Audio Wave: fixed 721-frame subsegment grid + per-segment select/crop + notes box auto-fill

Pivot to the user's model: segments are a fixed grid of subsegment_frames frames
(default 721 @ 24fps = one LTX clip), not arbitrary clicks. New inputs:
subsegment_frames (grid size) and segment_select (0=all, N=output ONLY chunk N —
crops waveform_image, AUDIO, and summary so you can generate/skip one beat at a time).
_render gains window-crop + frame markers + per-segment time/frame labels. JS rewritten:
draws the fixed grid, auto-fills the notes box with one segN: line per chunk (type or
dblclick to note), click-to-seek, playhead + time readout. Workflows updated for the
new widgets.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

README.md

@@ -112,12 +112,19 @@ adds to feed the judge's `system_prompt`/`user_prompt`/`axes` sockets.)
 ### Interactive: `Audio Wave + Segments`
 
 Same outputs, but with an in-node waveform widget: **upload** an audio clip, **play** it,
-and **click the waveform** to place segment boundaries — click to add a split, drag to
-move, double-click a segment for its note, shift/right-click to delete. The boundaries +
-notes are saved to a hidden `segments_json` and drive the segmentation (falls back to
-auto-split / the `notes` syntax if none placed). Outputs `waveform_image`, `audio_summary`,
-and `audio`. Needs `torchaudio`/`soundfile`/`librosa` to load the file (torchaudio usually
-ships with torch).
+click to **seek**. The audio is split into a **fixed grid of `subsegment_frames` frames**
+(default **721 @ 24fps ≈ 30.04s** — the LTX clip length), drawn as the green grid; each chunk
+is one beat. The `notes` box **auto-fills with one `segN:` line per chunk** — type your motion
+note after each (or double-click a chunk on the waveform to set its note). The `waveform_image`
+carries per-chunk labels (energy, start time, frame count).
+
+- **`subsegment_frames`** — the fixed chunk size (0 = fall back to manual boundaries in `segments_json`).
+- **`segment_select`** — `0` = whole clip; `N` = output **only subsegment N**: the `waveform_image`
+  is cropped to that chunk, the `audio` output is cropped to it, and the summary is just that beat —
+  so you can generate/skip **one beat at a time**.
+
+Outputs `waveform_image`, `audio_summary`, `audio`. Needs `torchaudio`/`soundfile`/`librosa`
+to load the file (torchaudio usually ships with torch).
 
 ## Performance / speed
 

nodes/audio_guide.py

@@ -130,27 +130,44 @@ def _segments(rms_n, times, duration, fps, max_segments, beats):
     return segs, bounds
 
 
-def _render(rms_n, times, duration, beats, bounds, segs):
-    """Render the envelope + beats + segment boundaries to a ComfyUI IMAGE tensor."""
+def _render(rms_n, times, duration, beats, bounds, segs,
+            fps=None, frames_marker=0, window=None):
+    """Render the envelope + beats + segment boundaries + time labels to a ComfyUI IMAGE.
+    window=(t0,t1) crops to that time span (for a selected segment). frames_marker draws
+    a green line every N frames (N/fps seconds) — LTX clip-length grid."""
     W, H = 1024, 256
     img = Image.new("RGB", (W, H), (18, 18, 22))
     d = ImageDraw.Draw(img)
-    dur = max(duration, 1e-6)
+    t0, t1 = window if window else (0.0, duration)
+    span = max(t1 - t0, 1e-6)
 
     def X(t):
-        return int(max(0, min(W - 1, t / dur * (W - 1))))
+        return int(max(0, min(W - 1, (t - t0) / span * (W - 1))))
 
-    pts = [(0, H)] + [(X(times[i]), H - int(rms_n[i] * (H - 26))) for i in range(len(rms_n))] + [(W - 1, H)]
+    pts = [(0, H)] + [(X(times[i]), H - int(rms_n[i] * (H - 30)))
+                      for i in range(len(rms_n)) if t0 <= times[i] <= t1] + [(W - 1, H)]
+    if len(pts) > 2:
         d.polygon(pts, fill=(60, 140, 220))
+    if fps and frames_marker:                         # frame grid (green)
+        step = frames_marker / fps
+        k = 1
+        while k * step < duration + 1e-6:
+            mt = k * step
+            if t0 <= mt <= t1:
+                d.line([(X(mt), 0), (X(mt), H)], fill=(70, 200, 120), width=1)
+                d.text((X(mt) + 2, H - 13), f"{frames_marker * k}f", fill=(70, 200, 120))
+            k += 1
     for b in beats:                                   # beat markers (orange)
+        if t0 <= b <= t1:
             d.line([(X(b), 0), (X(b), H)], fill=(230, 110, 60), width=1)
-    for bd in bounds:                                 # segment boundaries (white)
-        d.line([(X(bd), 0), (X(bd), H)], fill=(240, 240, 240), width=1)
-    for s in segs:
-        x = X(s["start_s"]) + 4
-        d.text((x, 4), f"S{s['segment']} {s['energy']}", fill=(255, 255, 255))
+    for s in segs:                                    # boundaries + label + time + note
+        if not (t0 <= s["start_s"] <= t1):
+            continue
+        x = X(s["start_s"])
+        d.line([(x, 0), (x, H)], fill=(240, 240, 240), width=1)
+        d.text((x + 4, 4), f"S{s['segment']} {s['energy']} {s['start_s']}s/{s['frames']}f", fill=(255, 255, 255))
         if s.get("note"):                             # per-segment note (amber)
-            d.text((x, 18), s["note"][:30], fill=(255, 210, 110))
+            d.text((x + 4, 18), s["note"][:30], fill=(255, 210, 110))
     arr = np.asarray(img, dtype=np.float32) / 255.0
     return torch.from_numpy(arr)[None, ...]           # [1, H, W, 3]
 

nodes/audio_wave_segments.py

@@ -88,6 +88,33 @@ def _audio_files():
     return files or [_NO_AUDIO]
 
 
+def _segments_fixed(rms_n, times, duration, fps, sub_frames, beats):
+    """Split into fixed chunks of `sub_frames` frames (the LTX clip length). The last
+    chunk gets whatever's left (snapped to 8n+1)."""
+    step = max(sub_frames, 1) / max(fps, 1)           # seconds per chunk
+    n = max(1, int(np.ceil(duration / step - 1e-6)))
+    stages = ["establish", "build", "peak", "settle"]
+    segs, bounds = [], [0.0]
+    for i in range(n):
+        t0 = i * step
+        t1 = min(duration, (i + 1) * step)
+        bounds.append(round(t1, 3))
+        mask = (times >= t0) & (times < t1)
+        e = float(rms_n[mask].mean()) if mask.any() else 0.0
+        peak_t = float(times[mask][np.argmax(rms_n[mask])]) if mask.any() else t0
+        dur = round(t1 - t0, 2)
+        frames = sub_frames if i < n - 1 else _snap8(round(dur * fps))
+        label = "high" if e > 0.66 else ("medium" if e > 0.33 else "low")
+        stage = stages[i] if i < len(stages) else ("peak" if e > 0.6 else "settle")
+        segs.append({
+            "segment": i + 1, "start_s": round(t0, 2), "duration_s": dur, "frames": frames,
+            "energy": label, "energy_val": round(e, 3), "peak_s": round(peak_t, 2),
+            "stage_hint": stage, "note": "",
+            "beats_in": [round(b, 2) for b in beats if t0 <= b < t1],
+        })
+    return segs, np.array(bounds)
+
+
 class AudioWaveSegments:
     CATEGORY = "prompt_calibrator"
     FUNCTION = "run"
@@ -101,25 +128,29 @@ class AudioWaveSegments:
                 # Pick a file from ComfyUI/input, or use the widget's "upload" button (JS).
                 "audio": (_audio_files(),),
                 "fps": ("INT", {"default": 24, "min": 1, "max": 120}),
-                "max_segments": ("INT", {"default": 6, "min": 3, "max": 12}),
+                # Fixed LTX clip length. Segments = chunks of this many frames (721@24fps
+                # ~= 30.04s). This is also the waveform grid. Set 0 to use segments_json.
+                "subsegment_frames": ("INT", {"default": 721, "min": 0, "max": 100000}),
+                # 0 = whole clip. N = output ONLY subsegment N (crops image + audio + summary),
+                # so you can generate/skip one beat at a time.
+                "segment_select": ("INT", {"default": 0, "min": 0, "max": 999}),
                 "notes": ("STRING", {"default": "", "multiline": True}),
-                # Written by the JS waveform widget: [{"start_s": 0.0, "note": "..."}, ...].
-                # Leave as "[]" to auto-split. Also editable by hand.
+                # Machine field: per-chunk notes/boundaries from the JS widget (hidden).
                 "segments_json": ("STRING", {"default": "[]"}),
             },
         }
 
     @classmethod
-    def IS_CHANGED(cls, audio, fps, max_segments, notes, segments_json):
+    def IS_CHANGED(cls, audio, fps, subsegment_frames, segment_select, notes, segments_json):
         try:
             import folder_paths
             p = folder_paths.get_annotated_filepath(audio)
             mt = os.path.getmtime(p) if os.path.isfile(p) else ""
         except Exception:
             mt = ""
-        return f"{audio}|{fps}|{max_segments}|{notes}|{segments_json}|{mt}"
+        return f"{audio}|{fps}|{subsegment_frames}|{segment_select}|{notes}|{segments_json}|{mt}"
 
-    def run(self, audio, fps, max_segments, notes, segments_json):
+    def run(self, audio, fps, subsegment_frames, segment_select, notes, segments_json):
         try:
             import folder_paths
             path = folder_paths.get_annotated_filepath(audio)
@@ -135,28 +166,44 @@ class AudioWaveSegments:
         rms_n, times = _rms_envelope(y, sr)
         bpm, beats = _tempo_beats(y, sr)
 
-        # User boundaries from the JS widget, else auto-split.
-        starts, seg_notes = [], {}
+        if subsegment_frames > 0:                     # fixed LTX-clip grid
+            segs, bounds = _segments_fixed(rms_n, times, duration, fps, subsegment_frames, beats)
+        else:                                         # manual boundaries from the JS widget
+            starts = []
             try:
-            data = json.loads(segments_json) if segments_json.strip() else []
-            for i, seg in enumerate(data):
+                for seg in (json.loads(segments_json) if segments_json.strip() else []):
                     starts.append(float(seg.get("start_s", 0.0)))
-                seg_notes[i + 1] = str(seg.get("note", "") or "")
             except Exception as e:
-            print(f"[AudioWaveSegments] bad segments_json ({e}); auto-splitting.")
-            data = []
-
-        if data:
+                print(f"[AudioWaveSegments] bad segments_json ({e})")
             segs, bounds = _segments_from_boundaries(rms_n, times, duration, fps, starts, beats)
-            for s in segs:                            # notes placed on the waveform
-                s["note"] = seg_notes.get(s["segment"], "")
-        else:
-            segs, bounds = _segments(rms_n, times, duration, fps, max_segments, beats)
 
-        global_notes = _attach_notes(segs, notes)     # merge the text-box notes on top
-        image = _render(rms_n, times, duration, beats, bounds, segs)
+        # per-chunk notes from segments_json (JS) then the notes-box (segN: syntax) on top.
+        try:
+            for i, seg in enumerate(json.loads(segments_json) if segments_json.strip() else []):
+                if i < len(segs) and seg.get("note"):
+                    segs[i]["note"] = str(seg["note"])
+        except Exception:
+            pass
+        global_notes = _attach_notes(segs, notes)
+
+        window, render_segs = None, segs
+        audio_out = {"waveform": wav.unsqueeze(0), "sample_rate": sr}
+        sel = int(segment_select)
+        if 1 <= sel <= len(segs):                     # crop to one subsegment
+            seg = segs[sel - 1]
+            s0 = seg["start_s"]
+            s1 = min(duration, s0 + seg["duration_s"])
+            a, b = int(s0 * sr), int(s1 * sr)
+            if b > a:
+                audio_out = {"waveform": wav[:, a:b].unsqueeze(0), "sample_rate": sr}
+            window, render_segs = (s0, s1), [seg]
+            summary = (f"SELECTED SUBSEGMENT {sel} of {len(segs)} — generate ONLY this beat.\n"
+                       + _summary(duration, sr, bpm, beats, [seg], global_notes))
+        else:
             summary = _summary(duration, sr, bpm, beats, segs, global_notes)
-        audio_out = {"waveform": wav.unsqueeze(0), "sample_rate": sr}  # [1, C, N]
+
+        image = _render(rms_n, times, duration, beats, bounds, render_segs,
+                        fps=fps, frames_marker=subsegment_frames, window=window)
         return (image, summary, audio_out)
 
 

web/audio_wave.js

@@ -1,5 +1,5 @@
-// Audio Wave + Segments — waveform display + playback + click-to-segment.
-// Writes the segment boundaries/notes into the node's hidden `segments_json` widget.
+// Audio Wave + Segments — waveform + playback, a fixed subsegment grid (721 frames @ fps),
+// per-chunk notes auto-filled into the `notes` box, playhead + click-to-seek.
 // First cut: open the browser console for [audiowave] logs if something misbehaves.
 import { app } from "../../scripts/app.js";
 import { api } from "../../scripts/api.js";
@@ -13,20 +13,16 @@ function computePeaks(buf, n) {
   let max = 1e-6;
   for (let i = 0; i < n; i++) {
     let m = 0;
-    const start = i * block;
-    for (let j = 0; j < block && start + j < data.length; j++) {
-      const v = Math.abs(data[start + j]);
-      if (v > m) m = v;
-    }
-    peaks[i] = m;
-    if (m > max) max = m;
+    const s = i * block;
+    for (let j = 0; j < block && s + j < data.length; j++) { const v = Math.abs(data[s + j]); if (v > m) m = v; }
+    peaks[i] = m; if (m > max) max = m;
   }
   for (let i = 0; i < n; i++) peaks[i] /= max;
   return peaks;
 }
 
 function setupWave(node) {
-  const st = { duration: 0, peaks: null, boundaries: [], notes: {}, audio: new Audio(), playing: false };
+  const st = { duration: 0, peaks: null, audio: new Audio(), playing: false };
   node._wave = st;
 
   const wrap = document.createElement("div");
@@ -34,29 +30,60 @@ function setupWave(node) {
   const bar = document.createElement("div");
   bar.style.cssText = "display:flex;gap:6px;align-items:center;font-size:10px;color:#bbb;flex-wrap:wrap;";
   const mk = (t) => { const b = document.createElement("button"); b.textContent = t; b.style.cssText = "font-size:10px;padding:1px 6px;"; return b; };
-  const playBtn = mk("▶ play"), clearBtn = mk("clear"), upBtn = mk("upload");
-  const hint = document.createElement("span");
-  hint.textContent = "click=add split · drag=move · dblclick=note · shift/right-click=delete";
-  bar.append(playBtn, clearBtn, upBtn, hint);
+  const playBtn = mk("▶ play"), upBtn = mk("upload");
+  const readout = document.createElement("span"); readout.textContent = "0.00 / 0.00s";
+  const hint = document.createElement("span"); hint.textContent = "click=seek · dblclick a chunk=note";
+  bar.append(playBtn, upBtn, readout, hint);
   const canvas = document.createElement("canvas");
-  canvas.width = 640; canvas.height = 150;
-  canvas.style.cssText = "width:100%;height:150px;background:#141418;border-radius:4px;cursor:crosshair;";
+  canvas.width = 640; canvas.height = 160;
+  canvas.style.cssText = "width:100%;height:160px;background:#141418;border-radius:4px;cursor:pointer;";
   wrap.append(bar, canvas);
   node.addDOMWidget("wave", "wave", wrap, { serialize: false });
   const ctx = canvas.getContext("2d");
 
-  // Hide the raw segments_json widget — it's machine data driven by the waveform.
-  const sjw = getW(node, "segments_json");
+  const sjw = getW(node, "segments_json");            // machine field — hide it
   if (sjw) { sjw.hidden = true; sjw.computeSize = () => [0, -4]; }
 
+  const fps = () => Math.max(1, Number(getW(node, "fps")?.value) || 24);
+  const subFrames = () => Math.max(0, Number(getW(node, "subsegment_frames")?.value) || 0);
   const t2x = (t) => (st.duration ? (t / st.duration) * canvas.width : 0);
   const x2t = (x) => (st.duration ? (x / canvas.width) * st.duration : 0);
-  const starts = () => [0, ...st.boundaries.slice().sort((a, b) => a - b)];
 
-  function serialize() {
-    const segs = starts().map((s, i) => ({ start_s: Math.round(s * 100) / 100, note: st.notes[i] || "" }));
-    const w = getW(node, "segments_json");
-    if (w) { w.value = JSON.stringify(segs); w.callback?.(w.value); }
+  function chunkStarts() {
+    const step = subFrames() / fps();
+    const arr = [];
+    if (!st.duration || !step) return [0];
+    for (let t = 0; t < st.duration - 1e-6; t += step) arr.push(t);
+    return arr.length ? arr : [0];
+  }
+
+  function parseNotes(txt) {
+    const seg = {}, glob = [];
+    (txt || "").split("\n").forEach((l) => {
+      const m = l.match(/^\s*(?:seg(?:ment)?|s)?\s*(\d+)\s*[:)]\s*(.*)$/i);
+      if (m) seg[parseInt(m[1])] = m[2]; else if (l.trim()) glob.push(l);
+    });
+    return { seg, glob };
+  }
+
+  function syncNotesBox() {                            // one "segN:" line per chunk, preserve notes
+    const nw = getW(node, "notes"); if (!nw) return;
+    const { seg, glob } = parseNotes(nw.value);
+    const n = chunkStarts().length;
+    const lines = [];
+    for (let i = 1; i <= n; i++) lines.push(`seg${i}: ${seg[i] !== undefined ? seg[i] : ""}`);
+    const val = lines.concat(glob).join("\n");
+    if (nw.value !== val) { nw.value = val; nw.callback?.(val); }
+  }
+
+  function setChunkNote(i, val) {
+    const nw = getW(node, "notes"); if (!nw) return;
+    const { seg, glob } = parseNotes(nw.value);
+    seg[i] = val;
+    const n = chunkStarts().length;
+    const lines = [];
+    for (let k = 1; k <= n; k++) lines.push(`seg${k}: ${seg[k] !== undefined ? seg[k] : ""}`);
+    nw.value = lines.concat(glob).join("\n"); nw.callback?.(nw.value);
   }
 
   function draw() {
@@ -64,57 +91,49 @@ function setupWave(node) {
     if (st.peaks) {
       ctx.fillStyle = "#3c8cdc";
       const n = st.peaks.length, bw = canvas.width / n;
-      for (let i = 0; i < n; i++) { const h = st.peaks[i] * (canvas.height - 26); ctx.fillRect(i * bw, canvas.height - h, Math.max(1, bw), h); }
+      for (let i = 0; i < n; i++) { const h = st.peaks[i] * (canvas.height - 30); ctx.fillRect(i * bw, canvas.height - h, Math.max(1, bw), h); }
     }
     ctx.font = "10px monospace";
-    starts().forEach((s, i) => {
+    const { seg } = parseNotes(getW(node, "notes")?.value);
+    const starts = chunkStarts();
+    starts.forEach((s, i) => {
       const x = t2x(s);
-      ctx.strokeStyle = "#eee"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, canvas.height); ctx.stroke();
-      ctx.fillStyle = "#fff"; ctx.fillText("S" + (i + 1), x + 3, 11);
-      if (st.notes[i]) { ctx.fillStyle = "#ffd27a"; ctx.fillText(st.notes[i].slice(0, 24), x + 3, 23); }
+      ctx.strokeStyle = "#7ec8a0"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, canvas.height); ctx.stroke();
+      ctx.fillStyle = "#fff"; ctx.fillText(`S${i + 1} ${s.toFixed(1)}s`, x + 3, 11);
+      if (seg[i + 1]) { ctx.fillStyle = "#ffd27a"; ctx.fillText(seg[i + 1].slice(0, 22), x + 3, 23); }
     });
-    if (st.playing) { const x = t2x(st.audio.currentTime); ctx.strokeStyle = "#ff5a3c"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, canvas.height); ctx.stroke(); }
+    if (st.playing || st.audio.currentTime) {
+      const x = t2x(st.audio.currentTime);
+      ctx.strokeStyle = "#ff5a3c"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, canvas.height); ctx.stroke();
+    }
+    readout.textContent = `${(st.audio.currentTime || 0).toFixed(2)} / ${st.duration.toFixed(2)}s`;
   }
 
   async function loadFile(name) {
-    if (!name) return;
+    if (!name || name.startsWith("(")) return;
     const url = `/view?filename=${encodeURIComponent(name)}&type=input&subfolder=`;
     try {
       const ab = await (await fetch(url)).arrayBuffer();
       const ac = new (window.AudioContext || window.webkitAudioContext)();
       const buf = await ac.decodeAudioData(ab.slice(0));
       st.duration = buf.duration; st.peaks = computePeaks(buf, canvas.width);
-      st.audio.src = url; draw();
+      st.audio.src = url; syncNotesBox(); draw();
     } catch (e) { console.error("[audiowave] could not load/decode", name, e); }
   }
 
-  const tol = () => x2t(6);
-  const nearBoundary = (t) => { let bi = -1, bd = 1e9; st.boundaries.forEach((b, i) => { const d = Math.abs(b - t); if (d < tol() && d < bd) { bd = d; bi = i; } }); return bi; };
-  let dragIdx = -1;
-
-  canvas.addEventListener("mousedown", (e) => {
+  canvas.addEventListener("mousedown", (e) => {         // click = seek
     const r = canvas.getBoundingClientRect();
-    const t = x2t(((e.clientX - r.left) / r.width) * canvas.width);
-    if (e.button === 2 || e.shiftKey) { const bi = nearBoundary(t); if (bi >= 0) { st.boundaries.splice(bi, 1); serialize(); draw(); } e.preventDefault(); return; }
-    const bi = nearBoundary(t);
-    if (bi >= 0) { dragIdx = bi; }
-    else { st.boundaries.push(t); serialize(); draw(); }
-  });
-  canvas.addEventListener("mousemove", (e) => {
-    if (dragIdx < 0) return;
-    const r = canvas.getBoundingClientRect();
-    st.boundaries[dragIdx] = Math.max(0.01, Math.min(st.duration - 0.01, x2t(((e.clientX - r.left) / r.width) * canvas.width)));
+    st.audio.currentTime = Math.max(0, Math.min(st.duration, x2t(((e.clientX - r.left) / r.width) * canvas.width)));
     draw();
   });
-  window.addEventListener("mouseup", () => { if (dragIdx >= 0) { serialize(); dragIdx = -1; draw(); } });
-  canvas.addEventListener("dblclick", (e) => {
+  canvas.addEventListener("dblclick", (e) => {          // dblclick a chunk = edit its note
     const r = canvas.getBoundingClientRect();
     const t = x2t(((e.clientX - r.left) / r.width) * canvas.width);
-    const ss = starts(); let si = 0; for (let i = 0; i < ss.length; i++) if (t >= ss[i]) si = i;
-    const val = window.prompt(`Note for segment ${si + 1}:`, st.notes[si] || "");
-    if (val !== null) { st.notes[si] = val; serialize(); draw(); }
+    const starts = chunkStarts(); let i = 0; for (let k = 0; k < starts.length; k++) if (t >= starts[k]) i = k;
+    const { seg } = parseNotes(getW(node, "notes")?.value);
+    const val = window.prompt(`Note for subsegment ${i + 1}:`, seg[i + 1] || "");
+    if (val !== null) { setChunkNote(i + 1, val); draw(); }
   });
-  canvas.addEventListener("contextmenu", (e) => e.preventDefault());
 
   const loop = () => { if (!st.playing) return; draw(); requestAnimationFrame(loop); };
   playBtn.onclick = () => {
@@ -122,7 +141,6 @@ function setupWave(node) {
     else { st.audio.pause(); st.playing = false; playBtn.textContent = "▶ play"; }
   };
   st.audio.onended = () => { st.playing = false; playBtn.textContent = "▶ play"; draw(); };
-  clearBtn.onclick = () => { st.boundaries = []; st.notes = {}; serialize(); draw(); };
 
   upBtn.onclick = () => {
     const inp = document.createElement("input"); inp.type = "file"; inp.accept = "audio/*";
@@ -132,19 +150,19 @@ function setupWave(node) {
       try {
         const res = await api.fetchApi("/upload/image", { method: "POST", body: fd });
         const j = await res.json(); const name = j.name || f.name;
-        const w = getW(node, "audio");
-        if (w) { if (!w.options.values.includes(name)) w.options.values.push(name); w.value = name; }
+        const w = getW(node, "audio"); if (w) { if (!w.options.values.includes(name)) w.options.values.push(name); w.value = name; }
         await loadFile(name);
       } catch (e) { console.error("[audiowave] upload failed (drop the file in ComfyUI/input instead)", e); }
     };
     inp.click();
   };
 
-  // react to the audio combo changing + initial load
-  const aw = getW(node, "audio");
-  if (aw) { const cb = aw.callback; aw.callback = function () { const r = cb ? cb.apply(this, arguments) : undefined; loadFile(aw.value); return r; }; if (aw.value) loadFile(aw.value); }
-  // restore boundaries/notes from a reloaded workflow
-  try { const sj = getW(node, "segments_json"); if (sj && sj.value) { const arr = JSON.parse(sj.value); st.boundaries = arr.slice(1).map((s) => s.start_s); arr.forEach((s, i) => { if (s.note) st.notes[i] = s.note; }); } } catch (e) { /* ignore */ }
+  // re-grid + re-fill notes when the audio, fps or subsegment_frames change
+  const hook = (name, fn) => { const w = getW(node, name); if (w) { const cb = w.callback; w.callback = function () { const r = cb ? cb.apply(this, arguments) : undefined; fn(); return r; }; } };
+  hook("audio", () => loadFile(getW(node, "audio")?.value));
+  hook("fps", () => { syncNotesBox(); draw(); });
+  hook("subsegment_frames", () => { syncNotesBox(); draw(); });
+  const aw = getW(node, "audio"); if (aw?.value) loadFile(aw.value);
   draw();
 }
 

workflow/workflow_ltx_audio.json

@@ -113,7 +113,8 @@
       "widgets_values": [
         "audio.wav",
         24,
-        6,
+        721,
+        0,
         "",
         "[]"
       ]

workflow/workflow_ltx_audio_api.json

@@ -13,7 +13,8 @@
     "inputs": {
       "audio": "audio.wav",
       "fps": 24,
-      "max_segments": 6,
+      "subsegment_frames": 721,
+      "segment_select": 0,
       "notes": "",
       "segments_json": "[]"
     },