SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2411.17697 · 2024

StableAnimator: High-Quality Identity-Preserving Human Image Animation

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 8 failures of our instrument; 1 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2411.17697.json · badge: /api/ran/2411.17697.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
Francis-Rings/StableAnimator canonical (classifier label)
classifier basis: census
10 ran · 1 no recorded run
FunctionStatusWhere it lives
FeedForward Ran
ran; our contract draft was wrong, not the code
Francis-Rings/StableAnimator/animation/modules/id_encoder.py
pointer only (licence: MIT); the stored copy get_code serves, from rainbowluocs/openomni, is pointer only (licence: NONE) · get_code("5105747c2711b1cb")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from rainbowluocs/openomni, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

alpha_blend_color Ran
ran; our contract draft was wrong, not the code
Francis-Rings/StableAnimator/DWPose/skeleton_extraction.py
pointer only (licence: MIT); the stored copy get_code serves, from lihxxx/DisPose, is pointer only (licence: NONE) · get_code("d5db4dc031f72a81")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from lihxxx/DisPose, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

draw_bodypose Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Francis-Rings/StableAnimator/DWPose/skeleton_extraction.py
code served (permissive licence) · get_code("68e1817b5a0cb440")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family int, shape (100, 150, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import cv2
import math

# Create a small canvas (H=100, W=150, C=3)
H, W, C = 100, 150, 3
canvas = np.ones((H, W, C), dtype=np.uint8) * 255

# candidate: array of keypoint positions normalized to [0, 1]
# Shape should be (num_keypoints, 2) where each row is [x_norm, y_norm]
# 18 keypoints for body pose
candidate = np.array([
    [0.5, 0.3],  # keypoint 0
    [0.5, 0.35], # keypoint 1
    [0.5, 0.4],  # keypoint 2
    [0.45, 0.45], # keypoint 3
    [0.4, 0.5],  # keypoint 4
    [0.55, 0.45], # keypoint 5
    [0.6, 0.5],  # keypoint 6
    [0.65, 0.55], # keypoint 7
    [0.7, 0.6],  # keypoint 8
    [0.48, 0.55], # keypoint 9
    [0.46, 0.65], # keypoint 10
    [0.44, 0.75], # keypoint 11
    [0.52, 0.55], # keypoint 12
    [0.54, 0.65], # keypoint 13
    [0.56, 0.75], # keypoint 14
    [0.42, 0.3],  # keypoint 15
    [0.58, 0.3],  # keypoint 16
    [0.4, 0.25],  # keypoint 17
], dtype=np.float32)

# subset: array mapping persons to keypoint indices
# Shape (num_persons, 18) where each value is the index into candidate array
# -1 means keypoint not detected for that person
subset = np.array([
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17],
    [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, -1, -1, -1, -1],
], dtype=np.float32)

# score: confidence scores for each keypoint per person
# Shape (num_persons, 18) where each value is in [0, 1]
score = np.array([
    [0.9, 0.85, 0.88, 0.92, 0.87, 0.91, 0.89, 0.86, 0.84, 0.9, 0.88, 0.85, 0.89, 0.87, 0.86, 0.91, 0.90, 0.88],
    [0.8, 0.82, 0.78, 0.85, 0.79, 0.83, 0.81, 0.77, 0.75, 0.8, 0.78, 0.76, 0.79, 0.77, -1, -1, -1, -1],
], dtype=np.float32)

ARGS = {
    "canvas": canvas,
    "candidate": candidate,
    "subset": subset,
    "score": score,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (100, 150, 3) — shape (100, 150, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
draw_bodypose_aligned Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Francis-Rings/StableAnimator/DWPose/training_skeleton_extraction.py
code served (permissive licence) · get_code("44ef0c50e1ed61e7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family int, shape (100, 150, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import cv2

# Create a small canvas image (height=100, width=150, channels=3)
canvas = np.ones((100, 150, 3), dtype=np.uint8) * 200

# candidate: array of keypoint coordinates, shape (num_keypoints, 2)
# Each keypoint has normalized x, y coordinates (values between 0 and 1)
candidate = np.array([
    [0.3, 0.4],
    [0.5, 0.5],
    [0.6, 0.6],
    [0.7, 0.7],
    [0.8, 0.8],
    [0.4, 0.3],
    [0.5, 0.4],
    [0.6, 0.5],
    [0.7, 0.6],
    [0.2, 0.5],
    [0.3, 0.6],
    [0.4, 0.7],
    [0.1, 0.4],
    [0.2, 0.3],
    [0.3, 0.2],
    [0.2, 0.1],
    [0.3, 0.1],
    [0.4, 0.2]
], dtype=np.float32)

# subset: array of shape (num_people, 18) where each row contains indices
# into the candidate array for each of the 18 keypoints. -1 means not detected.
subset = np.array([
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17],
    [0, 1, 2, 3, 4, 5, 6, 7, 8, -1, -1, -1, -1, -1, -1, -1, -1, -1]
], dtype=np.float32)

# score: array of shape (num_people, 18) with confidence scores for each keypoint
# Values should be between 0 and 1
score = np.array([
    [0.9, 0.8, 0.85, 0.7, 0.75, 0.8, 0.82, 0.78, 0.9, 0.6, 0.65, 0.7, 0.8, 0.75, 0.85, 0.8, 0.82, 0.79],
    [0.95, 0.85, 0.9, 0.8, 0.82, 0.88, 0.86, 0.84, 0.92, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5]
], dtype=np.float32)

ARGS = {
    "canvas": canvas,
    "candidate": candidate,
    "subset": subset,
    "score": score
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (100, 150, 3) — shape (100, 150, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
draw_handpose Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Francis-Rings/StableAnimator/DWPose/skeleton_extraction.py
pointer only (licence: MIT); the stored copy get_code serves, from lihxxx/DisPose, is pointer only (licence: NONE) · get_code("b408758fe55668b9")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from lihxxx/DisPose, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

draw_handpose_aligned Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Francis-Rings/StableAnimator/DWPose/training_skeleton_extraction.py
code served (permissive licence) · get_code("1e8a79e43e77ad58")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ndarray, ndim 3, dtype family int, shape (480, 640, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import cv2
import numpy as np
import matplotlib

# Create a small canvas (height=480, width=640, channels=3)
canvas = np.ones((480, 640, 3), dtype=np.uint8) * 255

# Create hand peaks - 21 keypoints per hand, each with normalized (x, y) coordinates
# Using 2 hands as examples
all_hand_peaks = [
    np.array([
        [0.1, 0.1], [0.15, 0.12], [0.2, 0.14], [0.25, 0.16], [0.3, 0.18],  # thumb
        [0.1, 0.3], [0.12, 0.35], [0.14, 0.4], [0.16, 0.45],                # index
        [0.1, 0.5], [0.12, 0.55], [0.14, 0.6], [0.16, 0.65],                # middle
        [0.1, 0.7], [0.12, 0.75], [0.14, 0.8], [0.16, 0.85],                # ring
        [0.1, 0.9], [0.12, 0.92], [0.14, 0.94], [0.16, 0.96]                # pinky
    ], dtype=np.float32),
    np.array([
        [0.5, 0.2], [0.55, 0.22], [0.6, 0.24], [0.65, 0.26], [0.7, 0.28],
        [0.5, 0.4], [0.52, 0.45], [0.54, 0.5], [0.56, 0.55],
        [0.5, 0.6], [0.52, 0.65], [0.54, 0.7], [0.56, 0.75],
        [0.5, 0.8], [0.52, 0.85], [0.54, 0.9], [0.56, 0.95],
        [0.5, 0.9], [0.52, 0.92], [0.54, 0.94], [0.56, 0.96]
    ], dtype=np.float32)
]

# Create hand scores - confidence scores for each keypoint (21 per hand)
all_hand_scores = [
    np.array([0.9, 0.85, 0.88, 0.87, 0.86, 0.92, 0.91, 0.89, 0.88, 
              0.85, 0.84, 0.83, 0.82, 0.88, 0.87, 0.86, 0.85, 0.9, 0.89, 0.88, 0.87], dtype=np.float32),
    np.array([0.88, 0.86, 0.87, 0.85, 0.84, 0.91, 0.90, 0.88, 0.87,
              0.83, 0.82, 0.81, 0.80, 0.86, 0.85, 0.84, 0.83, 0.89, 0.88, 0.87, 0.86], dtype=np.float32)
]

ARGS = {
    'canvas': canvas,
    'all_hand_peaks': all_hand_peaks,
    'all_hand_scores': all_hand_scores
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, matplotlib, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 3, dtype family int, shape (480, 640, 3) — shape (480, 640, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_images_from_folder Ran
ran, honoured the contract we drafted
Francis-Rings/StableAnimator/inference_basic.py
code served (permissive licence) · get_code("66ae98065714bafd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 4, dtype family int; no output shape is stated. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:59:35Z. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
from PIL import Image

# Create a temporary folder with test PNG images
temp_dir = tempfile.mkdtemp()

# Create a few test PNG images with naming convention that the function expects
for i in range(3):
    img = Image.new('RGB', (100, 100), color=(73, 109, 137))
    img.save(os.path.join(temp_dir, f'image_{i}.png'))

ARGS = {
    'folder': temp_dir,
    'width': 64,
    'height': 48
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 4, dtype family int — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_images_from_folder Ran
ran, honoured the contract we drafted
Francis-Rings/StableAnimator/train_single.py
code served (permissive licence) · get_code("b6014c1c431aa69e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 4, dtype family int; no output shape is stated. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:59:35Z. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import os
import tempfile
from PIL import Image, ImageDraw

# Create a temporary directory with PNG files
temp_dir = tempfile.mkdtemp()

# Create a few simple PNG images with naming convention expected by the function
for i in range(3):
    img = Image.new('RGB', (10, 10), color=(i*80, i*80, i*80))
    filename = f"image_{i}.png"
    img.save(os.path.join(temp_dir, filename))

ARGS = {
    'folder': temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 4, dtype family int — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
reshape_tensor Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
Francis-Rings/StableAnimator/animation/modules/id_encoder.py
pointer only (licence: MIT); the stored copy get_code serves, from Minyeol/MIDAS, is pointer only (licence: NONE) · get_code("4cb2e2a2ca0bec9f")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from Minyeol/MIDAS, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

validate_and_convert_image Ran
ran; our contract draft was wrong, not the code
Francis-Rings/StableAnimator/train_single.py
code served (permissive licence) · get_code("f38487057d980dce")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Image. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:59:35Z. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from PIL import Image
import numpy as np

torch.manual_seed(0)

# Create test image as a PIL Image
pil_image = Image.new('RGB', (512, 512), color=(100, 150, 200))

# Create test image as a torch tensor (3, 4, 5) - 3 channels, 4 height, 5 width
tensor_image = torch.randint(0, 256, (3, 4, 5), dtype=torch.uint8).float() / 255.0

ARGS = {
    'image': tensor_image,
    'target_size': (256, 256)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int describe that coercion, not the output, and are not stated
Environment
image syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_image_grid Not yet run
unverified (harvested, no recorded run)
Francis-Rings/StableAnimator/train_single.py
code served (permissive licence) · get_code("b40e78b0b1ea6160")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 5 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 1 the same month, 5 later) and 6 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
FeedForward 47 other papers
28 dated earlier, 2 the same month, 16 later, 1 same year, order not determinable · 4 only as an assembly brick
2204.14198 (2022-04)
Flamingo: a Visual Language Model for Few-Shot Learning
2308.02463 (2023-08)
Towards Generalist Foundation Model for Radiology by Leveraging Web-scale 2D&3D Medical Data
2308.06721 (2023-08)
IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models
and 44 more not named here
reshape_tensor 30 other papers
20 dated earlier, 10 later
2308.06721 (2023-08)
IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models
2312.04461 (2023-12)
PhotoMaker: Customizing Realistic Human Photos via Stacked ID Embedding
2312.16693 (2023-12)
I2V-Adapter: A General Image-to-Video Adapter for Diffusion Models
and 27 more not named here
alpha_blend_color 1 other paper
1 later
2412.09349 (2024-12)
DisPose: Disentangling Pose Guidance for Controllable Human Image Animation
draw_bodypose 1 other paper
1 later
2412.09349 (2024-12)
DisPose: Disentangling Pose Guidance for Controllable Human Image Animation
draw_handpose 1 other paper
1 later
2412.09349 (2024-12)
DisPose: Disentangling Pose Guidance for Controllable Human Image Animation

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Current diffusion models for human image animation struggle to ensure identity (ID) consistency. This paper presents StableAnimator, the first end-to-end ID-preserving video diffusion framework, which synthesizes high-quality videos without any post-processing, conditioned on a reference image and a sequence of poses. Building upon a video diffusion model, StableAnimator contains carefully designed modules for both training and inference striving for identity consistency. In particular, StableAnimator begins by computing image and face embeddings with off-the-shelf extractors, respectively and face embeddings are further refined by interacting with image embeddings using a global content-aware Face Encoder. Then, StableAnimator introduces a novel distribution-aware ID Adapter that prevents interference caused by temporal layers while preserving ID via alignment. During inference, we propose a novel Hamilton-Jacobi-Bellman (HJB) equation-based optimization to further enhance the face quality. We demonstrate that solving the HJB equation can be integrated into the diffusion denoising process, and the resulting solution constrains the denoising path and thus benefits ID preservation. Experiments on multiple benchmarks show the effectiveness of StableAnimator both qualitatively and quantitatively.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2411.17697")
get_code_for_paper("2411.17697")
have("2411.17697")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2411.17697.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2411.17697.svg)](https://syntology.ai/paper/2411.17697)

Connect an agent — have() is free.