Perle-ai/T2I-benchmark

T2I Benchmark: Frontier Text-to-Image Models on Image Description Prompts

Accompanying dataset for Benchmarking Frontier Text-to-Image Models on the Image Description Prompts (Perle AI).

Four frontier text-to-image systems are compared on the 48 hardest prompts in the DataSeeds.AI Sample Dataset (DSD). Every prompt is a verbatim, human-written description of a real photograph — no prompt engineering. Every generated image is graded against a per-prompt weighted rubric by an independent-judge pipeline in which the model that writes the rubric never scores it.

This is a strict matched-N comparison: all four systems are scored on the identical 48 prompts.

Benchmark construction

1. Complexity scoring. All ~7,000 valid rows of the DSD training split were scored 1–10 for generation difficulty by an automated classifier, weighing factors such as exact object counting, legible text rendering, spatial binding, conditional logic, and attribute binding.
2. Hardest-48 selection. The highest-complexity prompts were kept; ties at the ceiling score were broken by a seeded random shuffle before a stable sort. The generation prompt is the verbatim concatenation of the DSD image_description and scene_description fields.
3. Generation. Each prompt was submitted identically to all four systems via their public APIs — single sample, default settings, no negative prompts, no seed, no best-of-k.
4. Rubric writing. For each prompt, GPT-5.4-Pro wrote a weighted rubric from the prompt text alone: positive criteria (what the prompt requests) and negative criteria (failure modes to avoid), each with an integer weight.
5. Independent scoring. Gemini 3.1 Pro Preview scored each generated image against that rubric with vision input. Each criterion gets a met/triggered verdict, a failure code, and a written explanation. final_score normalizes earned positive weight against incurred negative weight onto 0–100.
6. Matched-N aggregation. Only prompts where all four systems both generated and graded successfully are included (N = 48). Prompts declined by a provider's content-moderation system were excluded rather than retried or circumvented.

Failure-code taxonomy

MISCOUNT · TEXT_GARBLING · SPATIAL_RELATION_ERROR · ATTRIBUTE_BINDING_ERROR · MISSING_STEP · CONSTRAINT_BREACH · FORMAT_VIOLATION · INCOMPLETE_TRAJECTORY · GEOMETRY_ARTIFACT · HALLUCINATED_OBSERVATION · PHYSICS_VIOLATION

Positive-axis codes mark a request the image failed to satisfy; negative-axis codes mark an active defect the image introduced.

Results

Identical 48 prompts, all four systems. Net score is the mean unnormalized weighted balance (earned − incurred); unmet criteria and active errors are mean counts per prompt.

Rank	Model	N	Final score (0–100)	Net score	Unmet criteria	Active errors
1	Gemini 3 Pro Image (Nano Banana Pro)	48	84.8	+44.3	0.5	0.8
2	Black Forest Labs FLUX.2	48	82.3	+42.6	0.5	1.0
3	Ideogram 3.0	48	65.7	+34.1	1.4	1.5
4	Hunyuan 3.0	48	63.3	+33.3	1.4	1.7

Top failure codes on the matched-48 set

Model	Missed requests (positive axis)	Active errors (negative axis)
Gemini 3 Pro Image	MISCOUNT (9), SPATIAL_RELATION_ERROR (7), MISSING_STEP (2), TEXT_GARBLING (2), CONSTRAINT_BREACH (2)	TEXT_GARBLING (16), GEOMETRY_ARTIFACT (12), HALLUCINATED_OBSERVATION (9), PHYSICS_VIOLATION (3)
FLUX.2	MISCOUNT (8), MISSING_STEP (4), TEXT_GARBLING (3), SPATIAL_RELATION_ERROR (3), CONSTRAINT_BREACH (2)	TEXT_GARBLING (19), GEOMETRY_ARTIFACT (16), HALLUCINATED_OBSERVATION (7), PHYSICS_VIOLATION (7)
Ideogram 3.0	MISSING_STEP (24), TEXT_GARBLING (12), MISCOUNT (10), SPATIAL_RELATION_ERROR (9), CONSTRAINT_BREACH (5)	TEXT_GARBLING (27), GEOMETRY_ARTIFACT (24), HALLUCINATED_OBSERVATION (12), PHYSICS_VIOLATION (5)
Hunyuan 3.0	TEXT_GARBLING (20), MISCOUNT (12), MISSING_STEP (10), SPATIAL_RELATION_ERROR (10), CONSTRAINT_BREACH (5)	TEXT_GARBLING (29), GEOMETRY_ARTIFACT (27), HALLUCINATED_OBSERVATION (15), PHYSICS_VIOLATION (6)

Legible text rendering is the single most common active defect for every system, and exact counting is the most common missed request for the two leaders.

Configs

prompts — 48 rows

Field	Type	Description
row_idx	int	Row index in the DSD training split
image_id	string	DSD image UUID — join key to the source dataset's reference photograph
image_title	string	DSD image title
image_description	string	Verbatim DSD field
scene_description	string	Verbatim DSD field
prompt	string	image_description + scene_description — the exact string submitted to every model
complexity_score	int	1–10 difficulty score
complexity_rationale	string	Classifier's reasoning for the score
hardness_factors	list[string]	Named difficulty factors, e.g. ["Precise counting (three cabins)", "Legible text generation (Disel-Bensin)"]

evaluations — 192 rows (48 prompts × 4 models)

One fully graded rubric per model per prompt: 2,658 individual criterion verdicts in total.

Field	Type	Description
model_key / model_display_name	string	Candidate system
row_idx / image_id	int / string	Join keys to prompts
rubric_writer	string	gpt5pro — GPT-5.4-Pro wrote the rubric
rubric_scorer	string	gemini — Gemini 3.1 Pro Preview scored the image
positive_criteria	list[struct]	{statement, weight, met, failure_code, explanation}
negative_criteria	list[struct]	{statement, weight, triggered, failure_code, explanation}
total_positive_weight_earned / _possible	int	Positive-axis totals
total_negative_weight_incurred	int	Negative-axis total
failure_count	int	Unmet positive criteria
active_error_count	int	Triggered negative criteria
severity_summary	string	Scorer's one-line summary of the image
net_score	float	Raw weighted balance (earned − incurred)
final_score	float	Normalized 0–100

Usage

from datasets import load_dataset

prompts = load_dataset("Perle-ai/T2I-benchmark", "prompts", split="test")
evals   = load_dataset("Perle-ai/T2I-benchmark", "evaluations", split="test")

# Reproduce the leaderboard
import collections
scores = collections.defaultdict(list)
for e in evals:
    scores[e["model_display_name"]].append(e["final_score"])
for name, s in sorted(scores.items(), key=lambda kv: -sum(kv[1]) / len(kv[1])):
    print(f"{name:40s} N={len(s)}  {sum(s) / len(s):.1f}")

# Failure-code frequencies for one model
codes = collections.Counter(
    c["failure_code"]
    for e in evals if e["model_key"] == "hunyuan3"
    for c in e["negative_criteria"] if c["triggered"] and c["failure_code"]
)
print(codes.most_common())

Reference photographs for every prompt are available in the source DSD dataset by image_id.

Evaluated systems

Gemini 3 Pro Image (Nano Banana Pro) · Black Forest Labs FLUX.2 · Ideogram 3.0 · Hunyuan 3.0 (HunyuanImage-3.0)

Limitations

One generation per prompt per model with no fixed seed and no resampling, so scores reflect default out-of-the-box quality rather than best-of-k. Grades come from a single automated judge pair (GPT-5.4-Pro → Gemini 3.1 Pro Preview); human-SME agreement has not yet been measured. Prompt selection is deliberately biased toward maximum difficulty, so absolute scores are not comparable to general-purpose prompt sets, and N = 48 bounds the resolution of small between-model gaps.

License and provenance

Prompts are derived verbatim from the DataSeeds.AI Sample Dataset (DSD), released under Apache-2.0. Rubrics, criterion verdicts, scores, and metadata in this repository are released under Apache-2.0.

Citation

@misc{perle2026t2ibenchmark,
  title        = {Benchmarking Frontier Text-to-Image Models on the Image Description Prompts},
  author       = {Perle AI},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/Perle-ai/T2I-benchmark}}
}

Please also cite the source dataset:

@misc{dataseeds2025,
  title  = {Peer-Ranked Precision: Creating a Foundational Dataset for Fine-Tuning Vision Models from GuruShots' Annotated Imagery},
  author = {Abdoli, Sajjad and others},
  year   = {2025},
  eprint = {2506.05673},
  archivePrefix = {arXiv}
}