---
title: "Image-to-Video Motion and Identity Test (Preregistered) | Made Research"
description: "WAN 2.2 Turbo vs Kling 2.5 Turbo vs Veo 3.1 in a preregistered, blinded benchmark: 3 runs per prompt, published failure rates, no provider sponsorship."
url: https://www.made.app/research/image-to-video-motion-identity-2026-q3
---

[Research](https://www.made.app/research) / Image-to-Video Motion and Identity Test

# Image-to-Video Motion and Identity Test

Preregistered

The protocol below is frozen. No outputs have been generated or scored yet, so this page makes no quality claims — it commits to how the claims will be made.

Protocol frozen 2026-08-13 · methodology v1.0 · 3 runs per prompt per model

## Models under test

WAN 2.2 Turbo

15 credits / video

[Full pricing](https://www.made.app/price-sheet/wan-2.2-turbo)

Kling 2.5 Turbo

4.2 credits / second

[Full pricing](https://www.made.app/price-sheet/kling-2-5-turbo)

Veo 3.1

40 credits / second

[Full pricing](https://www.made.app/price-sheet/veo-3-1)

## Metrics

| Metric | How it's scored | Scale | Better |
| --- | --- | --- | --- |
| Identity or subject preservation | human rating | 1-5 | Higher |
| Motion quality | human rating | 1-5 | Higher |
| Unwanted scene change | human rating | 0-4 severity | Lower |
| Temporal artifact count | measured and rated | count | Lower |
| Failure rate | measured | percent | Lower |

## Test protocol

**Aspect ratio**

match-reference-or-nearest-supported

**Comparison window seconds**

4

**Source policy**

Use the same Made-owned synthetic reference file for every model and run of a prompt.

## Rating & integrity

- 2 blinded raters score every output. Randomize model labels and output order independently for every prompt.
- Publish both scores, their absolute difference, and the adjudication rate.
- Count provider errors, safety refusals, corrupt media, and empty outputs as failures; do not silently rerun beyond the preregistered run count.
- Exclude only confirmed infrastructure failures affecting every compared model; record the reason and original run ID.
- Cherry-picking prohibited; no sponsored compute; no provider review before publication.

## Results — not yet published

This study publishes no scores until every preregistered run is complete, both blinded raters have scored every output, and two reviewers have signed off. When that happens the summary tables and the full raw run data will appear here — including failure rates and rater disagreement, not just the winning numbers.

## Frozen prompt set: image-to-video-2026-q3-v1

All 20 image-to-video prompts, frozen 2026-08-13 at version 2026-q3-v1. Every model runs every prompt 3 times. Download as [JSON](https://www.made.app/research-data/image-to-video-2026-q3-v1.json).

1. Keep the person recognizable while they blink once and turn their head slightly toward the window; locked camera, subtle natural movement.
2. Preserve the product shape and label-free surface while the camera makes a slow 10-degree orbit from left to right; reflections remain coherent.
3. Animate only the ocean surface and distant clouds; keep the foreground observatory and camera completely stable.
4. The adult subject takes two natural steps toward camera while clothing moves gently in the wind; preserve face, body proportions, and background.
5. Open the gift box lid slowly with both hands and pause at the reveal; preserve the box geometry and every visible finger.
6. Create a gentle push-in toward the plated food while steam rises continuously; do not change the ingredients or plate arrangement.
7. Move from the supplied opening room composition to the supplied ending composition with one smooth lateral camera slide; preserve all furniture.
8. The ceramic artist lifts one finished cup and turns it slightly toward camera; preserve identity, hand count, cup shape, and workshop layout.
9. Animate a slow clockwise half rotation of the watch while the pedestal and background remain fixed; maintain exact metal geometry.
10. Add a light breeze to hair and fabric while the portrait subject remains otherwise still; preserve facial features and camera framing.
11. The train advances slowly along the existing track and stops before the frame edge; preserve station architecture and track alignment.
12. Create one slow rack focus from the foreground bottle to the background bottle; no object movement and no camera translation.
13. The dog looks toward frame left and wags its tail twice; preserve coat pattern, leg count, and the grassy environment.
14. Animate only the candle flames and their reflections; keep every candle, table object, and camera position unchanged.
15. The dancer raises both arms and completes one controlled quarter turn; preserve identity, limb count, costume, and studio mirrors.
16. Create a slow vertical camera rise above the city rooftop while the adult subject remains centered and recognizable.
17. The paper boat drifts a short distance forward while rain ripples appear around it; preserve boat color, fold geometry, and puddle boundaries.
18. The sneaker descends gently onto the plinth while one light sweep moves across it; preserve sole shape, material, and background.
19. The adult chef adds one garnish with the right hand and withdraws the hand; preserve plate contents, face, and kitchen surfaces.
20. Move from the start frame to the supplied end frame with one continuous forward dolly; no cuts, no new objects, and no typography.

## Limitations

- Reference-image difficulty varies even within a controlled set, so results must include prompt-level score spread.
- Automated face similarity may not reflect perceived identity and will be reported separately from human ratings.

## Disclosures

- All identifiable people in the reference set are synthetic adults created for the benchmark.
- No provider sponsors this study.

All studies in the program are listed on the [research index](https://www.made.app/research); the models under test are priced on the [price sheet](https://www.made.app/price-sheet).

---

Markdown version of https://www.made.app/research/image-to-video-motion-identity-2026-q3. Every public Made page has one at its URL plus .md (https://www.made.app/research/image-to-video-motion-identity-2026-q3.md). Site summary for AI assistants: https://www.made.app/llms.txt