Few-step text-to-image models increasingly replace slower generators, yet acceleration can silently change distributions over unspecified attributes even when individual outputs remain plausible and aligned. We call these distributions semantic defaults and their change under replacement semantic default shift. Existing quality, preference, and diversity evaluations do not test whether a replacement preserves its reference model's semantic defaults. We introduce DefaultShift, a paired audit that labels repeated samples with closed semantic vocabularies, measures probability-mass movement, and separates interpretable ranking from confirmatory cross-fit inference. Across 14 reference and replacement pairs, adjusted color discrepancies range from 0.054 to 0.303 with recipe-specific directions. A 1,000-image human audit reproduces the ordering. We further introduce DefaultShift-Select, an offline calibration method that reduces human-measured shift by 10.3 percent to 35.1 percent across Turbo, DMD2, and FLUX without material quality loss. Under balanced evaluation, selected data recover 4.3 accuracy points and 7.5 worst-group points over uncalibrated replacement data. DefaultShift makes semantic preservation under acceleration measurable and actionable.
Figures & tables
Figure 1: The same prompt can induce different semantic defaults. A reference model and its accelerated replacement can produce individually plausible outputs while following different semantic distributions. DefaultShift measures both the magnitude and direction of this change.
Figure 2: Overview of DefaultShift. The audit generates matched samples from a declared reference and accelerated replacement, maps each image to a closed semantic vocabulary, and compares their conditional distributions. Cell-level discrepancies are aggregated for recipe ranking and directional analysis. DefaultShift-Select uses reference statistics and a quality-filtered replacement pool to construct a distribution-matched offline subset.
Figure 3: Acceleration recipes differ in magnitude, attribute, and direction. a The unified q=64 audit covers all 14 replacement pairs. Lines are 95% object-cluster intervals. b Color dominates the four-attribute profile, while viewpoint shifts are small. c Replacement-minus-reference changes in gray and warm mass reveal opposing recipe fingerprints.
Recipe
q=32
q=64
95% CI
Recipe
q=32
q=64
95% CI
NitroFusion-4
0.269
0.303
[0.252, 0.354]
TDM-SD3
0.155
0.170
[0.136, 0.205]
Turbo
0.226
0.254
[0.215, 0.293]
PixArt-LCM
0.107
0.120
[0.092, 0.148]
DMD2
0.210
0.235
[0.196, 0.273]
Flash-SDXL
0.101
0.114
[0.088, 0.141]
NitroFusion-1
0.205
0.228
[0.184, 0.273]
Hyper-SD15
0.073
0.079
[0.061, 0.098]
FLUX-schnell
0.192
0.203
[0.169, 0.239]
SD1.5-LCM
0.060
0.066
[0.050, 0.083]
SwD-FLUX
0.171
0.187
[0.150, 0.225]
Lightning
0.040
0.060
[0.044, 0.077]
Table 1: Unified color audit. The q=32 screen and q=64 endpoint use identical cells. Brackets give 95% object-cluster intervals for the q=64 endpoint.
Figure 4: Reliability and specificity. a Query-budget convergence on a 24-object plain-prompt diagnostic, not the main endpoint. b Explicitly specifying color suppresses shift across five recipes. c Conventional metrics explain little object-level color shift. Bands denote 95% cell-bootstrap intervals.
Figure 5: Reference-guided offline calibration. a DefaultShift-Select approaches a candidate-pool oracle as the budget grows. b Improvements remain positive under an independent evaluator and human labels.
Estimator
Null bias
Type I error
Rank ρ
Raw TV
0.197
0.980
0.991
TVadj
0.008
0.560
0.996
Signed cross-fit
0.001
0.052
0.981
Dirichlet posterior
0.034
0.071
0.989
Table 2: Estimator simulation at q=64 . Null calibration and alternative ranking are computed from frozen multinomial families.
Attribute
Clear n
Cov.
Acc.
Macro-F1
Fleiss κ
Color
261
0.87
0.91
0.91
0.71
Background
246
0.82
0.87
0.87
0.64
Table 3: Independent human validation. Each attribute is judged on 300 images. Clear n excludes abstentions, while coverage retains them in the denominator. Accuracy and macro-F1 compare Qwen with majority vote.
Figure 6: Independent human labels reproduce recipe separation. VLM circles and human diamonds show 95% intervals. Ordering is compared because label granularity differs.
Training source
Accuracy
Worst group
Reference
78.4 [76.9, 79.8]
70.1 [67.2, 73.0]
Replacement
72.1 [70.5, 73.6]
59.8 [56.5, 63.1]
Quality-only
72.6
61.2
DefaultShift-Select
76.4 [74.9, 77.9]
67.3 [64.2, 70.3]
Teacher-first hybrid
76.9
68.2
Table 4: Downstream classifier under balanced evaluation. Intervals are bootstrap 95% intervals.
Table 5: Evidence map for the claims in the main paper. Every supplementary entry uses the same terminology and numerical convention as the main text.
Reference family
Replacement
Resolution
Steps
CFG or recipe setting
SDXL
NitroFusion-Realism-4step
1024
30 to 4
5 to official recipe
SDXL
Turbo
512
30 to 4
5 to 0
SDXL
DMD2
1024
30 to 4
5 to 0, LCM scheduler
SDXL
NitroFusion-Realism-1step
1024
30 to 1
5 to official recipe
SDXL
Flash-SDXL
1024
30 to 4
5 to 0, LCM scheduler
SDXL
Lightning
768
30 to 4
5 to official recipe
Table 6: Generation configurations for the 14 reference and replacement pairs. Entries in the final column report reference to replacement settings.
Replacement pair
q=32
q=64
Raw TV
Cross-fit
Unknown
95% CI
SDXL to NitroFusion-Realism-4step
0.269
0.303
0.454
0.271
0.031
[0.252, 0.354]
SDXL to Turbo
0.226
0.254
0.402
0.228
0.028
[0.215, 0.293]
SDXL to DMD2
0.210
0.235
0.384
0.209
0.026
[0.196, 0.273]
SDXL to NitroFusion-Realism-1step
0.205
0.228
0.378
0.203
0.030
[0.184, 0.273]
FLUX.1-dev to FLUX.1-schnell
0.192
0.203
0.346
0.181
0.021
[0.169, 0.239]
FLUX.1-dev to SwD-FLUX
0.171
0.187
0.331
0.166
0.023
[0.150, 0.225]
Table 7: Complete unified color benchmark. The q=32 and q=64 columns use identical cells. Confidence intervals, raw TV, signed cross-fit TV, and unknown rates correspond to the q=64 endpoint.
Pair
Color
Background
Lighting
Viewpoint
NitroFusion-4
0.303 [0.252, 0.354]
0.115 [0.087, 0.145]
0.090 [0.066, 0.116]
0.030 [0.014, 0.047]
Turbo
0.254 [0.215, 0.293]
0.100 [0.078, 0.123]
0.080 [0.058, 0.103]
0.028 [0.013, 0.044]
DMD2
0.235 [0.196, 0.273]
0.086 [0.066, 0.108]
0.066 [0.046, 0.087]
0.022 [0.008, 0.037]
FLUX-schnell
0.203 [0.169, 0.239]
0.060 [0.038, 0.083]
0.050 [0.031, 0.070]
0.018 [0.004, 0.033]
Lightning
0.060 [0.044, 0.077]
0.030 [0.018, 0.043]
0.022 [0.010, 0.035]
0.009 [-0.003, 0.021]
LCM
0.054 [0.040, 0.069]
0.028 [0.015, 0.041]
0.024 [0.011, 0.038]
0.008 [-0.004, 0.020]
Table 8: Confirmatory q=64 results across four semantic attributes. Every entry reports TVadj followed by its 95% object-cluster interval.
Figure 7: A concrete q=64 semantic marginal for the exact prompt a photo of a car . Both FLUX models produce plausible cars, while the reference and accelerated color histograms differ in category identity and probability mass. This cell-level view complements the aggregate benchmark.
Comparison
TVadj
95% CI
SDXL to DMD2-4step
0.249
[0.221, 0.276]
DMD2-4step to NitroFusion-4step
0.163
[0.141, 0.187]
SDXL to NitroFusion-4step
0.269
[0.239, 0.298]
DMD2-4step to NitroFusion-1step
0.170
[0.146, 0.195]
SDXL to NitroFusion-1step
0.206
[0.182, 0.231]
Table 9: Lineage-aware NitroFusion analysis over 192 cells at q=32 .
Figure 8: Supplementary robustness summary. a The 24-object plain-prompt diagnostic supports q=64 for confirmatory estimation. b Explicitly naming color sharply reduces drift. c Aggressive-minus-mild separation persists across five prompt suites. d Recipe rankings remain stable across distribution distances and signed cross-fit TV. Error bars and bands give 95% intervals.
q
Turbo
LCM
Image slots
8
0.257 [0.188, 0.321]
0.032 [-0.019, 0.080]
576
16
0.279 [0.213, 0.349]
0.031 [-0.008, 0.075]
1,152
24
0.301 [0.246, 0.354]
0.049 [0.014, 0.084]
1,728
32
0.302 [0.243, 0.358]
0.052 [0.018, 0.087]
2,304
48
0.315 [0.261, 0.370]
0.046 [0.016, 0.079]
3,456
64
0.330 [0.277, 0.382]
0.038 [0.011, 0.066]
4,608
Table 10: Prefix-seed convergence on 24 plain-prompt objects. Brackets give 95% cell-bootstrap intervals.
Estimator
Null bias
Type I
Coverage
Alt. bias
Rank ρ
Raw empirical TV
0.197
0.980
0.000
+0.071
0.991
TVadj
0.008
0.560
0.440
-0.034
0.996
Signed cross-fit
0.001
0.052
0.947
-0.029
0.981
Dirichlet shrinkage
0.034
0.071
0.918
+0.004
0.989
Permutation test
n/a
0.049
n/a
n/a
n/a
Multinomial LRT
n/a
0.058
n/a
n/a
n/a
Table 11: Estimator operating characteristics at q=64 .
q
Type I
Coverage
Power
8
0.094
0.881
0.312
16
0.071
0.912
0.518
32
0.058
0.934
0.746
64
0.052
0.947
0.921
96
0.050
0.951
0.968
Table 12: Signed cross-fit calibration and power across query budgets for a true TV of 0.15 .
Pair
Primary
Unknown class
Coverage adjusted
Ref. unknown
Repl. unknown
95% CI
NitroFusion-4
0.115
0.190
0.125
0.100
0.060
[0.087, 0.145]
Turbo
0.100
0.185
0.113
0.125
0.100
[0.078, 0.123]
DMD2
0.086
0.176
0.095
0.125
0.055
[0.066, 0.108]
FLUX-schnell
0.060
0.090
0.062
0.024
0.037
[0.038, 0.083]
Lightning
0.030
0.095
0.034
0.125
0.095
[0.018, 0.043]
LCM
0.028
0.115
0.031
0.125
0.073
[0.015, 0.041]
Table 13: Background estimates under three treatments of unknown labels.
Table 15: Matched plain-prompt negative controls over 20 prompts and 32 seeds. The specified prompt inserts only the frozen color token.
Figure 9: Complete clear-case confusion matrices. Rows are human majority labels and columns are VLM predictions. Remaining errors concentrate among nearby color groups and between urban, nature, and abstract backgrounds.
Attribute
Class
Support
Precision
Recall
F1
Attribute
Class
Support
Precision
Recall
F1
Color
white
42
0.93
0.93
0.93
Background
plain or studio
58
0.87
0.90
0.88
Color
black
33
0.91
0.91
0.91
Background
indoor
52
0.94
0.87
0.90
Color
gray
38
0.85
0.87
0.86
Background
outdoor nature
54
0.89
0.89
0.89
Color
blue
36
0.97
0.94
0.96
Background
outdoor urban
44
0.80
0.84
0.82
Color
red
28
0.93
0.89
0.91
Background
abstract pattern
38
0.84
0.84
0.84
Color
green
24
0.96
0.96
0.96
Background
macro average
246
0.87
0.87
0.87
Table 16: Per-class VLM performance against human majority vote.
Statistic
Color
Background
Fleiss κ , all labels
0.71 [0.65, 0.77]
0.64 [0.57, 0.71]
Fleiss κ , clear cases
0.76
0.70
Fleiss κ , coarse groups
0.81
0.75
Krippendorff α
0.70
0.63
Mean pairwise agreement
0.83
0.78
Three-of-three agreement
0.62
0.54
Table 17: Inter-rater agreement. Confidence intervals refer to Fleiss κ with all legal labels retained.
Attribute
Quantity
Estimate
Difference
95% CI
Color
Reference split-half floor
0.16
n/a
[0.11, 0.21]
Color
Raw reference to Turbo
0.34
n/a
[0.26, 0.42]
Color
Raw reference to LCM
0.26
n/a
[0.19, 0.34]
Color
Bias-reduced reference to Turbo
0.18
n/a
[0.09, 0.27]
Color
Bias-reduced reference to LCM
0.10
n/a
[0.02, 0.19]
Color
Turbo minus LCM
n/a
0.08
[0.03, 0.13]
Table 18: Human grouped-label distribution distances with image-level paired bootstrap.
Pair
VLM q=64
Human
Human minus VLM
Human 95% CI
Direction
NitroFusion-4
0.303
0.246
-0.057
[0.158, 0.334]
agrees
Turbo
0.254
0.208
-0.046
[0.124, 0.293]
agrees
DMD2
0.235
0.193
-0.042
[0.108, 0.279]
agrees
FLUX-schnell
0.203
0.168
-0.035
[0.082, 0.254]
agrees
LCM
0.054
0.051
-0.003
[-0.030, 0.133]
mild pair swaps
Lightning
0.060
0.047
-0.013
[-0.034, 0.129]
mild pair swaps
Table 19: Human reproduction of the six-pair ranking. Human labels use five grouped color categories.
Recipe
Δ Gray
Δ Warm
Color TVadj
DMD2
-0.124
+0.159
0.235
Turbo
-0.069
+0.082
0.254
LCM
-0.044
+0.046
0.054
Lightning
-0.038
+0.048
0.060
FLUX-schnell
+0.019
-0.070
0.203
Table 20: Signed color fingerprints. Changes are replacement minus reference probability mass.
Figure 10: Selected and excluded candidates for the prompt a photo of a helicopter . Reference samples define the target. Real FLUX.1-schnell candidates are separated into selected, quality-eligible but unselected, and below-floor outcomes. Selection changes raw TV from 0.406 to 0.000 for this object. Aggregate evidence uses the held-out 24-object evaluation in Table 21 .
Method
Turbo
DMD2
FLUX
Random
0.0%
0.0%
0.0%
Quality-only
1.2%
0.9%
3.1%
DINO k-center
4.8%
5.3%
8.4%
Semantic entropy
7.9%
7.2%
15.3%
Rejection sampling
10.4%
11.8%
29.6%
Quota sampling
12.7%
15.1%
38.9%
Table 21: Cross-fit TV reduction from the shared fourfold candidate pool.
Recipe
Primary VLM
Second VLM
Human
Human 95% CI
Turbo
14.8%
12.1%
10.3%
[2.9, 17.8]
DMD2
16.8%
13.9%
11.7%
[3.8, 20.6]
FLUX
44.8%
38.2%
35.1%
[24.3, 46.2]
Table 22: Select reduction under independent evaluators and human labels.
Metric
Margin
Change
95% CI
Result
CLIP score
-0.010
+0.001
[-0.003, 0.005]
passes
Aesthetic score
-0.100
-0.015
[-0.070, 0.040]
passes
PickScore
-0.010
0.000
[-0.004, 0.004]
passes
Human quality preference
-0.050
-0.010
[-0.040, 0.020]
passes
Within-color DINO diversity
-5%
-2.1%
[-4.3, 0.8]
passes
Within-color Vendi
-5%
-1.8%
[-4.0, 1.1]
passes
Table 23: Quality noninferiority. A result passes when the interval lower bound remains above the frozen negative margin.
Training source
Balanced accuracy
95% CI
Worst-color accuracy
95% CI
ECE
Reference
78.4%
[76.9, 79.8]
70.1%
[67.2, 73.0]
0.061
Replacement random
72.1%
[70.5, 73.6]
59.8%
[56.5, 63.1]
0.094
Quality-only
72.6%
[71.0, 74.1]
61.2%
[57.9, 64.4]
0.089
DefaultShift-Select
76.4%
[74.9, 77.9]
67.3%
[64.2, 70.3]
0.071
Teacher-first hybrid
76.9%
[75.4, 78.3]
68.2%
[65.2, 71.1]
0.068
Table 24: Complete downstream results under balanced evaluation.
Figure 12: Illustrative reference and accelerated outputs for bird, cat, and dog prompts. Each row contains six experimental images. Samples are displayed to make semantic concentration visually inspectable. Quantitative conclusions use the complete frozen label banks rather than this curated subset.
Role
Identifier
Cached revision
Primary VLM
Qwen/Qwen2.5-VL-3B-Instruct
66285546d2b821cf421d4f5eb2576359d3770cd3
SDXL reference
stabilityai/stable-diffusion-xl-base-1.0
462165984030d82259a11f4367a4eed129e94a7b
Turbo
stabilityai/sdxl-turbo
71153311d3dbb46851df1931d3ca6e939de83304
LCM adapter
latent-consistency/lcm-lora-sdxl
a18548dd4956b174ec5b0d78d340c8dae0a129cd
DMD2 adapter
tianweiy/DMD2
be22767697a1f3ca656b73c776e15fa335c86c6c
FLUX reference
black-forest-labs/FLUX.1-dev
3de623fc3c33e44ffbe2bad470d0f45bccf2eb21
Table 25: Principal model identifiers and cached revisions.
Diffusion models have consistently driven progress in text-to-image generation. However, it is challenging to attribute recent progress to specific modeling and data choices: state-of-the-art open-weight models provide limited ablations, and do not disclose their training data and full training details. The research community needs fully open (weights, data, and code) models as a foundation for further research; yet existing fully open models still fall significantly short of leading models in performance. In this project, we conduct a systematic investigation of the modeling and data design choices in text-to-image diffusion training and inference with 300+ controlled experiments totaling 700K+ TPU v6e hours. Our experiments highlight several empirical findings (e.g., equal weighting is a strong default for mixing curated datasets) and simple design decisions (e.g., larger text encoder adapters improve performance with minimal added parameters) for training strong models. Guided by these insights, we train i1, a 3B-parameter text-to-image diffusion model using only publicly available datasets. i1 is competitive with leading models on five representative benchmarks (GenEval, DPG, PRISM, CVTG-2K, and LongText), and outperforms the best existing fully open model by 29.5 absolute percentage points on average. We provide the i1 checkpoints, training and inference code, and the data processing pipeline. Together, our findings and the i1 recipe establish a practical foundation for future open research in text-to-image diffusion models. Our code is available at https://github.com/zlab-princeton/i1.
Text-to-image models are typically reported on average-case prompts, which understates the gap between systems on compositionally demanding requests involving precise object counts, multi-object attribute binding, legible embedded text, and explicit spatial constraints. We evaluate four production text-to-image systems: Hunyuan 3.0, Gemini 3 Pro Image ("Nano Banana Pro"), Black Forest Labs FLUX.2, and Ideogram 3.0. The evaluation uses the 48 hardest prompts drawn from the DataSeeds.AI Sample Dataset (DSD), selected through an automated complexity-scoring pass over the full corpus. Every generated image is graded using an independent-judge rubric. GPT-5.4-Pro authors an atomic, weighted, mutually exclusive and collectively exhaustive (MECE) evaluation rubric, while Gemini 3.1 Pro Preview independently determines whether each criterion is satisfied. Gemini 3 Pro Image ranks first with a score of 84.8/100, narrowly ahead of FLUX.2 at 82.3/100. Ideogram 3.0 and Hunyuan 3.0 score 65.7/100 and 63.3/100, respectively. Failure analysis shows that the leading systems primarily lose points through object miscounting and geometric artifacts, whereas the trailing systems more frequently produce garbled text. Ideogram 3.0 also frequently omits requested elements. Full per-sample rubrics, scores, and failure annotations are available from the authors upon request.
Text-to-Image (T2I) generation is primarily driven by Diffusion Models (DM) which rely on random Gaussian noise. Thus, like playing the slots at a casino, a DM will produce different results given the same user-defined inputs. This imposes a gambler's burden: To perform multiple generation cycles to obtain a satisfactory result. However, even though DMs use stochastic sampling to seed generation, the distribution of generated content quality highly depends on the prompt and the generative ability of a DM with respect to it. To account for this, we propose Naïve PAINE for improving the generative quality of Diffusion Models by leveraging T2I preference benchmarks. We directly predict the numerical quality of an image from the initial noise and given prompt. Naïve PAINE then selects a handful of quality noises and forwards them to the DM for generation. Further, Naïve PAINE provides feedback on the DM generative quality given the prompt and is lightweight enough to seamlessly fit into existing DM pipelines. Experimental results demonstrate that Naïve PAINE outperforms existing approaches on several prompt corpus benchmarks.
Joong Ho Kim, Nicholas Thai, Souhardya Saha Dip +2