<report>

Black Forest Labs

FLUX 3

v-benchmark v2 by Megaton benchmark
Evaluated · August 11, 2026Leaderboard rank #3 of 11Last updated · August 11, 2026

76.51

/ 100

Categories

C01

Prompt AdherenceWhether every explicit, observable instruction in the prompt is fulfilled.

This Video

100.0

This Video · Assessment

This video fulfills its explicit, observable instructions with near-complete consistency.

This Category

87.1

60 videos were rated for this category

This Category · Assessment

Prompt requirements are fulfilled reliably across the set, with only isolated misses or ambiguities.

Models that scored similarly on this category

C02

Scene ConsistencyWhether established subjects, objects, spatial relationships, environment, and lighting remain coherent over time.

This Video

93.1

This Video · Assessment

This video maintains subjects, objects, spatial relationships, environment, and lighting over time with near-complete consistency.

This Category

94.8

60 videos were rated for this category

This Category · Assessment

Subjects, objects, spatial relationships, environment, and lighting remain coherent across nearly all evaluated videos.

Models that scored similarly on this category

C03

PhysicsWhether the central motion, force, material, and reaction behave coherently in the requested world and style.

This Video

60.0

This Video · Assessment

Coherent but noticeably synthetic or physically strained.

This Category

70.6

17 videos were rated for this category

This Category · Assessment

Motion and interaction are convincing overall, with recurring but non-disruptive physical anomalies.

Models that scored similarly on this category

C04

Human FidelityWhether faces, eyes, skin, bodies, hands, motion, and human presence remain anatomically and perceptually convincing.

This Video

100.0

This Video · Assessment

Fully convincing within the requested style, with no meaningful uncanny or anatomical defects.

This Category

73.7

27 videos were rated for this category

This Category · Assessment

Human fidelity is strong overall, with recurring but non-disruptive issues.

Models that scored similarly on this category

C05

2D Animation Craft & ArtistryThe quality and artistic authorship of intentionally 2D animation within its requested conventions.

This Video

70.0

This Video · Assessment

Strong and convincingly 2D, with visible craft limitations or less-developed artistic choices.

This Category

60.4

12 videos were rated for this category

This Category · Assessment

2D animation is generally coherent and capable, but synthetic motion or underdeveloped artistic choices recur.

Models that scored similarly on this category

C06

3D Animation Craft & ArtistryThe quality, coherence, and artistic authorship of intentionally 3D animation.

This Video

50.0

This Video · Assessment

The intended 3D animation is recognizable, but repeated technical or expressive defects leave it only partly convincing.

This Category

66.0

5 videos were rated for this category

This Category · Assessment

3D animation is generally coherent, but synthetic, mechanical, stiff, or underdeveloped execution recurs.

Models that scored similarly on this category

C07

Cinematography Craft & ArtistryShot design, framing, camera movement, lighting, focus, visual hierarchy, and intentional filmmaking authorship.

This Video

60.0

This Video · Assessment

Coherent and competent but synthetic, mechanical, or visually underdeveloped.

This Category

71.4

7 videos were rated for this category

This Category · Assessment

Shot design is technically strong overall, though recurring choices are more conventional or less expressive.

Models that scored similarly on this category

C08

Taste & Art DirectionThe strength, specificity, coherence, and artistic authorship of the result as a finished creative choice.

This Video

80.0

This Video · Assessment

Polished, professional, and coherent, but more conventional or less distinctive.

This Category

65.0

60 videos were rated for this category

This Category · Assessment

Art direction is generally capable, but generic, unresolved, or less-confident choices recur.

Models that scored similarly on this category

C09

Animal & Creature FidelityWhether real animals and designed creatures preserve convincing anatomy, identity, motion, behavior, and interaction.

This Video

50.0

This Video · Assessment

The intended animal or creature is clearly recognizable, but repeated anatomical or behavioral defects leave it only partly convincing.

This Category

75.7

7 videos were rated for this category

This Category · Assessment

Fidelity is strong overall, with recurring but non-disruptive anatomical or behavioral issues.

Models that scored similarly on this category

C10

Object & Product FidelityWhether requested objects have correct identity, structure, materials, details, motion, and interactions.

This Video

60.0

This Video · Assessment

This video preserves the requested objects’ identity, structure, materials, details, motion, and interactions in a generally coherent way, though visible failures recur.

This Category

83.8

8 videos were rated for this category

This Category · Assessment

Object and product fidelity is highly reliable, with only isolated detail or material failures.

Models that scored similarly on this category

C11

Causal & Semantic CoherenceWhether events, relationships, transformations, and cause-and-effect remain semantically coherent, including surreal premises.

This Video

100.0

This Video · Assessment

Actions, roles, relationships, and causal progression are completely coherent within the premise.

This Category

80.9

11 videos were rated for this category

This Category · Assessment

Causal and semantic coherence is highly reliable, with only small localized ambiguities.

Models that scored similarly on this category

C12

Text FidelityWhether requested text content, typography, placement, attachment, legibility, and temporal stability are correct.

This Video

100.0

This Video · Assessment

This video preserves requested text content, typography, placement, attachment, legibility, and temporal stability with near-complete consistency.

This Category

82.4

7 videos were rated for this category

This Category · Assessment

Text fidelity is highly reliable, with only isolated content, placement, or temporal defects.

Models that scored similarly on this category

C13

Character Performance & ActingWhether character intention, emotion, timing, reactions, eyelines, gesture, and performance read convincingly.

This Video

80.0

This Video · Assessment

Professional, clear, and convincing, but more conventional or less nuanced.

This Category

68.0

21 videos were rated for this category

This Category · Assessment

Acting is generally believable, but synthetic timing, expression, attention, or specificity recurs.

Models that scored similarly on this category

Differences between other models

CategoryFLUX 3Seedance 2.5Seedance 2.0Kling 3 ProMiniMax Hailuo 03
C01Prompt Adherence87.187.887.484.790.3
C03Physics70.684.766.570.056.5
C02Scene Consistency94.897.292.290.290.2
C04Human Fidelity73.793.382.290.084.4
C052D Animation Craft & Artistry60.480.066.759.257.5
C063D Animation Craft & Artistry66.078.070.058.067.5
C11Causal & Semantic Coherence80.993.680.976.477.3
C08Taste & Art Direction65.075.263.265.865.3
C07Cinematography Craft & Artistry71.480.068.667.178.6
C09Animal & Creature Fidelity75.795.780.080.072.9
C10Object & Product Fidelity83.8100.098.289.781.6
C12Text Fidelity82.493.385.824.390.0
C13Character Performance & Acting68.075.066.778.063.0

Leaderboard

Methodology

The Megaton Index is the published 0–100 weighted summary for v-benchmark v2. This public report shows the model's Index and category scores; the full calibration evidence, decision records, and scoring implementation are reserved for the full report.

Read the full methodology →

Changelog

Videos generated by
Pika API on August 11, 2026
Evaluated
August 11, 2026
Benchmark
v-benchmark v2 by Megaton
Publication
1.0.0