Rapidata/image-to-video-realism-human-feedback-benchmark

Rapidata image-to-video realism & visual quality Benchmark

Built by Rapidata, collected with the Rapidata Python SDK.

To evaluate your own models and create a leaderboard, check out our MRI.

Live Leaderboard

Explore the full interactive leaderboard — filter by model, question and prompt, and inspect individual head-to-head matchups on Rapidata.

Open the live leaderboard

At a glance

Head-to-head comparisons (rows)	2,981
Human votes	522,210
Models compared	19
Prompts	25

The leaderboards

Leaderboard	Question shown to annotators	Prompt shown?
Photorealism	Which video looks more like it was shot by a real camera?	No
Visual integrity	Which video has more glitches and is more likely to be AI generated?	No

Dataset structure

One row per head-to-head comparison of two samples generated from the same prompt.

Column	Type	Meaning
prompt	string	Prompt segment prompt (text).
prompt_asset	image	Prompt segment prompt_asset: the image itself, embedded.
video1	string	Compared video 1: public URL of the video.
video2	string	Compared video 2: public URL of the video.
model1	string	Name of the model that produced video1.
model2	string	Name of the model that produced video2.
weighted_results_video1_photorealism	float32	Sum of the userScore weights of the votes for video1 on the Photorealism leaderboard.
weighted_results_video2_photorealism	float32	Sum of the userScore weights of the votes for video2 on the Photorealism leaderboard.
detailed_results_photorealism	string	Every vote on the Photorealism leaderboard as a JSON array; votedFor is A for side 1 and B for side 2.
weighted_results_video1_visual_integrity	float32	Sum of the userScore weights of the votes for video1 on the Visual integrity leaderboard.
weighted_results_video2_visual_integrity	float32	Sum of the userScore weights of the votes for video2 on the Visual integrity leaderboard.
detailed_results_visual_integrity	string	Every vote on the Visual integrity leaderboard as a JSON array; votedFor is A for side 1 and B for side 2.

The weighted_results_* pair is a userScore-weighted vote sum per side, not a probability: the two values of a row do not sum to 1. Divide by their sum for a normalised win share.

Embedded image columns are not format-homogeneous: one column may mix JPEG, WebP and GIF (an animated GIF keeps all its frames).

Overall ranking

#	Model	ELO
1	ground_truth_baseline	1469.65
2	minimax-h3	1324.98
3	grok-imagine-1.5	1315.7
4	kling-3.0-pro	1290.87
5	gemini-omni-flash-1.1	1288.05
6	flux-3-video	1277.01
7	veo-3.1	1266.6
8	wan-3.0	1249.97
9	minimax-h3-max	1194.15
10	seedance-2.5	1175.61
11	happy-horse-1.1	1137.92
12	pixverse-v5.6	1085.73
13	vidu-q3-pro	1083.15
14	cosmos-3-super	987.226
15	cosmos-3-nano	717.128
16	alaya-world	269.276
17	yume-1-5	143.746
18	sana-wm	-276.77
19	hidream-o1-video	—

Photorealism ranking

#	Model	ELO
1	ground_truth_baseline	1475.11
2	grok-imagine-1.5	1303.62
3	kling-3.0-pro	1299.02
4	gemini-omni-flash-1.1	1297.21
5	minimax-h3	1283.28
6	veo-3.1	1270.62
7	flux-3-video	1246.95
8	wan-3.0	1246.1
9	minimax-h3-max	1166.16
10	happy-horse-1.1	1161.2
11	vidu-q3-pro	1107.51
12	seedance-2.5	1105.53
13	pixverse-v5.6	1076.94
14	cosmos-3-super	985.992
15	cosmos-3-nano	743.64
16	alaya-world	225.838
17	yume-1-5	209.592
18	sana-wm	-204.315
19	hidream-o1-video	—

Visual integrity ranking

#	Model	ELO
1	ground_truth_baseline	1463.65
2	minimax-h3	1369.01
3	grok-imagine-1.5	1329.51
4	flux-3-video	1309.06
5	kling-3.0-pro	1284.29
6	gemini-omni-flash-1.1	1280.52
7	veo-3.1	1264.04
8	wan-3.0	1255.51
9	seedance-2.5	1247.85
10	minimax-h3-max	1224.02
11	happy-horse-1.1	1115.8
12	pixverse-v5.6	1096.09
13	vidu-q3-pro	1059.7
14	cosmos-3-super	989.838
15	cosmos-3-nano	690.277
16	alaya-world	311.785
17	yume-1-5	68.9843
18	sana-wm	-359.954
19	hidream-o1-video	—

Citation

If you use this dataset, please cite it:

@misc{giger2026realism,
  title     = {{image-to-video-realism-human-feedback-benchmark}},
  author    = {Giger, Lino and Caldarone, Alex and Lin, Deli and Sundararajan, Aniruddha and Teikari, Elias and Vikat, Karl and Kannwischer, Marian},
  year      = {2026},
  publisher = {Hugging Face},
  doi       = {10.57967/hf/10987},
  url       = {https://huggingface.co/datasets/Rapidata/image-to-video-realism-human-feedback-benchmark}
}

Giger, L., Caldarone, A., Lin, D., Sundararajan, A., Teikari, E., Vikat, K., & Kannwischer, M. (2026). image-to-video-realism-human-feedback-benchmark [Data set]. Hugging Face. https://doi.org/10.57967/hf/10987

About Rapidata

Rapidata's technology makes collecting human feedback at scale faster and more accessible than ever before. Visit rapidata.ai to learn more.