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.
Python ai-filmmakingai-videoai-video-generatorapple-siliconhailuoimage-to-videolip-synclocal-ailora-trainingltx-2ltx-2-5ltx-videomacosminimax-h3mlxmusic-generationoffline-aipinokiotext-to-videovideo-generation
helper: distinguish silent SIGKILL deaths from clean exits + step breadcrumbs
mrbizarro752458146bec
diff --git a/mlx_ltx_panel.py b/mlx_ltx_panel.pyindex 03d1803..d169be4 100644--- a/mlx_ltx_panel.py+++ b/mlx_ltx_panel.py@@ -1019,7 +1019,19 @@ class WarmHelper:raise RuntimeError(f"helper stdin closed: {exc}")ev = self._read_until(["done", "error", "exit"])if ev is None:- raise RuntimeError("helper died mid-job (no event)")+ # Silent death — no exit event, no Python error, just stdout pipe+ # closing. This is the OOM / SIGKILL / Metal-segfault fingerprint+ # because nothing else can kill the helper without giving Python+ # a chance to emit "exit" or "error". The atexit handler emits+ # "python_normal_exit" on normal shutdown; the SIGTERM handler+ # emits "sigterm(...)" before exit. If we get nothing at all,+ # SIGKILL or hard fault is the only explanation.+ raise RuntimeError(+ "helper killed by the OS (likely out of memory or a Metal fault). "+ "No Python exception was emitted before the helper died — that's "+ "the SIGKILL signature. Close memory-heavy apps and retry, "+ "or switch to Draft quality."+ )if ev.get("event") == "error":raise RuntimeError(ev.get("error", "helper error"))if ev.get("event") == "exit":@@ -3758,12 +3770,16 @@ async function poll() {// event back. Tell the user how to recover instead of leaving them// with the engine wording.const raw = (last.error || 'unknown error');+ const rawLower = raw.toLowerCase();let friendly, hint;- if (raw.toLowerCase().includes('helper died') || raw.toLowerCase().includes('helper exited')) {- friendly = 'Helper crashed — likely out of memory.';+ if (rawLower.includes('killed by the os') || rawLower.includes('sigkill') ||+ rawLower.includes('helper died') || rawLower.includes('helper exited')) {+ friendly = 'Helper killed by the OS — likely out of memory.';hint = 'Close memory-heavy apps (Chrome, Slack, iOS Simulator) and try again. ' +- 'Or switch Quality to Draft, which uses about half the RAM.';- } else if (raw.toLowerCase().includes('q8') || raw.toLowerCase().includes('keyframe')) {+ 'Or switch Quality to Draft, which uses about half the RAM. ' ++ 'If it keeps happening, check ~/Library/Logs/DiagnosticReports/ for ' ++ 'a crash report named python3.11_*.crash and share it on the GitHub repo.';+ } else if (rawLower.includes('q8') || rawLower.includes('keyframe')) {friendly = 'This mode needs the Q8 model.';hint = raw;} else {diff --git a/mlx_warm_helper.py b/mlx_warm_helper.pyindex e09caa3..5b2999f 100644--- a/mlx_warm_helper.py+++ b/mlx_warm_helper.py@@ -76,6 +76,43 @@ class LineEmitter(io.TextIOBase):sys.stdout = LineEmitter()sys.stderr = LineEmitter()+# ---- exit / signal tracing ---------------------------------------------------+# When users hit "no error, just stopped" silent crashes (cocktailpeanut on+# I2V, ~10s in), we currently can't tell if the helper exited cleanly, was+# SIGTERM'd by the panel, was SIGKILL'd by jetsam (macOS OOM), or hit a+# C-level fault in MLX/Metal. atexit fires on graceful Python exit (which+# also runs after our own emit({"exit"})), and the SIGTERM handler emits+# before the raise/exit. Neither catches SIGKILL — that's the diagnostic+# fingerprint. If the panel sees no exit/sigterm event AND the pipe closes,+# we KNOW it was SIGKILL (jetsam OOM) or a segfault.+import atexit, signal+_exit_emitted = False++def _emit_exit(reason: str) -> None:+ global _exit_emitted+ if _exit_emitted:+ return+ _exit_emitted = True+ try:+ emit({"event": "exit", "reason": reason})+ except Exception:+ pass++atexit.register(lambda: _emit_exit("python_normal_exit"))++def _sigterm_handler(signum, frame):+ _emit_exit(f"sigterm({signum})")+ sys.exit(0)++# SIGTERM is what the panel sends on /helper/restart and at panel shutdown.+# SIGINT is Ctrl+C from a user running the helper directly. Both exit cleanly.+# SIGKILL can't be caught — that's by design, and is the OOM fingerprint.+for _sig in (signal.SIGTERM, signal.SIGINT):+ try:+ signal.signal(_sig, _sigterm_handler)+ except (ValueError, OSError):+ pass+# ---- idle reaper -------------------------------------------------------------_last_activity = time.time()_is_busy = False # set during active generation; reaper skips while True@@ -338,7 +375,16 @@ for line in sys.__stdin__:_is_busy = Truetry:t0 = time.time()+ # Granular breadcrumbs so a silent helper death is traceable:+ # if the panel's last log line is "step:get_pipe" then we died+ # during pipeline init (likely OOM or weight-load issue). If+ # the last line is "step:generate_and_save" we died during+ # the actual gen (denoising / VAE decode). Without this the+ # last visible line was the original "Loading I2V pipeline..."+ # message and users had no idea where to look.+ emit({"event": "log", "line": f"step:get_pipe kind={('i2v' if needs_image else 't2v')}"})pipe = get_pipe("i2v" if needs_image else "t2v")+ emit({"event": "log", "line": "step:get_pipe done"})kwargs = dict(prompt=p["prompt"],@@ -368,7 +414,9 @@ for line in sys.__stdin__:else:kwargs["image"] = None+ emit({"event": "log", "line": f"step:generate_and_save mode={mode} {kwargs['width']}x{kwargs['height']} {kwargs['num_frames']}f steps={kwargs['num_steps']}"})out_path = pipe.generate_and_save(**kwargs)+ emit({"event": "log", "line": "step:generate_and_save done"})elapsed = round(time.time() - t0, 2)_last_activity = time.time()emit({