konradre/video-production-skills / skills/video-refs-continuity
video-refs-continuity
VerifiedContinuity through consistency for generated video: builds the DERIVED reference set and start image every shot needs, keeps the continuity ledger between shots, and runs the refs gate in code before any generation call. Use when a shot must match the shot before it (room, cast, seats, wardrobe, held props, door swing, a named state), when a character, product or room reference is built or cut from a keeper, when a plate or start image is generated, or when seeds keep drifting between shots. Triggers — "continuity", "refs gate", "reference set", "start image", "continuity ledger", "backfill from accepted", "the door is on the wrong wall", "not the same guy", "the same shape in all videos". Not for the cost line, the GO and the submit itself — use video-gen-cost-gate. Not for the prompt's wording — use video-prompt-dialects. Not for upscale, film look, grain or the delivery encode — use video-finish.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 31.6 KB |
| references/ASSET-HYGIENE.md | markdown | 6.1 KB |
| references/CONTINUITY-AXES.md | markdown | 6.3 KB |
| references/REFERENCE-CONTRACT.md | markdown | 19.0 KB |
| references/REPAIR-LADDER.md | markdown | 6.0 KB |
| references/SCENE-PROXY.md | markdown | 24.5 KB |
| references/ZOOM-OUT-CHAIN.md | markdown | 4.1 KB |
| references/refs-required.example.json | config | 3.1 KB |
| scripts/air_pieces.pyexecutable | script | 3.9 KB |
| scripts/cast_stress.pyexecutable | script | 17.8 KB |
| scripts/comfy_preflight.py | script | 22.3 KB |
| scripts/comfy_ready.pyexecutable | script | 2.7 KB |
| scripts/comfy_up.shexecutable | script | 2.6 KB |
| scripts/knit_letters.pyexecutable | script | 7.3 KB |
| scripts/plate_pieces.pyexecutable | script | 4.9 KB |
| scripts/ref_dag.pyexecutable | script | 5.2 KB |
| scripts/refs_gate.pyexecutable | script | 41.1 KB |
| scripts/scene_blockout.py | script | 6.4 KB |
| scripts/scene_proxy.py | script | 16.6 KB |
| scripts/split_sheet.pyexecutable | script | 1.4 KB |
| scripts/sprinkle_pieces.pyexecutable | script | 2.8 KB |
scripts/scene_blockout.py
View on GitHub#!/usr/bin/env python3
"""Keeper frame -> labelled grey-box scene, headless, through the ComfyUI server that carries OmniCam.
python3 scene_blockout.py <frame.png> [--out PREFIX] [--labels a,b,c] [--mode blockout|hybrid|depth_mesh]
[--threshold 0.55] [--no-walls] [--host URL]
Writes <PREFIX>.scene.json (the MotionScene + summary the route returned) and prints the object table.
Needs: a RUNNING ComfyUI server (the browser is not needed) with the OmniCam custom node, a MoGe checkpoint
under models/geometry_estimation and a sam3* checkpoint under models/checkpoints. The host comes from
--host or the COMFY_HOST env var; it is never written into this skill. Cost: $0, ~15-25 s per frame.
Fails closed: a server that does not answer, a missing checkpoint or a failed job stops with the reason.
"""
import argparse, json, mimetypes, os, sys, time, uuid, urllib.error, urllib.request
from pathlib import Path
DEFAULT_LABELS = ["person", "sofa", "armchair", "chair", "table", "coffee table", "bed", "door", "window",
"lamp", "television", "fireplace", "shelf", "plant", "car", "building", "tree"]
SUBFOLDER = "scene-proxy"
def call(host, path, body=None, files=None, timeout=90):
if files:
boundary = uuid.uuid4().hex
chunks = []
for name, value in (body or {}).items():
chunks.append(f"--{boundary}\r\nContent-Disposition: form-data; name=\"{name}\"\r\n\r\n{value}\r\n".encode())
for name, p in files.items():
ctype = mimetypes.guess_type(p)[0] or "application/octet-stream"
chunks.append(f"--{boundary}\r\nContent-Disposition: form-data; name=\"{name}\"; filename=\"{Path(p).name}\"\r\n"
f"Content-Type: {ctype}\r\n\r\n".encode() + Path(p).read_bytes() + b"\r\n")
chunks.append(f"--{boundary}--\r\n".encode())
data, headers = b"".join(chunks), {"Content-Type": f"multipart/form-data; boundary={boundary}"}
else:
data = None if body is None else json.dumps(body).encode()
headers = {"Content-Type": "application/json"}
req = urllib.request.Request(host + path, data=data, headers=headers, method="POST" if data else "GET")
try:
with urllib.request.urlopen(req, timeout=timeout) as r:
return json.load(r)
except urllib.error.HTTPError as e:
sys.exit(f"HTTP {e.code} on {path}: {e.read().decode(errors='replace')[:500]}")
except urllib.error.URLError as e:
sys.exit(f"ComfyUI server not reachable at {host} ({e.reason}). It must be RUNNING (no browser needed).")
def main():
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("image")
ap.add_argument("--out", help="output prefix (default: <image stem>-scene)")
ap.add_argument("--labels", help="comma list of SAM3 labels; set it PER SCENE to kill phantoms")
ap.add_argument("--mode", default="blockout", choices=["blockout", "hybrid", "depth_mesh"])
ap.add_argument("--threshold", type=float, default=0.55, help="SAM3 score floor; raise to drop phantoms")
ap.add_argument("--no-walls", action="store_true")
ap.add_argument("--host", default=os.environ.get("COMFY_HOST", ""), help="ComfyUI URL (or COMFY_HOST)")
a = ap.parse_args()
if not a.host:
sys.exit("set --host or COMFY_HOST to the ComfyUI server URL")
host = a.host.rstrip("/")
out = a.out or (Path(a.image).stem + "-scene")
labels = [s.strip() for s in a.labels.split(",")] if a.labels else DEFAULT_LABELS
caps = call(host, "/majoor/omnicam/reconstruction/capabilities")
geo = {p["provider_id"]: p for p in caps.get("geometry", [])}
seg = {p["provider_id"]: p for p in caps.get("segmentation", [])}
for pid, table in (("comfy_moge", geo), ("comfy_sam3", seg)):
if pid == "comfy_sam3" and a.mode == "depth_mesh":
continue
if not table.get(pid, {}).get("available"):
sys.exit(f"provider {pid} unavailable: {table.get(pid, {}).get('reason', 'not registered')}")
up = call(host, "/upload/image", {"subfolder": SUBFOLDER, "overwrite": "true"}, files={"image": a.image})
ref = f"{up['subfolder']}/{up['name']} [input]" if up.get("subfolder") else f"{up['name']} [input]"
body = {"node_id": f"scene-{Path(a.image).stem[:40]}", "client_id": "sceneproxy",
"source": {"kind": "annotated_input", "value": ref},
"settings": {"mode": a.mode, "quality": "balanced", "provider": "comfy_moge",
"segmentation_provider": "comfy_sam3", "sam3_threshold": a.threshold,
"semantic_labels": labels, "max_blockout_objects": 24, "detect_ground": True,
"detect_walls": not a.no_walls, "source_texture": False}}
job = call(host, "/majoor/omnicam/reconstruction/jobs", body)
jid, t0, last = job["job_id"], time.time(), ""
while True:
s = call(host, f"/majoor/omnicam/reconstruction/jobs/{jid}?client_id=sceneproxy")
line = f"{s['state']} {s.get('stage', '')} {s.get('message', '')}"
if line != last:
print(f" {time.time() - t0:5.1f}s {line}", file=sys.stderr); last = line
if str(s["state"]).upper() in ("DONE", "FAILED", "STOPPED"):
break
if time.time() - t0 > 600:
sys.exit("job timeout after 600 s")
time.sleep(2)
if str(s["state"]).upper() != "DONE":
sys.exit(f"job {s['state']}: {json.dumps(s.get('error'))}")
res = call(host, f"/majoor/omnicam/reconstruction/jobs/{jid}/result?client_id=sceneproxy")
res["source_image"] = str(Path(a.image).resolve()); res["labels"] = labels
Path(f"{out}.scene.json").write_text(json.dumps(res, indent=1))
ms = res["motion_scene"]; cam = ms["cameras"][0]["track"]["keyframes"][0]["camera"]
print(f"saved {out}.scene.json | {time.time() - t0:.0f} s | warnings {res.get('warnings')}")
print(f"canvas {ms['canvas']['width']}x{ms['canvas']['height']} | source camera at {[round(v, 2) for v in cam['position']]} "
f"fov {cam['fov']:.0f} | relative units, NOT metric")
print(f"{'object':28} {'type':6} {'position x,y,z':26} {'size w,h,d':22} conf")
for o in ms["objects"]:
if o["type"] == "null":
continue
r = o.get("reconstruction", {})
print(f"{o['id'][:28]:28} {o['type']:6} {str([round(v, 2) for v in o['position']]):26} "
f"{str([round(v, 2) for v in o['size']]):22} {r.get('confidence', 0):.2f}")
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 472k | Updated | Claude CodeCodex | Unverified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |