SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2405.02793 · EMNLP · 2024

ImageInWords: Unlocking Hyper-Detailed Image Descriptions

Andrea Burns, Jason Baldridge, Yonatan Bitton, Burcu Karagol, Radu Soricut, Ranjay Krishna, Yasumasa Onoe, Roopal Garg, Ceslee Montgomery, Andrew Bunner

arXiv · PDF · Open in the Atlas

Code that ran

8 of 10 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; 0 honoured the contract we drafted, 0 violated it and 8 have no contract verdict; 0 failures 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/2405.02793.json · badge: /api/ran/2405.02793.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
google-deepmind/proactive_t2i_agents community (classifier label)
classifier basis not recorded on the link
8 ran · 2 no recorded run
FunctionStatusWhere it lives
entity_parser_prompt Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/prompts.py
code served (permissive licence) · get_code("6fdfee8ae85731ff")
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 str. The image it ran in was not recorded. 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 textwrap

# The function takes a user_prompt string as input
# Looking at the function body, it simply formats this string into a detailed prompt template
# using textwrap.dedent and f-string formatting

ARGS = {
    "user_prompt": "a red car parked in front of a modern building with blue sky"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
entity_parser_prompt_w_background Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/prompts.py
code served (permissive licence) · get_code("96bd654b5fe8cafd")
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 str. The image it ran in was not recorded. 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 textwrap

# The function only takes one parameter: user_prompt (a string)
# Looking at the function body, it simply takes the user_prompt and inserts it
# into a formatted string using f-string and textwrap.dedent()

user_prompt = "generate an image of a lionhead rabbit running on grass with sun shining. There is no trees in the background."

ARGS = {
    "user_prompt": user_prompt
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
entropy Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/agent_classes.py
code served (permissive licence) · get_code("ac24c3867c0d6c29")
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 float64, ndim 0, 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 numpy as np

# The entropy function takes a list of floats (probabilities)
# and computes entropy: -sum(p * log(p) for p > 0)

# Create a valid probability distribution (sums to 1, all non-negative)
probs = [0.2, 0.3, 0.3, 0.2]

ARGS = {
    "probs": probs
}
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 float64, ndim 0, dtype family float — no output shape is stated: observe_signature.py recorded float ndim 0, which disagrees with the fixture run
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
is_confident Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/belief_utils.py
code served (permissive licence) · get_code("4337cf11cd0ec49c")
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 bool, ndim 0, dtype family float, 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

class Candidate:
    def __init__(self, probability):
        self.probability = probability

class Attribute:
    def __init__(self, importance_score, value):
        self.importance_score = importance_score
        self.value = value

class Entity:
    def __init__(self, importance_score, attributes):
        self.importance_score = importance_score
        self.attributes = attributes

class Relation:
    def __init__(self, importance_score, value):
        self.importance_score = importance_score
        self.value = value

class BeliefState:
    def __init__(self, all_entities, all_relations):
        self.all_entities = all_entities
        self.all_relations = all_relations

# Create a belief state with entities, attributes, and relations
# All probabilities should be confident (either very high or very low)
candidate1 = Candidate(probability=0.95)
candidate2 = Candidate(probability=0.05)

attribute1 = Attribute(importance_score=0.5, value=[candidate1])
attribute2 = Attribute(importance_score=0.5, value=[candidate2])

entity1 = Entity(importance_score=0.5, attributes=[attribute1, attribute2])

relation1 = Relation(importance_score=0.5, value=[Candidate(probability=0.92)])

belief_state = BeliefState(
    all_entities=[entity1],
    all_relations=[relation1]
)

ARGS = {
    "belief_state": belief_state,
    "confidence_threshold": 0.9,
    "importance_threshold": 0.1
}
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 bool, ndim 0, dtype family float, 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
load_img Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/image_utils.py
code served (permissive licence) · get_code("912864da44ac4f8d")
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 PngImageFile. The image it ran in was not recorded. 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 PIL.Image
import tempfile
import os

# Create a temporary directory and a simple test image
temp_dir = tempfile.mkdtemp()
temp_image_path = os.path.join(temp_dir, "test_image.png")

# Create a simple PIL Image and save it
img = PIL.Image.new('RGB', (10, 10), color='red')
img.save(temp_image_path)

ARGS = {
    'path': temp_image_path
}
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 PngImageFile — a PngImageFile output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PngImageFile output the recorded ndim 3 and dtype family int describe that coercion, not the output, and are not stated
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
merge_info_prompt Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/prompts.py
code served (permissive licence) · get_code("9969f5aeb6be4c91")
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 str. The image it ran in was not recorded. 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 numpy as np

torch.manual_seed(0)

ARGS = {
    "prompt": "A serene landscape with mountains and a lake",
    "additional_info": "The sky should be clear and blue, without any clouds"
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
most_likely_deterministic_state Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/belief_utils.py
code served (permissive licence) · get_code("2d562b2e2b98bf22")
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 BeliefState. The image it ran in was not recorded. 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 copy

class Candidate:
    def __init__(self, value, probability):
        self.value = value
        self.probability = probability

class Attribute:
    def __init__(self, name, value_candidates):
        self.name = name
        self.value = value_candidates

class Relation:
    def __init__(self, name, value_candidates):
        self.name = name
        self.value = value_candidates

class Entity:
    def __init__(self, entity_id, probability, attributes):
        self.entity_id = entity_id
        self.probability = probability
        self.attributes = attributes

class BeliefState:
    def __init__(self, all_entities, all_relations):
        self.all_entities = all_entities
        self.all_relations = all_relations

belief_state = BeliefState(
    all_entities=[
        Entity(
            entity_id=0,
            probability=0.95,
            attributes=[
                Attribute("color", [
                    Candidate("red", 0.8),
                    Candidate("blue", 0.2)
                ]),
                Attribute("size", [
                    Candidate("large", 0.75),
                    Candidate("small", 0.25)
                ])
            ]
        ),
        Entity(
            entity_id=1,
            probability=0.65,
            attributes=[
                Attribute("shape", [
                    Candidate("circle", 0.9),
                    Candidate("square", 0.1)
                ])
            ]
        )
    ],
    all_relations=[
        Relation("next_to", [
            Candidate("entity_0", 0.85),
            Candidate("entity_1", 0.15)
        ]),
        Relation("above", [
            Candidate("entity_1", 0.6),
            Candidate("entity_0", 0.4)
        ])
    ]
)

ARGS = {
    "belief_state": belief_state,
    "remove_uncertain_entities": True,
    "remove_uncertain_attributes": True,
    "prob_threshold": 0.7
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BeliefState — a BeliefState output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a BeliefState output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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
show_image Ran
ran on a synthesized input
google-deepmind/proactive_t2i_agents/agent/image_utils.py
code served (permissive licence) · get_code("dce5b9144b3351bc")
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 AxesImage. The image it ran in was not recorded. 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 numpy as np
from PIL import Image

torch.manual_seed(0)

# Create a small PIL Image object
# show_image expects a PIL Image that can be displayed with plt.imshow()
image_array = np.random.randint(0, 256, (4, 3, 3), dtype=np.uint8)
image_pil = Image.fromarray(image_array, mode='RGB')

ARGS = {
    'image_pil': image_pil
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, matplotlib, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type AxesImage — a AxesImage output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a AxesImage output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
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_image_description Not yet run
unverified (harvested, no recorded run)
google-deepmind/proactive_t2i_agents/agent/image_utils.py
code served (permissive licence) · get_code("7d93f6d7eb593350")
How it ran

no invocation recorded.

negative_log_likelihood Not yet run
unverified (harvested, no recorded run)
google-deepmind/proactive_t2i_agents/agent/belief_metrics.py
code served (permissive licence) · get_code("b9dcd4f5f46998b9")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 10 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.

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

Despite the longstanding adage "an image is worth a thousand words," generating accurate hyper-detailed image descriptions remains unsolved. Trained on short web-scraped imagetext, vision-language models often generate incomplete descriptions with visual inconsistencies. We address this via a novel data-centric approach with ImageInWords (IIW), a carefully designed human-in-the-loop framework for curating hyper-detailed image descriptions. Human evaluations on IIW data show major gains compared to recent datasets (+66%) and GPT-4V (+48%) across comprehensiveness, specificity, hallucinations, and more. We also show that fine-tuning with IIW data improves these metrics by +31% against models trained with prior work, even with only 9k samples. Lastly, we evaluate IIW models with text-to-image generation and vision-language reasoning tasks. Our generated descriptions result in the highest fidelity images, and boost compositional reasoning by up to 6% on ARO, SVO-Probes, and Winoground datasets. We release the IIW-Eval benchmark with human judgement labels, object and image-level annotations from our framework, and existing image caption datasets enriched via IIW-model.

For agents

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

get_harvested_code_for_paper("2405.02793")
get_code_for_paper("2405.02793")
have("2405.02793")

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

curl https://syntology.ai/api/ran/2405.02793.json

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

[![Syntology run record](https://syntology.ai/api/ran/2405.02793.svg)](https://syntology.ai/paper/2405.02793)

Connect an agent — have() is free.