SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2211.01324 · 2022

eDiff-I: Text-to-Image Diffusion Models with an Ensemble of Expert Denoisers

arXiv · PDF · Open in the Atlas

Code that ran

11 of 13 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 8 have no contract verdict; 1 failure of our instrument; 2 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/2211.01324.json · badge: /api/ran/2211.01324.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
cross-domain-compositing/cross-domain-compositing community (classifier label)
classifier basis not recorded on the link
8 ran · 2 no recorded run
cloneofsimo/paint-with-words-sd reimplementation (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
always_round Ran
ran, honoured the contract we drafted
cloneofsimo/paint-with-words-sd/paint_with_words/paint_with_words.py
code served (permissive licence) · get_code("a611fe4aef89ea52")
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) and returned int, ndim 0, dtype family int, shape () 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 torch
import numpy as np

torch.manual_seed(0)

# Analyze the function:
# - Takes parameter 'x' which is used with int(), modulo, comparison, and round()
# - x should be a numeric type (int, float, or numeric tensor)
# - The function performs arithmetic and logical operations on x

# Test with a simple float value
x = 3.7

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () 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
cross_attention_masks Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/cdc/prompt_masking.py
code served (permissive licence) · get_code("99b6da8de85fe2bc")
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 Tensor, ndim 4, dtype family float, shape (2, 77, 3, 4) 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 torch

torch.manual_seed(0)

# Construct a valid mask tensor: shape (batch, 1, height, width)
# The function extracts mask[:, 0, :, :], so we need at least 2 dimensions after batch
mask = torch.randn(2, 1, 3, 4)

# Construct token indices
tokens_in = [0, 1, 5]
tokens_out = [10, 15, 20]

# Weights for masking
w_in = 1.0
w_out = 0.5

# Maximum number of tokens in the output
max_tokens = 77

ARGS = {
    "mask": mask,
    "tokens_in": tokens_in,
    "tokens_out": tokens_out,
    "w_in": w_in,
    "w_out": w_out,
    "max_tokens": max_tokens,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 77, 3, 4) — shape (2, 77, 3, 4) 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
get_schedule Ran
ran, honoured the contract we drafted
cross-domain-compositing/cross-domain-compositing/cdc/scheduler.py
code served (permissive licence) · get_code("314afb570b35e5c0")
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 1, dtype family int; no output shape is stated. The image it ran in was not recorded. 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 torch
import numpy as np

# Analyze the function to understand what arguments are needed
# get_schedule(t_T, t_0, n_sample, n_steplength, debug=0)
# 
# From the function body:
# - t_T: upper bound for time values (used in comparisons and loops)
# - t_0: lower bound for time values (checked in _check_times)
# - n_sample: number of samples (checked > 1 if n_steplength > 1)
# - n_steplength: step length (used in min() and loops)
# - debug: optional debug flag (defaults to 0)
#
# From _check_times:
# - times[0] > times[1]: first element must be > second
# - times[-1] == -1: last element must be -1
# - consecutive elements differ by exactly 1
# - all elements in range [t_0, t_T]
#
# The function builds a times list starting from t_T and going down to -1
# So we need: t_T >= 0, t_0 <= -1 (since -1 must be in range), n_sample >= 1, n_steplength >= 1

ARGS = {
    't_T': 5,           # upper time bound
    't_0': -1,          # lower time bound (must allow -1)
    'n_sample': 2,      # number of samples (> 1 to avoid RuntimeError if n_steplength > 1)
    'n_steplength': 1,  # step length
    'debug': 0,         # debug flag (0 = no debug output)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
matplotlib, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, 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 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
get_schedule_jump Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/cdc/scheduler.py
code served (permissive licence) · get_code("aefaff0f568e5037")
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 1, dtype family int; no output shape is stated. The image it ran in was not recorded. 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 torch
import numpy as np

# Analyzing get_schedule_jump function:
# - t_T: integer representing total time steps
# - n_sample: integer for sampling parameter
# - jump_length: integer for jump length
# - jump_n_sample: integer for jump sample count
# - jump2_length: integer for second jump length (default 1)
# - jump2_n_sample: integer for second jump sample count (default 1)
# - jump3_length: integer for third jump length (default 1)
# - jump3_n_sample: integer for third jump sample count (default 1)
# - start_resampling: integer for resampling start point (default 100000000)
# - collapse_increasing: boolean flag (default False)

# The function generates a schedule of time steps with jumps
# It needs reasonable parameters that create a valid time schedule

ARGS = {
    "t_T": 100,
    "n_sample": 2,
    "jump_length": 10,
    "jump_n_sample": 2,
    "jump2_length": 5,
    "jump2_n_sample": 2,
    "jump3_length": 2,
    "jump3_n_sample": 1,
    "start_resampling": 100000000,
    "collapse_increasing": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, 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 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
get_word_inds Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/cdc/prompt_masking.py
code served (permissive licence) · get_code("2032ab3924e8cd43")
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 1, dtype family int, shape (10,) 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

class MockTokenizer:
    def encode(self, text):
        # Return a list of token IDs
        # We need to return something that when decoded and stripped gives us subword tokens
        words = text.split(' ')
        tokens = [0]  # start token
        for word in words:
            # Simulate tokenizing each word into subword tokens
            for char in word:
                tokens.append(ord(char) % 50 + 1)
        tokens.append(0)  # end token
        return tokens
    
    def decode(self, token_ids):
        # Decode a single token ID into a character/subword
        if not token_ids or token_ids[0] == 0:
            return ""
        token_id = token_ids[0]
        # Map token IDs back to characters
        char = chr((token_id - 1) % 50 + 97)
        return char

text = "hello world"
words = [0, 1]
tokenizer = MockTokenizer()

ARGS = {
    "text": text,
    "words": words,
    "tokenizer": tokenizer
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (10,) — shape (10,) 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
inj_forward Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cloneofsimo/paint-with-words-sd/paint_with_words/paint_with_words.py
code served (permissive licence) · get_code("de7114b637a2e1b6")
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) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 4) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 torch
import torch.nn as nn
import torch.nn.functional as F
import math

torch.manual_seed(0)

# Build a minimal self object that has all required attributes and methods
class MinimalAttention(nn.Module):
    def __init__(self):
        super().__init__()
        # to_q, to_k, to_v: linear layers that project hidden_states to query/key/value
        self.to_q = nn.Linear(4, 4)
        self.to_k = nn.Linear(4, 4)
        self.to_v = nn.Linear(4, 4)
        
        # to_out: sequential with two layers
        self.to_out = nn.Sequential(
            nn.Linear(4, 4),
            nn.Identity()
        )
        
        # scale factor for attention scores
        self.scale = 0.5
        
        # Number of heads (inferred from reshape operations)
        self.num_heads = 2
    
    def reshape_heads_to_batch_dim(self, tensor):
        # tensor shape: (batch_size, sequence_length, dim)
        # reshape to (batch_size * num_heads, sequence_length, dim // num_heads)
        batch_size, seq_len, dim = tensor.shape
        tensor = tensor.reshape(batch_size, seq_len, self.num_heads, dim // self.num_heads)
        tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size * self.num_heads, seq_len, dim // self.num_heads)
        return tensor
    
    def reshape_batch_dim_to_heads(self, tensor):
        # Reverse of reshape_heads_to_batch_dim
        batch_size_times_heads, seq_len, dim_per_head = tensor.shape
        batch_size = batch_size_times_heads // self.num_heads
        tensor = tensor.reshape(batch_size, self.num_heads, seq_len, dim_per_head)
        tensor = tensor.permute(0, 2, 1, 3).reshape(batch_size, seq_len, self.num_heads * dim_per_head)
        return tensor

self_obj = MinimalAttention()

# Create hidden_states: (batch_size=2, sequence_length=3, hidden_dim=4)
hidden_states = torch.randn(2, 3, 4)

# Create context as a dict with all required keys
def weight_function(w, sigma, attention_scores):
    """Minimal weight function that returns a tensor compatible with attention_scores"""
    if isinstance(w, int):
        return 0.0
    return torch.zeros_like(attention_scores)

context = {
    'CONTEXT_TENSOR': torch.randn(2, 3, 4),
    'WEIGHT_FUNCTION': weight_function,
    'CROSS_ATTENTION_WEIGHT_ORIG': torch.randn(3, 3, 1),  # img_h=3, img_w=3, nc=1
    'SIGMA': 1.0
}

# mask parameter (not used in the function body, but it's a parameter)
mask = None

ARGS = {
    'self': self_obj,
    'hidden_states': hidden_states,
    'context': context,
    'mask': mask
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 3, 4) — shape (2, 3, 4) 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_checkpoint Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/SVR/utils.py
code served (permissive licence) · get_code("619771c3e74f395b")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. 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 torch
import torch.nn as nn
import tempfile
import os

torch.manual_seed(0)

# Create a minimal model
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create a minimal optimizer
model = MinimalModel()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

# Create a checkpoint file
checkpoint_data = {
    'state_dict': model.state_dict(),
    'optimizer': optimizer.state_dict(),
    'epoch': 5,
    'bestloss': 0.123
}

# Write checkpoint to a temporary file
with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.pt') as f:
    temp_checkpoint_path = f.name

torch.save(checkpoint_data, temp_checkpoint_path)

# Create fresh model and optimizer for loading
fresh_model = MinimalModel()
fresh_optimizer = torch.optim.SGD(fresh_model.parameters(), lr=0.01)

ARGS = {
    'cp_filename': temp_checkpoint_path,
    'model': fresh_model,
    'optimizer': fresh_optimizer
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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 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_model Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/SVR/utils.py
code served (permissive licence) · get_code("f04426f214fbb1ec")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. 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 torch
import torch.nn as nn
import tempfile
import os

torch.manual_seed(0)

# Create a minimal model
class TinyModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create model instance
model = TinyModel()

# Create a checkpoint with required keys
checkpoint = {
    'state_dict': model.state_dict(),
    'epoch': 42
}

# Save checkpoint to a temporary file
with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.pt') as f:
    cp_filename = f.name

torch.save(checkpoint, cp_filename)

try:
    ARGS = {
        'cp_filename': cp_filename,
        'model': model
    }
finally:
    # Clean up temp file after defining ARGS
    pass
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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 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
preprocess Ran
ran on a synthesized input
cloneofsimo/paint-with-words-sd/paint_with_words/paint_with_words.py
code served (permissive licence) · get_code("cc073aabc2f49dab")
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) and returned Tensor, ndim 4, dtype family float, shape (1, 3, 96, 96) 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 PIL
import torch

torch.manual_seed(0)

# Create a small test image using PIL
# The function expects a PIL Image with .size attribute and .resize method
image = PIL.Image.new('RGB', (100, 96))  # Width=100, Height=96 (both divisible by 32)
# Fill with some pixel data
pixels = image.load()
for i in range(100):
    for j in range(96):
        pixels[i, j] = (i % 256, j % 256, (i + j) % 256)

ARGS = {
    'image': image
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 Tensor, ndim 4, dtype family float, shape (1, 3, 96, 96) — shape (1, 3, 96, 96) 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
process_image Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/SVR/augment_ShapeNet_data.py
code served (permissive licence) · get_code("ce5593d971584146")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. 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 numpy as np
from PIL import Image
import tempfile

# Create a temporary directory for saving outputs
temp_dir = tempfile.mkdtemp()

# Create a small RGBA test image (4 channels: R, G, B, A)
# Size 100x100 to keep it small
width, height = 100, 100
image_array = np.zeros((height, width, 4), dtype=np.uint8)
# Fill RGB channels with some color
image_array[:, :, 0] = 100  # Red channel
image_array[:, :, 1] = 150  # Green channel
image_array[:, :, 2] = 200  # Blue channel
# Alpha channel: make it non-zero in the center region (opaque)
image_array[20:80, 20:80, 3] = 255  # Opaque center
image_array[:20, :, 3] = 0    # Transparent top
image_array[80:, :, 3] = 0    # Transparent bottom
image_array[:, :20, 3] = 0    # Transparent left
image_array[:, 80:, 3] = 0    # Transparent right

# Save the test image
test_image_path = os.path.join(temp_dir, "test_image.png")
test_image = Image.fromarray(image_array, mode='RGBA')
test_image.save(test_image_path)

ARGS = {
    "image_file": test_image_path,
    "save_path": temp_dir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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 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
sample_grid_points Ran
ran on a synthesized input
cross-domain-compositing/cross-domain-compositing/SVR/utils.py
code served (permissive licence) · get_code("08767bfaed57712a")
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 4, dtype family float, shape (3, 4, 5, 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

ARGS = {
    "xnum": 3,
    "ynum": 4,
    "znum": 5,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 4, dtype family float, shape (3, 4, 5, 3) — shape (3, 4, 5, 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
download_models Not yet run
unverified (harvested, no recorded run)
cross-domain-compositing/cross-domain-compositing/notebook_helpers.py
code served (permissive licence) · get_code("4533e43e06a12dc5")
How it ran

no invocation recorded.

get_blend_operator_and_mask Not yet run
unverified (harvested, no recorded run)
cross-domain-compositing/cross-domain-compositing/cdc/cdc.py
code served (permissive licence) · get_code("0657ed0cbf2c4fc5")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 2 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 2 later) and 11 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
download_models 17 other papers
17 later
Dombrowski_Foreground-Background_Separation_through_Concept_Distillation_from_Generative_Image_Foundation_Models_ICCV_2023_paper (2023 (year only))
Foreground-Background Separation through Concept Distillation from Generative Image Foundation Models
2303.00836 (2023-03)
Ensemble flow reconstruction in the atmospheric boundary layer from spatially limited measurements through latent diffusion models
2303.09618 (2023-03)
HIVE: Harnessing Human Feedback for Instructional Visual Editing
and 14 more not named here
get_schedule 1 other paper
1 later
2312.11121 (2023-12)
Multi-scale Reconstruction of Turbulent Rotating Flows with Generative Diffusion Models

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

Large-scale diffusion-based generative models have led to breakthroughs in text-conditioned high-resolution image synthesis. Starting from random noise, such text-to-image diffusion models gradually synthesize images in an iterative fashion while conditioning on text prompts. We find that their synthesis behavior qualitatively changes throughout this process: Early in sampling, generation strongly relies on the text prompt to generate text-aligned content, while later, the text conditioning is almost entirely ignored. This suggests that sharing model parameters throughout the entire generation process may not be ideal. Therefore, in contrast to existing works, we propose to train an ensemble of text-to-image diffusion models specialized for different synthesis stages. To maintain training efficiency, we initially train a single model, which is then split into specialized models that are trained for the specific stages of the iterative generation process. Our ensemble of diffusion models, called eDiff-I, results in improved text alignment while maintaining the same inference computation cost and preserving high visual quality, outperforming previous large-scale text-to-image diffusion models on the standard benchmark. In addition, we train our model to exploit a variety of embeddings for conditioning, including the T5 text, CLIP text, and CLIP image embeddings. We show that these different embeddings lead to different behaviors. Notably, the CLIP image embedding allows an intuitive way of transferring the style of a reference image to the target text-to-image output. Lastly, we show a technique that enables eDiff-I's "paint-with-words" capability. A user can select the word in the input text and paint it in a canvas to control the output, which is very handy for crafting the desired image in mind. The project page is available at https://deepimagination.cc/eDiff-I/

For agents

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

get_harvested_code_for_paper("2211.01324")
get_code_for_paper("2211.01324")
have("2211.01324")

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

curl https://syntology.ai/api/ran/2211.01324.json

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

[![Syntology run record](https://syntology.ai/api/ran/2211.01324.svg)](https://syntology.ai/paper/2211.01324)

Connect an agent — have() is free.