Edit feature: slider + napis edit + recut endpoint

User insight: 'treba je narediti da ko se reels naredijo da jih lahko
popravljamo... delamo na avtomatiko ampak lahk pa tudi popravljam'

Avto pipeline ostane (Soniox + Claude + render). Po render-u uporabnik
lahko klikne ✏️ Edit gumb in:

1. **Slider za clip start/end**:
   - Vidi 16:9 original video
   - Drag start/end slider z živim preview-om
   - Dolžina prikazana real-time
   - Min 5s, max 60s

2. **Edit napisov** (collapsed, opcijsko):
   - Klik na vrstico → input za popravek besedila
   - Original timestamp ostane, samo besedilo se posodobi
   - Uporabno za 'doline IZBOR' → 'doline IZPOD' tip popravkov

3. **Re-render**:
   - Backend POST /api/jobs/{id}/recut z {start, end, custom_segments}
   - Worker preskoči Soniox + Claude (custom_clip flag)
   - Re-uporabi cached transcript + analysis
   - Re-render samo: clip → reframe → subtitle → output
   - ~30s namesto 3-5 min

New endpoints:
- GET /api/source-video/{id} — 16:9 original za editor preview
- GET /api/transcript/{id} — segmenti + clip range za editor
- POST /api/jobs/{id}/recut — re-render z user timestampi

Worker change: če job ima custom_clip=True, preskoči auto_chorus
analizo in samo re-uporabi obstoječi clip_range iz analysis.json
(updated by recut endpoint).
This commit is contained in:
Sebastjan Artič 2026-04-30 10:26:25 +00:00
parent f99574daff
commit 7cb4302dcd
3 changed files with 382 additions and 5 deletions

View File

@ -349,18 +349,40 @@ def generate_srt_from_segments(segments, clip_start, clip_end, output_path):
if words_in_clip and len(words_in_clip) >= 2:
# Group besede v chunke z max trajanjem MAX_CHUNK_DURATION
# Pravila:
# 1. Vsak chunk traja max MAX_CHUNK_DURATION (2.5s)
# 2. NE začni novega chunka če bi prejšnji ostal kratek (<3 besede ali <12 znakov)
# 3. NE pusti zadnji chunk siroto z 1 kratko besedo (<3 znakov) — združi nazaj
MIN_CHUNK_WORDS = 3
MIN_CHUNK_CHARS = 12
chunks = []
current_chunk = [words_in_clip[0]]
for w in words_in_clip[1:]:
chunk_start_time = current_chunk[0]["start"]
chunk_dur_so_far = w["end"] - chunk_start_time
# Bi začeli nov chunk?
if chunk_dur_so_far > MAX_CHUNK_DURATION:
chunks.append(current_chunk)
current_chunk = [w]
# Preveri ali bi current_chunk ostal "siroto kratek"
current_text = " ".join(c["text"] for c in current_chunk)
too_short = (len(current_chunk) < MIN_CHUNK_WORDS
or len(current_text) < MIN_CHUNK_CHARS)
if too_short:
# Daj še to besedo zraven (raje malo predolg chunk kot siroto kratek)
current_chunk.append(w)
else:
chunks.append(current_chunk)
current_chunk = [w]
else:
current_chunk.append(w)
if current_chunk:
chunks.append(current_chunk)
# Če zadnji chunk ima samo 1-2 kratki besedi, združi z zadnjim chunkom
last_text = " ".join(c["text"] for c in current_chunk)
if (chunks and
(len(current_chunk) < 2 or len(last_text) < 5)):
chunks[-1].extend(current_chunk)
else:
chunks.append(current_chunk)
# Generiraj SRT iz chunks (z dejanskimi word timestampi)
for chunk in chunks:
@ -490,7 +512,50 @@ def process_job(job_id):
print(f"⚠️ ACR error: {e}", flush=True)
# ── 2. Smart analysis (če auto_chorus) ──────────────────────────
if job.get("auto_chorus"):
# ÄŒe je custom_clip=True (user-edit mode), preskoÄi analizo
# in samo posodobi start/duration iz obstojeÄega clip_range v analysis.json
if job.get("custom_clip"):
update_job(job_id, current_step="User-edit recut (preskočim analizo)")
analysis_path = OUTPUT_DIR / f"{job_id}.analysis.json"
srt_from_claude = None
if analysis_path.exists():
try:
with open(analysis_path, "r", encoding="utf-8") as f:
analysis = json.load(f)
cr = analysis["clip_range"]
fade = analysis.get("fade", {"fade_in": 0.05, "fade_out": 0.5})
# Generiraj nov SRT iz obstoječega transkripta na novem range-u
if analysis.get("transcript", {}).get("segments") and not job.get("no_subs"):
srt_path_out = OUTPUT_DIR / f"{job_id}.subtitles.srt"
try:
# Če imamo custom_subtitle_segments (user popravil napise)
segs_to_use = analysis.get("custom_subtitle_segments") or analysis["transcript"]["segments"]
generate_srt_from_segments(
segs_to_use,
cr["start"], cr["end"],
srt_path_out,
)
srt_from_claude = str(srt_path_out)
except Exception as e:
print(f"⚠️ Recut SRT generation failed: {e}", flush=True)
update_job(
job_id,
start=cr["start"],
duration=cr["duration"],
fade_in=fade.get("fade_in", 0.05),
fade_out=fade.get("fade_out", 0.5),
claude_srt_path=srt_from_claude,
)
job = load_job(job_id)
except Exception as e:
update_job(job_id, status="failed", error=f"Recut analysis read: {e}")
return
else:
update_job(job_id, status="failed", error="Analysis manjka za recut")
return
elif job.get("auto_chorus"):
update_job(job_id, current_step="Analiza pesmi (transkript + energija)")
analysis_path = OUTPUT_DIR / f"{job_id}.analysis.json"
cmd = [
@ -1164,3 +1229,130 @@ async def delete_job(job_id: str, user: str = Depends(check_auth)):
Path(p).unlink(missing_ok=True)
job_path(job_id).unlink(missing_ok=True)
return {"deleted": job_id}
# ─── EDIT FEATURE ────────────────────────────────────────────────
@app.get("/api/source-video/{job_id}")
async def source_video(job_id: str, user: str = Depends(check_auth)):
"""Vrne 16:9 original video za predogled v editorju."""
job = load_job(job_id)
if not job:
raise HTTPException(404, "Ne obstaja")
src = job.get("input_path")
if not src or not Path(src).exists():
raise HTTPException(404, "Original video ne obstaja")
return FileResponse(
path=src,
media_type="video/mp4",
filename=f"source_{job_id}.mp4",
headers={"Accept-Ranges": "bytes"},
)
@app.get("/api/transcript/{job_id}")
async def get_transcript(job_id: str, user: str = Depends(check_auth)):
"""Vrne transkript (segmente + words) za editor."""
job = load_job(job_id)
if not job:
raise HTTPException(404, "Ne obstaja")
analysis_path = OUTPUTS_DIR / f"{job_id}.analysis.json"
if not analysis_path.exists():
raise HTTPException(404, "Analysis ne obstaja")
try:
analysis = json.loads(analysis_path.read_text())
transcript = analysis.get("transcript", {})
clip_range = analysis.get("clip_range", {})
return {
"segments": transcript.get("segments", []),
"language": transcript.get("language", "?"),
"clip_range": clip_range,
"video_duration": job.get("video_duration", 0),
}
except Exception as e:
raise HTTPException(500, f"Napaka pri branju: {e}")
class RecutRequest(BaseModel):
start: float
end: float
custom_segments: Optional[list] = None # [{start, end, text}] za override napisov
no_subs: Optional[bool] = None
@app.post("/api/jobs/{job_id}/recut")
async def recut_job(job_id: str, payload: RecutRequest, user: str = Depends(check_auth)):
"""Re-rendar reel z user-defined timestampi (in opcijsko popravljenimi napisi).
Veliko hitrejše od full pipeline-a:
- Brez Soniox transkripta (re-uporabi shranjenega)
- Brez Claude analize (uporabnik je odločil)
- Samo: clip reframe subtitle output
"""
job = load_job(job_id)
if not job:
raise HTTPException(404, "Ne obstaja")
src = job.get("input_path")
if not src or not Path(src).exists():
raise HTTPException(400, "Original video manjka")
if payload.end <= payload.start:
raise HTTPException(400, "End mora biti večji od start")
if payload.end - payload.start < 5:
raise HTTPException(400, "Trajanje vsaj 5s")
if payload.end - payload.start > 60:
raise HTTPException(400, "Trajanje največ 60s")
duration = payload.end - payload.start
no_subs = payload.no_subs if payload.no_subs is not None else job.get("no_subs", False)
# Naloži obstoječi analysis
analysis_path = OUTPUTS_DIR / f"{job_id}.analysis.json"
if not analysis_path.exists():
raise HTTPException(500, "Analysis manjka — re-uplad pesmi")
analysis = json.loads(analysis_path.read_text())
# Posodobi clip range
analysis["clip_range"] = {
"start": payload.start,
"end": payload.end,
"duration": duration,
"reason": f"USER EDIT: ročno popravljen clip {payload.start:.1f}-{payload.end:.1f}s",
"source": "user_edit",
}
# Če uporabnik popravil napise, override segmenti
if payload.custom_segments:
# Ohrani originalni transcript ampak shrani custom segments za SRT
analysis["custom_subtitle_segments"] = payload.custom_segments
analysis_path.write_text(json.dumps(analysis, indent=2, ensure_ascii=False))
# Re-queue job v processing (worker ga bo obdelal)
update_job(
job_id,
status="queued",
no_subs=no_subs,
custom_clip=True, # flag da preskoči Soniox + Claude
current_step="V vrsti za recut",
error=None,
chorus_error=None,
)
# Briši stari output
out_mp4 = OUTPUTS_DIR / f"{job_id}.mp4"
if out_mp4.exists():
out_mp4.unlink()
for ext in ("srt", "ass"):
p = OUTPUTS_DIR / f"{job_id}.subtitles.{ext}"
if p.exists():
p.unlink()
return {
"status": "queued",
"job_id": job_id,
"start": payload.start,
"end": payload.end,
"duration": duration,
}

View File

@ -1453,7 +1453,7 @@ PROSIM:
| Pesem | Refren (kar ide v reel) | Občutek |
|---|---|---|
| **BRAJDE** (FIRBCI) | "Ajmo Janezi! Pejd' greva, ajde, na traktorju od Majde, noben naju ne najde, ko peljem te v brajde. Da v senci hladni vžgem te po tazadn'i, ko s tebe trgam cunje, Gucci, Louis Vuitton!" | 2x ponovitev z "Ajmo Janezi" intro = 28s |
| **CVETELE SO MALINE** (Avsenik) | "Naj veter zdaj ponese moje sanje čez del neba, tja, kjer je ona doma. Zašepetaj ji bog tja pod kostanje, da sanjam še, kar sva nekoč oba." | PRVI nastop, ~15s, **NE OUTRO** |
| **CVETELE SO MALINE** (Avsenik) | "Naj veter zdaj ponese moje sanje čez del neba, tja, kjer je ona doma. Zašepetaj ji bog tja pod kostanje, da sanjam še, kar sva nekoč oba." | PRVI nastop, ~15s, **VKLJUČI vse do konca "oba" (zadnja beseda IZPETO)** če je instrumental break za besedo "oba", clip še VEDNO mora vključiti TO BESEDO + 0.5s po njej (NE odreži pred "oba"!) |
| **ENA BOLHA ZA POMOČ** (Avsenik) | "Kam se ti tako mudi, punca, voda ne gori, daj počakaj še malo! Mlada ledi, k' si in frej, dej se z fanti mal' igrej, živeti je treba zdej-ej-ej. Hvala, lupčka, lahko noč, ena bolha za pomoč in užiVAJ življenje. Dobro jutro, dober dan, saj ljubezen pač ni plan, in užiVAJ samski stan." | Dolg refren, ~30s |
| **ŽENA ME TEPE** (Avsenik) | "Žena me tepe, mi prazni žepe, da vidi, kje in s kom sem bil. Jaz pa razlagam, da ne vem, da nisem kriv, da šel naravnost sem domov, pa se izgubil. Žena me tepe, obrača žepe, preverja, kje in s kom sem pil. Saj bi ji rekel, naj se že enkrat ustav', pa vem, da ima po svoje prav." | 2 vrstici refrena ~30s |
| **STISN SE K MEN** (Stil) | "Stisni se k meni, tak ni noben. Ko bova stara, bova spala, do takrat pa boogie-woogie žgala, ne? Ne hecam se, ko se zaljubim, ljubim, kot da zadnjič je." | 2x ponovitev = ~30s |

View File

@ -923,6 +923,7 @@
if (job.status === "done") {
actions.push(`<button class="small" data-action="download" data-id="${job.id}">⬇ Download</button>`);
actions.push(`<button class="small ghost" data-action="preview" data-id="${job.id}">▶ Preview</button>`);
actions.push(`<button class="small ghost" data-action="edit" data-id="${job.id}">✏️ Edit</button>`);
}
actions.push(`<button class="small ghost" data-action="delete" data-id="${job.id}"></button>`);
@ -971,6 +972,8 @@
window.open(`/api/download/${id}`);
} else if (action === "preview") {
previewJob(id, title);
} else if (action === "edit") {
openEditModal(id, title);
} else if (action === "delete") {
deleteJob(id);
}
@ -982,6 +985,188 @@
refreshJobs();
}
// ─── EDIT MODAL ─────────────────────────────────────
async function openEditModal(jobId, title) {
// Fetch transcript + clip range
let data;
try {
const res = await fetch(`/api/transcript/${jobId}`);
if (!res.ok) {
alert("Napaka: ne morem naložiti transkripta");
return;
}
data = await res.json();
} catch (e) {
alert("Napaka: " + e.message);
return;
}
const startInit = data.clip_range?.start || 0;
const endInit = data.clip_range?.end || (startInit + 30);
const videoDuration = data.video_duration || endInit + 60;
const segments = data.segments || [];
const overlay = document.createElement("div");
overlay.className = "modal-overlay";
overlay.innerHTML = `
<div class="modal-content edit-modal" onclick="event.stopPropagation()" style="max-width:900px;">
<button class="modal-close" title="Zapri (ESC)">×</button>
<div class="modal-title" style="margin-bottom:12px;">✏️ Edit: ${escapeHtml(title)}</div>
<video id="edit-video" src="/api/source-video/${jobId}" controls preload="metadata" style="width:100%; max-height:50vh; background:#000;"></video>
<div style="display:grid; gap:14px; margin-top:14px;">
<div>
<label style="display:flex; justify-content:space-between; font-size:13px;">
<span>Začetek</span>
<span><b id="edit-start-val">${startInit.toFixed(1)}s</b> · trajanje: <b id="edit-duration">${(endInit-startInit).toFixed(1)}s</b></span>
</label>
<input type="range" id="edit-start" min="0" max="${videoDuration.toFixed(1)}" step="0.1" value="${startInit.toFixed(1)}" style="width:100%;">
</div>
<div>
<label style="display:flex; justify-content:space-between; font-size:13px;">
<span>Konec</span>
<span id="edit-end-val">${endInit.toFixed(1)}s</span>
</label>
<input type="range" id="edit-end" min="0" max="${videoDuration.toFixed(1)}" step="0.1" value="${endInit.toFixed(1)}" style="width:100%;">
</div>
<div style="display:flex; gap:8px;">
<button class="small" onclick="seekEditVideo('start')">▶ Predvajaj od začetka</button>
<button class="small ghost" onclick="seekEditVideo('end')">↪ Skoči na konec</button>
</div>
<details style="margin-top:8px;">
<summary style="cursor:pointer; font-size:13px; color:var(--muted);">📝 Edit napise (kliknite vrstico za popravek)</summary>
<div id="edit-segments" style="max-height:30vh; overflow:auto; margin-top:10px; padding:10px; background:rgba(255,255,255,0.03); border-radius:6px;">
${segments.filter(s => s.start < endInit && s.end > startInit).map((s, i) => `
<div class="seg-row" data-idx="${i}" style="margin-bottom:6px; padding:6px; background:rgba(255,255,255,0.04); border-radius:4px;">
<span style="font-size:11px; color:var(--muted); margin-right:8px;">[${s.start.toFixed(1)}s]</span>
<input type="text" data-orig="${escapeHtml(s.text || '')}" data-start="${s.start}" data-end="${s.end}" value="${escapeHtml(s.text || '').replace(/\\n/g, ' ').trim()}" style="width:calc(100% - 80px); padding:4px 8px; font-size:13px; background:rgba(0,0,0,0.3); border:1px solid rgba(255,255,255,0.1); border-radius:3px; color:#fff;">
</div>
`).join("")}
</div>
</details>
</div>
<div class="modal-actions" style="margin-top:18px;">
<button class="primary" id="edit-save-btn">✅ Shrani in re-render</button>
<button onclick="closeModal()">Prekliči</button>
</div>
<div id="edit-status" style="margin-top:10px; font-size:12px; color:var(--muted);"></div>
</div>
`;
document.body.appendChild(overlay);
document.body.style.overflow = "hidden";
// ESC key
const escHandler = (e) => {
if (e.key === "Escape") {
closeModal();
document.removeEventListener("keydown", escHandler);
}
};
document.addEventListener("keydown", escHandler);
overlay.querySelector(".modal-close").addEventListener("click", closeModal);
// Slider handlers
const video = document.getElementById("edit-video");
const startSlider = document.getElementById("edit-start");
const endSlider = document.getElementById("edit-end");
const startVal = document.getElementById("edit-start-val");
const endVal = document.getElementById("edit-end-val");
const durVal = document.getElementById("edit-duration");
function updateUI() {
const s = parseFloat(startSlider.value);
const e = parseFloat(endSlider.value);
startVal.textContent = s.toFixed(1) + "s";
endVal.textContent = e.toFixed(1) + "s";
durVal.textContent = (e - s).toFixed(1) + "s";
}
startSlider.addEventListener("input", () => {
if (parseFloat(startSlider.value) >= parseFloat(endSlider.value)) {
startSlider.value = parseFloat(endSlider.value) - 0.5;
}
updateUI();
if (video) video.currentTime = parseFloat(startSlider.value);
});
endSlider.addEventListener("input", () => {
if (parseFloat(endSlider.value) <= parseFloat(startSlider.value)) {
endSlider.value = parseFloat(startSlider.value) + 0.5;
}
updateUI();
if (video) video.currentTime = parseFloat(endSlider.value);
});
window.seekEditVideo = function(which) {
if (!video) return;
const t = which === "start" ? parseFloat(startSlider.value) : parseFloat(endSlider.value);
video.currentTime = t;
if (which === "start") video.play();
};
// Save button
document.getElementById("edit-save-btn").addEventListener("click", async () => {
const start = parseFloat(startSlider.value);
const end = parseFloat(endSlider.value);
if (end - start < 5) {
alert("Trajanje mora biti vsaj 5s");
return;
}
if (end - start > 60) {
alert("Trajanje največ 60s");
return;
}
// Zberi popravljene segmente
const segInputs = document.querySelectorAll("#edit-segments input");
const customSegments = [];
let hasChanges = false;
segInputs.forEach(inp => {
const orig = inp.dataset.orig.replace(/\\n/g, ' ').trim();
const newText = inp.value.trim();
if (orig !== newText) hasChanges = true;
customSegments.push({
start: parseFloat(inp.dataset.start),
end: parseFloat(inp.dataset.end),
text: newText,
});
});
const status = document.getElementById("edit-status");
status.textContent = "⏳ Pošiljam zahtevo...";
document.getElementById("edit-save-btn").disabled = true;
try {
const body = { start, end };
if (hasChanges) body.custom_segments = customSegments;
const res = await fetch(`/api/jobs/${jobId}/recut`, {
method: "POST",
headers: { "Content-Type": "application/json" },
body: JSON.stringify(body),
});
if (!res.ok) {
const err = await res.json();
status.textContent = "❌ " + (err.detail || "Napaka");
document.getElementById("edit-save-btn").disabled = false;
return;
}
status.textContent = "✅ Re-render v vrsti! Bo gotov v ~30s.";
setTimeout(() => {
closeModal();
refreshJobs();
}, 1500);
} catch (e) {
status.textContent = "❌ Napaka: " + e.message;
document.getElementById("edit-save-btn").disabled = false;
}
});
}
function previewJob(id, title) {
// Odpre velik modal z reel videom
const overlay = document.createElement("div");