VideoGenReview

What each AI video model's release notes say changed

Reference images in AI video: dated changes

35 entries on references or consistency from 13 of the 17 lines, 34 of them dated 6 December 2024 to 14 September 2026. The table lists each with its line and version; the chart counts them per line. As of 2026-09-25.

1Reading the notes

Reference notes describe keeping a person, object or place the same across shots by giving the model pictures of it. Runway's Gen-4 References in April 2025 is the earliest note here built around consistent characters and places; Veo 3.1 took up to three images in October 2025, Vidu one to seven, HappyHorse up to nine and Wan 2.7 up to five mixed images and videos. Sora's character references and Kling's elements carry the same idea under other names, and several 2026 notes tie a voice to the reference. Firefly's references work differently: a reference video lends a new generation its composition from July 2025 and its camera motion from December 2025.

Entries on references per lineCounted from the table. Lines without an entry of this kind are left out.Entries on references per lineWan66Kling AI55PixVerse55HappyHorse33Runway33Vidu33Adobe Firefly Video22Google Veo22MiniMax Hailuo22Grok Imagine11Luma Ray11Seedance11Sora11
Fig. 1 Counted from the table. Lines without an entry of this kind are left out.
Entries on references or consistency per line, read 2026-09-25.
LineEntries
Adobe Firefly Video2
Google Veo2
Grok Imagine1
HappyHorse3
Kling AI5
Luma Ray1
MiniMax Hailuo2
PixVerse5
Runway3
Seedance1
Sora1
Vidu3
Wan6

Inclusion rule. Lines with at least one entry of this kind. Order. Alphabetical by line.

2Every entry

Every entry on references or consistency, oldest first.
DateLineVersionWhat changed
6 December 2024RunwayAct-OneAct-One puts a performance onto characters inside existing videos
30 April 2025RunwayGen-4Gen-4 References keeps characters and locations consistent
14 May 2025WanWan 2.1Wan 2.1 VACE combines local editing, repainting, outpainting, extension and image references
25 June 2025RunwayGen-4An update to Gen-4 References improves object consistency
17 July 2025Adobe Firefly VideoFirefly Video ModelA reference video can pass its composition to a new generation
5 August 2025PixVerseNot named in the noteSound effects and Fusion (reference to video) arrive
26 August 2025ViduNot named in the noteReference video generation from one to seven images
15 October 2025Google VeoVeo 3.1Veo 3.1 preview adds extension, up to three reference images and first-and-last-frame input
21 October 2025ViduVidu Q2Q2 gains reference-to-video, text-to-video and extension
10 November 2025WanWan 2.2Wan 2.2 Animate moves a character photo with the performance from a reference video
15 December 2025Kling AIKling OmniThe Omni video model launches on a new API driven by prompt words
16 December 2025Adobe Firefly VideoFirefly Video ModelA start frame can be paired with a reference video whose camera motion the generation recreates
16 December 2025WanWan 2.6Wan 2.6 reference-to-video keeps a person's look and voice, with several characters at once
18 December 2025Luma RayRay3Ray3 Modify adds keyframe and character reference controls
22 December 2025Kling AINot named in the noteMotion Control in the API takes a reference image and a reference video together
13 January 2026Google VeoVeo 3.1On Vertex AI, reference-to-video takes 9:16 and 1080p or 4K outputs can be upsampled
23 January 2026Kling AINot named in the noteOther angles of an element can be filled in from its front view
29 January 2026WanWan 2.6A Flash tier of Wan 2.6 reference-to-video
12 February 2026PixVersePixVerse V5.6Fusion reference-to-video moves to V5.6
25 February 2026Kling AIKling 3.0Kling 3.0 and 3.0 Omni reach the API; elements can be built from video
12 March 2026SoraSora 2Character references, clips up to 20 seconds, 1080p on Sora 2 Pro, extensions and batch jobs
23 March 2026Kling AINot named in the noteElements made from several images can be tied to a voice
3 April 2026WanWan 2.7Wan 2.7 reference-to-video mixes up to five image or video references and clones a voice timbre
7 April 2026PixVersePixVerse C1The C1 model covers text, image, transition and reference-to-video
26 April 2026HappyHorseHappyHorse 1.0HappyHorse 1.0 reference-to-video takes up to nine reference images
26 April 2026HappyHorseHappyHorse 1.0HappyHorse 1.0 video edit changes parts of a clip from instructions and up to five images
21 May 2026PixVersePixVerse V6V6 Fusion makes reference names and types optional
16 June 2026HappyHorseHappyHorse 1.1HappyHorse 1.1 reference-to-video keeps subject and scene style steadier
26 July 2026PixVersePixVerse V6V6 reference-to-video takes video references in an omni mode
31 July 2026Grok ImagineGrok Imagine 1.5Grok Imagine video 1.5 adds reference-to-video with preset voices and native 1080p
31 July 2026MiniMax HailuoMiniMax H3MiniMax H3 reads text, image, video and audio together as creative context
5 August 2026MiniMax HailuoHailuo 3.0Runway's developer API adds a model it names MiniMax Hailuo 3.0
6 August 2026WanWan 3.0Wan 3.0 folds reference, editing, replication and driving into one model, with clips up to 30 seconds
14 September 2026ViduVidu Q3Vidu Q3 reference-to-video models for drama and for ads are listed on Model Studio
Undated version rowSeedanceSeedance 2.0Seedance 2.0: 4 to 15 seconds, up to 4K, multimodal references, editing and extension

Inclusion rule. All entries tagged with this kind of change, including platform arrivals and undated rows. Order. By date, oldest first; undated rows last.

3One line at a time

4Other kinds of change