diff --git a/README.md b/README.md index 9b9895d..2ba123a 100644 --- a/README.md +++ b/README.md @@ -111,17 +111,18 @@ 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, -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). +An in-node **mirrored waveform** widget: **upload** an audio clip, **play** it, **click to seek**. -- **`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**. +- **`group_frames`** (default **721 @ 24fps ≈ 30.04s** = one LTX clip) is the **hard split** — + bold grid lines every `group_frames`. Segments never cross a group line, so the last segment + in a group ends exactly on the boundary (e.g. frame 721). +- **Segments live inside groups.** By default each group is one segment; **double-click the + waveform to add a finer split** inside a group (shift-click a split to remove it). The `notes` + box **auto-fills with one `segN:` line per segment** — type your motion note after each. +- **`segment_select`** — `""`/`0` = whole clip; `"N"` = only segment N; **`"A-B"` = segments A..B**. + The `waveform_image`, the `audio` output, and the summary are all **cropped to that range**, so + you can generate/skip beats a range at a time (e.g. `6-12`). Labels sit along the top, per-segment + notes along the bottom; the selected range is shaded. Outputs `waveform_image`, `audio_summary`, `audio`. Needs `torchaudio`/`soundfile`/`librosa` to load the file (torchaudio usually ships with torch). diff --git a/nodes/audio_guide.py b/nodes/audio_guide.py index 1aa6876..eefa608 100644 --- a/nodes/audio_guide.py +++ b/nodes/audio_guide.py @@ -131,43 +131,50 @@ def _segments(rms_n, times, duration, fps, max_segments, beats): 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.""" + fps=None, group_frames=0, frames_marker=0, window=None, sel=None): + """Render a mirrored waveform + the 721-frame group grid (bold) + segment lines (thin), + labels along the top and per-segment notes along the bottom. window=(t0,t1) crops to a + range; sel=(a,b) shades segments a..b.""" W, H = 1024, 256 img = Image.new("RGB", (W, H), (18, 18, 22)) d = ImageDraw.Draw(img) t0, t1 = window if window else (0.0, duration) span = max(t1 - t0, 1e-6) + mid = H // 2 def X(t): return int(max(0, min(W - 1, (t - t0) / span * (W - 1)))) - 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 + if sel and not window: # shade the selected segment range + chosen = [s for s in segs if sel[0] <= s["segment"] <= sel[1]] + if chosen: + xa = X(chosen[0]["start_s"]) + xb = X(chosen[-1]["start_s"] + chosen[-1]["duration_s"]) + d.rectangle([xa, 0, xb, H], fill=(38, 54, 82)) + amp = H * 0.44 # mirrored waveform around the centre line + for i in range(len(rms_n)): + if t0 <= times[i] <= t1: + x = X(times[i]); h = int(rms_n[i] * amp) + d.line([(x, mid - h), (x, mid + h)], fill=(60, 140, 220)) + if fps and group_frames: # bold 721-frame group grid + gstep = group_frames / 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)) + while k * gstep < duration + 1e-6: + gt = k * gstep + if t0 <= gt <= t1: + d.line([(X(gt), 0), (X(gt), H)], fill=(235, 235, 242), width=2) k += 1 - for b in beats: # beat markers (orange) + for b in beats: # beat ticks (faint, centre band) if t0 <= b <= t1: - d.line([(X(b), 0), (X(b), H)], fill=(230, 110, 60), width=1) - for s in segs: # boundaries + label + time + note + d.line([(X(b), mid - 4), (X(b), mid + 4)], fill=(150, 90, 60), width=1) + for s in segs: # segment line + label top + note bottom 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 + 4, 18), s["note"][:30], fill=(255, 210, 110)) + d.line([(x, 0), (x, H)], fill=(120, 190, 150), width=1) + d.text((x + 3, 3), f"S{s['segment']} {s['start_s']}s/{s['frames']}f", fill=(240, 240, 240)) + if s.get("note"): + d.text((x + 3, H - 13), s["note"][:28], fill=(255, 210, 110)) arr = np.asarray(img, dtype=np.float32) / 255.0 return torch.from_numpy(arr)[None, ...] # [1, H, W, 3] diff --git a/nodes/audio_wave_segments.py b/nodes/audio_wave_segments.py index bd973f4..629cc5e 100644 --- a/nodes/audio_wave_segments.py +++ b/nodes/audio_wave_segments.py @@ -15,12 +15,13 @@ from __future__ import annotations import json import os +import re import numpy as np import torch from .audio_guide import ( - _rms_envelope, _tempo_beats, _snap8, _segments, _render, _summary, _attach_notes, + _rms_envelope, _tempo_beats, _snap8, _render, _summary, _attach_notes, ) @@ -88,31 +89,52 @@ 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))) +def _parse_range(sel, n): + """'' / '0' / 'all' -> (0,0) = whole clip; 'N' -> (N,N); 'A-B' -> (A,B). 1-based, clamped.""" + sel = (sel or "").strip() + if not sel or sel in ("0", "all"): + return (0, 0) + m = re.match(r"^(\d+)\s*[-–:]\s*(\d+)$", sel) + if m: + a, b = int(m.group(1)), int(m.group(2)) + else: + try: + a = b = int(sel) + except ValueError: + return (0, 0) + if a > b: + a, b = b, a + return (max(1, min(a, n)), max(1, min(b, n))) + + +def _segments_grouped(rms_n, times, duration, fps, group_frames, user_starts, beats): + """Segments = user split points UNION the fixed group boundaries (every group_frames + frames). The group lines are hard splits, so a segment never crosses one — the last + segment in a group ends exactly on the group boundary (e.g. frame 721).""" + gstep = max(group_frames, 1) / max(fps, 1) + gbounds, k = [], 1 + while k * gstep < duration - 1e-6: + gbounds.append(round(k * gstep, 3)); k += 1 + allb = sorted({round(b, 3) for b in list(user_starts) + gbounds if 0 < b < duration - 1e-6}) + starts = [0.0] + allb stages = ["establish", "build", "peak", "settle"] segs, bounds = [], [0.0] - for i in range(n): - t0 = i * step - t1 = min(duration, (i + 1) * step) + for i, t0 in enumerate(starts): + t1 = starts[i + 1] if i + 1 < len(starts) else duration 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], + "segment": i + 1, "group": 1 + sum(1 for gb in gbounds if t0 >= gb - 0.02), + "start_s": round(t0, 2), "duration_s": dur, "frames": _snap8(round(dur * fps)), + "energy": "high" if e > 0.66 else ("medium" if e > 0.33 else "low"), + "energy_val": round(e, 3), "peak_s": round(peak_t, 2), + "stage_hint": stages[i] if i < len(stages) else ("peak" if e > 0.6 else "settle"), + "note": "", "beats_in": [round(b, 2) for b in beats if t0 <= b < t1], }) - return segs, np.array(bounds) + return segs, np.array(bounds), gbounds class AudioWaveSegments: @@ -128,29 +150,29 @@ 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}), - # 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}), + # Hard split every this many frames (721@24fps ~= 30.04s = one LTX clip). + # Segments live inside a group and never cross a group line. + "group_frames": ("INT", {"default": 721, "min": 1, "max": 100000}), + # "" / "0" = whole clip. "N" = only segment N. "A-B" = segments A..B — + # crops image + audio + summary to that range (for generating/skipping beats). + "segment_select": ("STRING", {"default": ""}), "notes": ("STRING", {"default": "", "multiline": True}), - # Machine field: per-chunk notes/boundaries from the JS widget (hidden). + # Machine field: the JS widget writes only USER split points here (hidden). "segments_json": ("STRING", {"default": "[]"}), }, } @classmethod - def IS_CHANGED(cls, audio, fps, subsegment_frames, segment_select, notes, segments_json): + def IS_CHANGED(cls, audio, fps, group_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}|{subsegment_frames}|{segment_select}|{notes}|{segments_json}|{mt}" + return f"{audio}|{fps}|{group_frames}|{segment_select}|{notes}|{segments_json}|{mt}" - def run(self, audio, fps, subsegment_frames, segment_select, notes, segments_json): + def run(self, audio, fps, group_frames, segment_select, notes, segments_json): try: import folder_paths path = folder_paths.get_annotated_filepath(audio) @@ -166,44 +188,44 @@ class AudioWaveSegments: rms_n, times = _rms_envelope(y, sr) bpm, beats = _tempo_beats(y, sr) - 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: - for seg in (json.loads(segments_json) if segments_json.strip() else []): - starts.append(float(seg.get("start_s", 0.0))) - except Exception as e: - print(f"[AudioWaveSegments] bad segments_json ({e})") - segs, bounds = _segments_from_boundaries(rms_n, times, duration, fps, starts, beats) - - # per-chunk notes from segments_json (JS) then the notes-box (segN: syntax) on top. + user_starts, seg_notes = [], {} # user split points from the JS widget 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) + user_starts.append(float(seg.get("start_s", 0.0))) + if seg.get("note"): + seg_notes[round(float(seg.get("start_s", 0.0)), 2)] = str(seg["note"]) + except Exception as e: + print(f"[AudioWaveSegments] bad segments_json ({e})") - window, render_segs = None, segs + if group_frames < 1: + group_frames = 721 + segs, bounds, gbounds = _segments_grouped(rms_n, times, duration, fps, group_frames, user_starts, beats) + for s in segs: # notes attached by start time + if s["start_s"] in seg_notes: + s["note"] = seg_notes[s["start_s"]] + global_notes = _attach_notes(segs, notes) # notes-box (segN:) merges on top + + a, b = _parse_range(segment_select, len(segs)) + window, sel = None, None 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)) + if a >= 1 and segs: # crop to segment range A..B + s0 = segs[a - 1]["start_s"] + s1 = min(duration, segs[b - 1]["start_s"] + segs[b - 1]["duration_s"]) + i0, i1 = int(s0 * sr), int(s1 * sr) + if i1 > i0: + audio_out = {"waveform": wav[:, i0:i1].unsqueeze(0), "sample_rate": sr} + window, sel = (s0, s1), (a, b) + total = sum(s["frames"] for s in segs[a - 1:b]) + g0, g1 = segs[a - 1]["group"], segs[b - 1]["group"] + gtxt = f"group {g0}" + ("" if g0 == g1 else f"–{g1}") + summary = (f"SELECTED SEGMENTS {a}-{b} of {len(segs)} ({gtxt}), {total} frames total " + f"— generate these beats.\n" + + _summary(duration, sr, bpm, beats, segs[a - 1:b], global_notes)) else: summary = _summary(duration, sr, bpm, beats, segs, global_notes) - image = _render(rms_n, times, duration, beats, bounds, render_segs, - fps=fps, frames_marker=subsegment_frames, window=window) + image = _render(rms_n, times, duration, beats, bounds, segs, + fps=fps, group_frames=group_frames, window=window, sel=sel) return (image, summary, audio_out) diff --git a/web/audio_wave.js b/web/audio_wave.js index 4146afa..1918263 100644 --- a/web/audio_wave.js +++ b/web/audio_wave.js @@ -1,6 +1,6 @@ -// 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. +// Audio Wave + Segments — mirrored waveform + playback, a bold 721-frame group grid, +// optional fine user splits inside groups, per-segment notes auto-filled into the `notes` +// box, playhead + click-to-seek. Open the console for [audiowave] logs on trouble. import { app } from "../../scripts/app.js"; import { api } from "../../scripts/api.js"; @@ -12,8 +12,7 @@ function computePeaks(buf, n) { const peaks = new Float32Array(n); let max = 1e-6; for (let i = 0; i < n; i++) { - let m = 0; - const s = i * block; + let m = 0; 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; } @@ -21,39 +20,26 @@ function computePeaks(buf, n) { return peaks; } -// Keep a ComfyUI DOM widget at full node width (fixes the collapse-to-half-width bug on -// selection / re-layout). Portable helper; call once after addDOMWidget. +// Keep a ComfyUI DOM widget at full node width (fixes collapse-to-half on select). function keepDomWidgetFullWidth(node, container) { const GRID_SEL = '[data-testid="node-widgets"], .lg-node-widgets'; const ROW_SEL = '[data-testid="node-widget"], .lg-node-widget'; const MAX_MARGIN = 40; let enforcing = false, marginLogical = Infinity, gridObserver = null; - function refFromDom() { - const grid = container.closest(GRID_SEL); - if (!grid) return 0; + const grid = container.closest(GRID_SEL); if (!grid) return 0; let w = 0; - for (const row of Array.from(grid.querySelectorAll(ROW_SEL))) { - if (row.contains(container)) continue; - const c = row.lastElementChild; - if (c && c.clientWidth > w) w = c.clientWidth; - } - if (!w && grid.clientWidth > 0) { - const dot = grid.querySelector(`${ROW_SEL.split(",")[0]} > :first-child`); - w = grid.clientWidth - (dot?.offsetWidth ?? 0); - } + for (const row of Array.from(grid.querySelectorAll(ROW_SEL))) { if (row.contains(container)) continue; const c = row.lastElementChild; if (c && c.clientWidth > w) w = c.clientWidth; } + if (!w && grid.clientWidth > 0) { const dot = grid.querySelector(`${ROW_SEL.split(",")[0]} > :first-child`); w = grid.clientWidth - (dot?.offsetWidth ?? 0); } return w; } function refFromNodeSize(cw) { - const nodeW = node.size?.[0] ?? 0; - if (nodeW <= 0) return 0; + const nodeW = node.size?.[0] ?? 0; if (nodeW <= 0) return 0; if (cw > 0) { const m = nodeW - cw; if (m >= 0 && m < marginLogical) marginLogical = Math.min(m, MAX_MARGIN); } - const margin = Number.isFinite(marginLogical) ? marginLogical : MAX_MARGIN / 2; - return Math.round(nodeW - margin); + return Math.round(nodeW - (Number.isFinite(marginLogical) ? marginLogical : MAX_MARGIN / 2)); } function reference(cw) { - let w = refFromDom(); - if (!w) w = refFromNodeSize(cw); + let w = refFromDom(); if (!w) w = refFromNodeSize(cw); if (!w) for (const wd of node.widgets ?? []) { const el = wd.inputEl || wd.element; if (el && el !== container && el.offsetWidth > w) w = el.offsetWidth; } if (!w && container.parentElement) w = container.parentElement.clientWidth; return w; @@ -61,19 +47,16 @@ function keepDomWidgetFullWidth(node, container) { function enforce() { if (enforcing) return; if (!gridObserver) { const grid = container.closest(GRID_SEL); if (grid) { gridObserver = new ResizeObserver(enforce); gridObserver.observe(grid); } } - const cw = container.clientWidth; - const ref = reference(cw); + const cw = container.clientWidth, ref = reference(cw); if (ref > 0 && Math.abs(cw - ref) > 2) { enforcing = true; container.style.width = ref + "px"; requestAnimationFrame(() => { enforcing = false; }); } } - const ro = new ResizeObserver(enforce); - ro.observe(container); - const origOnResize = node.onResize; - node.onResize = function (size) { origOnResize?.call(this, size); enforce(); }; + const ro = new ResizeObserver(enforce); ro.observe(container); + const origOnResize = node.onResize; node.onResize = function (s) { origOnResize?.call(this, s); enforce(); }; return () => { ro.disconnect(); gridObserver?.disconnect(); }; } function setupWave(node) { - const st = { duration: 0, peaks: null, audio: new Audio(), playing: false }; + const st = { duration: 0, peaks: null, splits: [], audio: new Audio(), playing: false }; node._wave = st; const wrap = document.createElement("div"); @@ -83,85 +66,92 @@ function setupWave(node) { 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"), 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"; + const hint = document.createElement("span"); hint.textContent = "click=seek · dblclick=add split · shift-click a split=remove"; bar.append(playBtn, upBtn, readout, hint); const canvas = document.createElement("canvas"); - canvas.width = 640; canvas.height = 160; - canvas.style.cssText = "width:100%;height:160px;background:#141418;border-radius:4px;cursor:pointer;"; + canvas.width = 900; canvas.height = 170; + canvas.style.cssText = "width:100%;height:170px;background:#141418;border-radius:4px;cursor:pointer;"; wrap.append(bar, canvas); - // Neutral type (NOT "preview" — that triggers aspect-ratio sizing that collapses width), - // with an explicit height, then pin the width against the collapse-on-select bug. - node.addDOMWidget("wave", "wave", wrap, { serialize: false, getMinHeight: () => 205 }); + node.addDOMWidget("wave", "wave", wrap, { serialize: false, getMinHeight: () => 225 }); const cleanupWidth = keepDomWidgetFullWidth(node, wrap); const origRemoved = node.onRemoved; node.onRemoved = function () { try { cleanupWidth(); } catch (e) { /* ignore */ } return origRemoved?.apply(this, arguments); }; const ctx = canvas.getContext("2d"); - const sjw = getW(node, "segments_json"); // machine field — hide it - if (sjw) { sjw.hidden = true; sjw.computeSize = () => [0, -4]; } + const sjw = getW(node, "segments_json"); 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 groupFrames = () => Math.max(1, Number(getW(node, "group_frames")?.value) || 721); // guard: never 0 const t2x = (t) => (st.duration ? (t / st.duration) * canvas.width : 0); const x2t = (x) => (st.duration ? (x / canvas.width) * st.duration : 0); - 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 groupBounds() { + const step = groupFrames() / fps(); const arr = []; + if (!st.duration || step <= 0) return arr; + for (let t = step; t < st.duration - 1e-6; t += step) arr.push(t); + return arr; + } + function segStarts() { // union of 0 + user splits + group lines + const set = new Set([0]); + for (const s of st.splits) if (s > 0 && s < st.duration) set.add(Math.round(s * 100) / 100); + for (const g of groupBounds()) set.add(Math.round(g * 100) / 100); + return Array.from(set).sort((a, b) => a - b); } + function serialize() { // write ONLY user splits (small + stable) + const w = getW(node, "segments_json"); if (!w) return; + w.value = JSON.stringify(st.splits.map((s) => ({ start_s: Math.round(s * 100) / 100 }))); + w.callback?.(w.value); + } 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); - }); + (txt || "").split("\n").forEach((l) => { const m = l.match(/^\s*(?:seg(?:ment)?|s)?\s*(\d+)\s*[:)]\s*(.*)$/i); if (m) seg[parseInt(m[1])] = m[2]; else if (l.trim()) glob.push(l); }); return { seg, glob }; } - - function syncNotesBox() { // one "segN:" line per chunk, preserve notes + function syncNotesBox() { // one segN: line per segment, preserve notes const nw = getW(node, "notes"); if (!nw) return; const { seg, glob } = parseNotes(nw.value); - const n = chunkStarts().length; - const lines = []; + const n = segStarts().length; const lines = []; for (let i = 1; i <= n; i++) lines.push(`seg${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}: ${seg[k] !== undefined ? seg[k] : ""}`); - nw.value = lines.concat(glob).join("\n"); nw.callback?.(nw.value); + function selRange() { + const v = (getW(node, "segment_select")?.value || "").trim(); + const m = v.match(/^(\d+)\s*[-:]\s*(\d+)$/); if (m) return [+m[1], +m[2]].sort((a, b) => a - b); + if (/^\d+$/.test(v) && v !== "0") return [+v, +v]; + return null; } function draw() { ctx.fillStyle = "#141418"; ctx.fillRect(0, 0, canvas.width, canvas.height); - if (st.peaks) { - ctx.fillStyle = "#3c8cdc"; - const n = st.peaks.length, bw = canvas.width / n; - for (let i = 0; i < n; i++) { const h = st.peaks[i] * (canvas.height - 30); ctx.fillRect(i * bw, canvas.height - h, Math.max(1, bw), h); } + const H = canvas.height, mid = H / 2; + const starts = segStarts(); + const sel = selRange(); + if (sel) { // shade selected range + const a = starts[sel[0] - 1], b = sel[1] < starts.length ? starts[sel[1]] : st.duration; + if (a !== undefined) { ctx.fillStyle = "#26364f"; ctx.fillRect(t2x(a), 0, t2x(b) - t2x(a), H); } } - ctx.font = "10px monospace"; + if (st.peaks) { // mirrored waveform around centre + ctx.strokeStyle = "#3c8cdc"; ctx.beginPath(); + const n = st.peaks.length; + for (let i = 0; i < n; i++) { const x = (i / n) * canvas.width, h = st.peaks[i] * (H * 0.44); ctx.moveTo(x, mid - h); ctx.lineTo(x, mid + h); } + ctx.stroke(); + } + ctx.strokeStyle = "#ececf2"; ctx.lineWidth = 2; // bold group grid + for (const g of groupBounds()) { const x = t2x(g); ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, H); ctx.stroke(); } + ctx.lineWidth = 1; const { seg } = parseNotes(getW(node, "notes")?.value); - const starts = chunkStarts(); - starts.forEach((s, i) => { + ctx.font = "10px monospace"; + starts.forEach((s, i) => { // segment line + label top + note bottom const x = t2x(s); - ctx.strokeStyle = "#7ec8a0"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, canvas.height); ctx.stroke(); + ctx.strokeStyle = "#7ec8a0"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, H); ctx.stroke(); ctx.fillStyle = "#fff"; ctx.fillText(`S${i + 1} ${s.toFixed(1)}s`, x + 3, 11); - if (seg[i + 1]) { ctx.fillStyle = "#ffd27a"; ctx.fillText(seg[i + 1].slice(0, 22), x + 3, 23); } + if (seg[i + 1]) { ctx.fillStyle = "#ffd27a"; ctx.fillText(seg[i + 1].slice(0, 22), x + 3, H - 4); } }); - if (st.playing || st.audio.currentTime) { - const x = t2x(st.audio.currentTime); - ctx.strokeStyle = "#ff5a3c"; ctx.beginPath(); ctx.moveTo(x, 0); ctx.lineTo(x, canvas.height); ctx.stroke(); - } + ctx.strokeStyle = "#ff5a3c"; const px = t2x(st.audio.currentTime || 0); // playhead + ctx.beginPath(); ctx.moveTo(px, 0); ctx.lineTo(px, H); ctx.stroke(); readout.textContent = `${(st.audio.currentTime || 0).toFixed(2)} / ${st.duration.toFixed(2)}s`; } @@ -177,18 +167,21 @@ function setupWave(node) { } catch (e) { console.error("[audiowave] could not load/decode", name, e); } } - canvas.addEventListener("mousedown", (e) => { // click = seek - const r = canvas.getBoundingClientRect(); - st.audio.currentTime = Math.max(0, Math.min(st.duration, x2t(((e.clientX - r.left) / r.width) * canvas.width))); - draw(); - }); - canvas.addEventListener("dblclick", (e) => { // dblclick a chunk = edit its note + const tol = () => x2t(6); + canvas.addEventListener("mousedown", (e) => { const r = canvas.getBoundingClientRect(); const t = x2t(((e.clientX - r.left) / r.width) * canvas.width); - 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(); } + if (e.shiftKey) { // shift-click = remove nearest user split + let bi = -1, bd = 1e9; st.splits.forEach((s, i) => { const dd = Math.abs(s - t); if (dd < tol() && dd < bd) { bd = dd; bi = i; } }); + if (bi >= 0) { st.splits.splice(bi, 1); serialize(); syncNotesBox(); draw(); } + return; + } + st.audio.currentTime = Math.max(0, Math.min(st.duration, t)); draw(); // click = seek + }); + canvas.addEventListener("dblclick", (e) => { // add a user split + const r = canvas.getBoundingClientRect(); + const t = x2t(((e.clientX - r.left) / r.width) * canvas.width); + if (t > 0.05 && t < st.duration - 0.05) { st.splits.push(t); serialize(); syncNotesBox(); draw(); } }); const loop = () => { if (!st.playing) return; draw(); requestAnimationFrame(loop); }; @@ -213,11 +206,12 @@ function setupWave(node) { inp.click(); }; - // 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(); }); + hook("group_frames", () => { syncNotesBox(); draw(); }); + hook("segment_select", () => draw()); + try { const sj = getW(node, "segments_json"); if (sj?.value) st.splits = (JSON.parse(sj.value) || []).map((s) => s.start_s).filter((s) => s > 0); } catch (e) { /* ignore */ } const aw = getW(node, "audio"); if (aw?.value) loadFile(aw.value); draw(); } diff --git a/workflow/workflow_ltx_audio.json b/workflow/workflow_ltx_audio.json index dda18cf..29ca70a 100644 --- a/workflow/workflow_ltx_audio.json +++ b/workflow/workflow_ltx_audio.json @@ -114,7 +114,7 @@ "audio.wav", 24, 721, - 0, + "", "", "[]" ] diff --git a/workflow/workflow_ltx_audio_api.json b/workflow/workflow_ltx_audio_api.json index 1ddad8b..65b8f9e 100644 --- a/workflow/workflow_ltx_audio_api.json +++ b/workflow/workflow_ltx_audio_api.json @@ -13,8 +13,8 @@ "inputs": { "audio": "audio.wav", "fps": 24, - "subsegment_frames": 721, - "segment_select": 0, + "group_frames": 721, + "segment_select": "", "notes": "", "segments_json": "[]" },