@ignix-social/creator 0.0.0-stage → 0.1.1

package/dist/box-kit/comfyui/recast/LICENSE-WAN.txt

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/dist/box-kit/comfyui/recast/people.py

@@ -0,0 +1,115 @@
+"""
+WHO IS WHO in the video (ADR-0022 O6): every person, followed frame to frame,
+so a request replaces the ONE person the customer chose — the official Wan
+preprocessing takes the largest person of each frame and can jump between
+two. Pure geometry (no GPU, no model): unit-tested in test_people.py.
+
+A detection is (x1, y1, x2, y2, score) in pixels of the processed frames.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+
+Box = tuple[float, float, float, float]
+
+
+def iou(a: Box, b: Box) -> float:
+    ix = max(0.0, min(a[2], b[2]) - max(a[0], b[0]))
+    iy = max(0.0, min(a[3], b[3]) - max(a[1], b[1]))
+    inter = ix * iy
+    union = (a[2] - a[0]) * (a[3] - a[1]) + (b[2] - b[0]) * (b[3] - b[1]) - inter
+    return inter / union if union > 0 else 0.0
+
+
+def area(b: Box) -> float:
+    return max(0.0, b[2] - b[0]) * max(0.0, b[3] - b[1])
+
+
+@dataclass
+class Track:
+    id: int
+    boxes: dict[int, Box] = field(default_factory=dict)
+
+    def last(self) -> tuple[int, Box]:
+        f = max(self.boxes)
+        return f, self.boxes[f]
+
+
+def track(frames: list[list[Box]], min_iou: float = 0.3, max_gap: int = 8) -> list[Track]:
+    """Greedy IoU linking: each detection joins the track it overlaps most
+    (seen within `max_gap` frames), else starts a new one."""
+    tracks: list[Track] = []
+    for f, dets in enumerate(frames):
+        taken: set[int] = set()
+        for d in sorted(dets, key=area, reverse=True):
+            best, best_iou = None, min_iou
+            for t in tracks:
+                lf, lb = t.last()
+                if t.id in taken or f - lf > max_gap or lf == f:
+                    continue
+                o = iou(lb, d)
+                if o > best_iou:
+                    best, best_iou = t, o
+            if best is None:
+                best = Track(id=len(tracks))
+                tracks.append(best)
+            best.boxes[f] = d
+            taken.add(best.id)
+    return tracks
+
+
+def people(tracks: list[Track], frames: int, min_share: float = 0.3, min_size: float = 0.1) -> list[Track]:
+    """The tracks that are PEOPLE IN THE SHOT: present in at least `min_share`
+    of the frames and not tiny next to the biggest (a passer-by in the far
+    background is not someone to replace)."""
+    if not tracks:
+        return []
+    big = max(area(b) for t in tracks for b in t.boxes.values())
+    keep = []
+    for t in tracks:
+        typical = sorted(area(b) for b in t.boxes.values())[len(t.boxes) // 2]
+        if len(t.boxes) >= min_share * frames and typical >= min_size * big:
+            keep.append(t)
+    return sorted(keep, key=lambda t: -sum(area(b) for b in t.boxes.values()))
+
+
+class ManyPeople(Exception):
+    """Several people and nobody chosen: the customer must pick one."""
+
+
+class PersonNotFound(Exception):
+    """The chosen point is on nobody at that moment."""
+
+
+class NoPerson(Exception):
+    """Nobody to replace in the video."""
+
+
+def choose(
+    found: list[Track], frame: int | None, x: float | None, y: float | None, width: int, height: int
+) -> Track:
+    """The track the customer chose (a point in [0, 1] at a frame), or the
+    only person when there is one."""
+    if not found:
+        raise NoPerson()
+    if frame is None or x is None or y is None:
+        if len(found) > 1:
+            raise ManyPeople()
+        return found[0]
+    px, py = x * width, y * height
+    near = [t for t in found if frame in t.boxes]
+    inside = [t for t in near if t.boxes[frame][0] <= px <= t.boxes[frame][2] and t.boxes[frame][1] <= py <= t.boxes[frame][3]]
+    if inside:
+        # Overlapping boxes: the smallest one holding the point is the one touched.
+        return min(inside, key=lambda t: area(t.boxes[frame]))
+    raise PersonNotFound()
+
+
+def fill(t: Track, frames: int) -> list[Box]:
+    """A box for every frame: missing ones from the nearest frame it was seen."""
+    seen = sorted(t.boxes)
+    out: list[Box] = []
+    for f in range(frames):
+        nearest = min(seen, key=lambda s: abs(s - f))
+        out.append(t.boxes[nearest])
+    return out

package/dist/box-kit/comfyui/recast/persons.py

@@ -0,0 +1,90 @@
+"""
+EVERY PERSON in a frame (ADR-0022 O6), by their body: YOLOX-S (Apache-2.0,
+Megvii; the ONNX export from OpenCV Zoo), the same detector Creator runs on
+the customer's computer to ask "which one?" (the box's `people` op), so both
+sides count the same people. It replaces the detector bundled with Wan's
+preprocessing (yolov10m, from an AGPL-3.0 project).
+
+`decode` is pure numpy (unit-tested in test_persons.py); `Persons` runs the
+model with onnxruntime, on the GPU when there is one.
+"""
+from __future__ import annotations
+
+import numpy as np
+
+SIZE = 640
+MIN_SCORE = 0.4
+NMS_IOU = 0.5
+
+
+def _anchors() -> tuple[np.ndarray, np.ndarray]:
+    grids, strides = [], []
+    for st in (8, 16, 32):
+        n = SIZE // st
+        xv, yv = np.meshgrid(np.arange(n), np.arange(n))
+        grids.append(np.stack((xv, yv), 2).reshape(-1, 2))
+        strides.append(np.full((n * n, 1), st))
+    return np.concatenate(grids), np.concatenate(strides)
+
+
+GRIDS, STRIDES = _anchors()
+
+
+def letterbox(rgb: np.ndarray) -> tuple[np.ndarray, float]:
+    """The frame scaled into the model's square (top-left, grey padding)."""
+    import cv2
+
+    h, w = rgb.shape[:2]
+    r = min(SIZE / h, SIZE / w)
+    pad = np.full((SIZE, SIZE, 3), 114.0, np.float32)
+    small = cv2.resize(rgb, (int(w * r), int(h * r)))
+    pad[: small.shape[0], : small.shape[1]] = small
+    return pad.transpose(2, 0, 1)[None], r
+
+
+def _nms(boxes: np.ndarray, scores: np.ndarray) -> list[int]:
+    order = list(np.argsort(-scores))
+    keep: list[int] = []
+    while order:
+        i = order.pop(0)
+        keep.append(i)
+        x1 = np.maximum(boxes[i, 0], boxes[order, 0])
+        y1 = np.maximum(boxes[i, 1], boxes[order, 1])
+        x2 = np.minimum(boxes[i, 2], boxes[order, 2])
+        y2 = np.minimum(boxes[i, 3], boxes[order, 3])
+        inter = np.clip(x2 - x1, 0, None) * np.clip(y2 - y1, 0, None)
+        area = lambda b: (b[..., 2] - b[..., 0]) * (b[..., 3] - b[..., 1])  # noqa: E731
+        iou = inter / (area(boxes[i]) + area(boxes[order]) - inter + 1e-9)
+        order = [o for o, v in zip(order, iou) if v <= NMS_IOU]
+    return keep
+
+
+def decode(out: np.ndarray, ratio: float, width: int, height: int) -> list[tuple[float, float, float, float]]:
+    """The model's raw output (8400 x 85) -> person boxes (x1, y1, x2, y2) in
+    pixels of the frame, most confident first."""
+    xy = (out[:, :2] + GRIDS) * STRIDES
+    wh = np.exp(out[:, 2:4]) * STRIDES
+    score = out[:, 4] * out[:, 5]  # an object, and that object is a person
+    sel = score >= MIN_SCORE
+    if not sel.any():
+        return []
+    boxes = np.concatenate([xy[sel] - wh[sel] / 2, xy[sel] + wh[sel] / 2], 1) / ratio
+    boxes[:, [0, 2]] = boxes[:, [0, 2]].clip(0, width)
+    boxes[:, [1, 3]] = boxes[:, [1, 3]].clip(0, height)
+    s = score[sel]
+    return [tuple(float(v) for v in boxes[i]) for i in _nms(boxes, s) if boxes[i, 2] > boxes[i, 0] and boxes[i, 3] > boxes[i, 1]]
+
+
+class Persons:
+    def __init__(self, checkpoint: str):
+        import onnxruntime
+
+        have = onnxruntime.get_available_providers()
+        providers = [p for p in ("CUDAExecutionProvider", "CPUExecutionProvider") if p in have]
+        self.session = onnxruntime.InferenceSession(checkpoint, providers=providers)
+        self.input = self.session.get_inputs()[0].name
+
+    def __call__(self, rgb: np.ndarray) -> list[tuple[float, float, float, float]]:
+        blob, ratio = letterbox(rgb)
+        out = self.session.run(None, {self.input: blob})[0][0]
+        return decode(out, ratio, rgb.shape[1], rgb.shape[0])

package/dist/box-kit/comfyui/recast/prep.py

@@ -0,0 +1,204 @@
+"""
+"PUT ME IN THIS VIDEO" ON THIS COMPUTER (ADR-0022 O6): the recast's inputs for
+ComfyUI's own WanAnimateToVideo node, made with the ComfyUI box's Python and
+GPU. Creator talks to it over JSON lines on stdin/stdout (like OpenMontage's
+creator_runner.py); ComfyUI itself still runs with no custom nodes.
+
+The same steps as the Cloud Boost worker (engines/recast-worker/preprocess.py,
+derived from Wan2.2's replacement preprocessing): the video at 30 fps and
+720p area, every person found by their body (YOLOX-S, persons.py) and followed
+(people.py), the chosen one's pose (ViTPose-H wholebody) drawn as Wan draws
+it, their face crops, and the area to regenerate. That area is the person's
+box: the worker's SAM2 mask ends as its bounding box too (get_aug_mask with a
+1x1 grid), so no SAM2 here.
+
+Request:  {"id": 1, "op": "prep", "video": "/abs/in.mp4", "out": "/abs/dir",
+           "person": {"t": 2.0, "x": 0.4, "y": 0.5} | null}
+Response: {"id": 1, "ok": true, "result": {"frames": 177, "width": 720, "height": 1280, "people": 1}}
+          {"id": 1, "ok": false, "error": "many_people" | "person_not_found" | "no_person" | ...}
+Writes face.mp4 (512x512), pose.mp4, bg.mp4 and mask.mp4 into `out`.
+Models: CREATOR_BOX_DATA/detect/yolox_s.onnx, CREATOR_BOX_DATA/pose/vitpose_h_wholebody_model.onnx.
+"""
+from __future__ import annotations
+
+import json
+import os
+import sys
+import traceback
+from pathlib import Path
+
+HERE = Path(__file__).resolve().parent
+sys.path.insert(0, str(HERE))
+
+FPS = 30
+AREA = 1280 * 720
+MAX_FRAMES = 15 * FPS
+# The box grown like the worker's mask dilation (7 px kernel, 3 iterations).
+GROW = 10
+
+
+def _onnx():
+    # torch first: its CUDA libraries are the ones onnxruntime's CUDA provider loads.
+    try:
+        import torch  # noqa: F401
+    except ImportError:
+        pass
+    import onnxruntime
+
+    try:
+        onnxruntime.preload_dlls()
+    except Exception:  # noqa: BLE001 - older builds have no preload: torch's libraries are loaded
+        pass
+    return onnxruntime
+
+
+class ViTPose:
+    """ViTPose-H wholebody (133 keypoints) through onnxruntime — Wan2.2's
+    pose2d.ViTPose without its torch dependency (Apache-2.0, Alibaba Wan Team)."""
+
+    def __init__(self, checkpoint: str):
+        ort = _onnx()
+        have = ort.get_available_providers()
+        providers = [p for p in ("CUDAExecutionProvider", "CPUExecutionProvider") if p in have]
+        self.session = ort.InferenceSession(checkpoint, providers=providers)
+        self.name = self.session.get_inputs()[0].name
+
+    def __call__(self, rgb, box):
+        import numpy as np
+        from wan_pose_utils import bbox_from_detector, crop, keypoints_from_heatmaps
+
+        bbox = np.array([*box, 1.0])
+        if (bbox[2] - bbox[0]) < 10 or (bbox[3] - bbox[1]) < 10:
+            bbox = np.array([0, 0, rgb.shape[1], rgb.shape[0], 1.0])
+        center, scale = bbox_from_detector(bbox, (256, 192), rescale=1.25)
+        img, _, _, _ = crop(rgb, center, scale, (256, 192))
+        mean, std = np.array([0.485, 0.456, 0.406]), np.array([0.229, 0.224, 0.225])
+        x = ((img / 255.0 - mean) / std).transpose(2, 0, 1).astype(np.float32)[None]
+        heat = self.session.run([], {self.name: x})[0]
+        points, prob = keypoints_from_heatmaps(
+            heatmaps=heat, center=np.array(center)[None], scale=np.array(scale)[None] * 200, unbiased=True, use_udp=False
+        )
+        return np.concatenate([points, prob], axis=2)
+
+
+def read_frames(path: str):
+    """The video at 30 fps (by time), each frame resized to a 720p area."""
+    import av
+    import numpy as np
+    from wan_utils import resize_by_area
+
+    out = []
+    with av.open(path) as c:
+        stream = c.streams.video[0]
+        next_t = 0.0
+        for frame in c.decode(stream):
+            t = float(frame.time or 0.0)
+            if t + 1e-6 < next_t:
+                continue
+            rgb = frame.to_ndarray(format="rgb24")
+            while next_t <= t + 1e-6 and len(out) < MAX_FRAMES:
+                out.append(resize_by_area(rgb, AREA, divisor=16))
+                next_t += 1.0 / FPS
+            if len(out) >= MAX_FRAMES:
+                break
+    return [np.ascontiguousarray(f) for f in out]
+
+
+def write_video(path: str, frames, gray: bool = False):
+    """A near-lossless MPEG-4 file ComfyUI's LoadVideo reads (FFmpeg's own encoder)."""
+    import av
+    import numpy as np
+
+    h, w = frames[0].shape[:2]
+    with av.open(path, "w") as c:
+        s = c.add_stream("mpeg4", rate=FPS)
+        s.width, s.height, s.pix_fmt = w, h, "yuv420p"
+        s.bit_rate = 40_000_000
+        s.options = {"qscale": "1"}
+        for f in frames:
+            rgb = np.repeat(f[:, :, None], 3, axis=2) if gray else f
+            for p in s.encode(av.VideoFrame.from_ndarray(rgb, format="rgb24")):
+                c.mux(p)
+        for p in s.encode():
+            c.mux(p)
+
+
+def prep(req: dict) -> dict:
+    import cv2
+    import numpy as np
+
+    import people
+    from persons import Persons
+    from wan_pose_draw import draw_aapose_by_meta_new
+    from wan_pose_utils import AAPoseMeta, load_pose_metas_from_kp2ds_seq
+    from wan_utils import get_face_bboxes
+
+    data = Path(os.environ.get("CREATOR_BOX_DATA", ""))
+    out = Path(req["out"])
+    out.mkdir(parents=True, exist_ok=True)
+    frames = read_frames(req["video"])
+    if not frames:
+        return {"ok": False, "error": "invalid_input"}
+    h, w = frames[0].shape[:2]
+    detector = Persons(str(data / "detect" / "yolox_s.onnx"))
+    found = people.people(people.track([detector(f) for f in frames]), len(frames))
+    at = req.get("person") or {}
+    frame_at = None if at.get("t") is None else min(len(frames) - 1, max(0, round(float(at["t"]) * FPS)))
+    try:
+        chosen = people.choose(found, frame_at, at.get("x"), at.get("y"), w, h)
+    except people.ManyPeople:
+        return {"ok": False, "error": "many_people"}
+    except people.PersonNotFound:
+        return {"ok": False, "error": "person_not_found"}
+    except people.NoPerson:
+        return {"ok": False, "error": "no_person"}
+    boxes = people.fill(chosen, len(frames))
+
+    pose = ViTPose(str(data / "pose" / "vitpose_h_wholebody_model.onnx"))
+    kp2ds = np.concatenate([pose(f, b) for f, b in zip(frames, boxes)], 0)
+    metas = load_pose_metas_from_kp2ds_seq(kp2ds, width=w, height=h)
+
+    faces, poses, bgs, masks = [], [], [], []
+    for frame, meta, box in zip(frames, metas, boxes):
+        x1, x2, y1, y2 = get_face_bboxes(meta["keypoints_face"][:, :2], scale=1.3, image_shape=(h, w))
+        crop = frame[y1:y2, x1:x2]
+        faces.append(cv2.resize(crop if crop.size else frame, (512, 512)))
+        poses.append(draw_aapose_by_meta_new(np.zeros_like(frame), AAPoseMeta.from_humanapi_meta(meta)))
+        bx1, by1 = max(0, int(box[0]) - GROW), max(0, int(box[1]) - GROW)
+        bx2, by2 = min(w, int(box[2]) + GROW), min(h, int(box[3]) + GROW)
+        mask = np.zeros((h, w), np.uint8)
+        mask[by1:by2, bx1:bx2] = 255
+        bg = frame.copy()
+        bg[by1:by2, bx1:bx2] = 0
+        bgs.append(bg)
+        masks.append(mask)
+    write_video(str(out / "face.mp4"), faces)
+    write_video(str(out / "pose.mp4"), poses)
+    write_video(str(out / "bg.mp4"), bgs)
+    write_video(str(out / "mask.mp4"), masks, gray=True)
+    return {"ok": True, "result": {"frames": len(frames), "width": w, "height": h, "people": len(found)}}
+
+
+OPS = {"hello": lambda _req: {"ok": True, "result": {"prep": 1}}, "prep": prep}
+
+
+def main() -> None:
+    for line in sys.stdin:
+        line = line.strip()
+        if not line:
+            continue
+        msg_id = None
+        try:
+            msg = json.loads(line)
+            msg_id = msg.get("id")
+            op = OPS.get(msg.get("op"))
+            reply = op(msg) if op else {"ok": False, "error": "unsupported"}
+        except Exception:  # noqa: BLE001 - one bad request must not end the process
+            traceback.print_exc(file=sys.stderr)
+            reply = {"ok": False, "error": "failed"}
+        sys.stdout.write(json.dumps({"id": msg_id, **reply}) + "\n")
+        sys.stdout.flush()
+
+
+if __name__ == "__main__":
+    main()