Which AI video models take several keyframes?
3 of the 24 models: LTX, Moonvalley Marey and Runway. The other 21 do not, in the docs read. As of 2026-09-25.
| Model | Frames | What the docs say |
|---|---|---|
| LTX | Several keyframes | The model supports several keyframes, not only a first frame |
| Moonvalley Marey | Several keyframes | Several reference images are placed on a timeline and the model makes the transitions |
| Runway | Several keyframes | Image to video uses an input image; keyframes moved to an Animate Frames app |
Inclusion rule. Models whose own docs meet the condition in the question. Order. Alphabetical by model name.
1Why it matters
Several keyframes turn a clip into a planned sequence of positions rather than a start and an end.
2From the docs
Multiple keyframe support
Lightricks, LTX-Video repository
Upload multiple reference images and place them on a timeline.
Moonvalley, Marey
3The rest of the register
Not matching: Adobe Firefly, Google Veo, Grok Imagine, Higgsfield, HunyuanVideo 1.5, Kling AI, Leonardo Motion 2.0, LTX Studio, Luma Dream Machine, Midjourney, MiniMax Hailuo, Open-Sora 2.0, Pika, PixVerse, SceneMixer, Seedance, SkyReels, Sora, Step-Video-TI2V, Vidu and Wan.
- Advanced featuresThe model supports several keyframes, not only a first frame
- KeyframingSeveral reference images are placed on a timeline and the model makes the transitions
- ReplacementsImage to video uses an input image; keyframes moved to an Animate Frames app
- FrameA first and a last keyframe image can be uploaded as fixed points for the clip
- Documentation contentsA separate mode generates video from a first and a last frame
- OverviewVideo is generated from text or from a still image
- Camera controls page read; first and last frames are not describedNot described in the docs read (2026-09-25)
- AbstractText-to-video and image-to-video in one framework
- Capabilities tableStart and end frames to video is supported in VIDEO 2.6 and 3.0
- Turning images into a videoA start frame from an uploaded or generated image
- How to usePresets are applied to a clip made from a first-frame image on the LTXV model
- KeyframesKeyframes set a start image, an end image or both
- Loop and end parametersThe image is the starting frame; --end sets an ending frame and --loop makes the two the same
- First/last-frame image-to-videoA first-frame image, a last-frame image or both can be set
- Image-to-video generationImage-to-video from a prompt and a reference image
- PikaframesPikaframes takes a first and a last frame as still images
- API contentsA Transition endpoint generates video between a first and a last frame
- Storyboard page read; first and last frames are not describedNot described in the docs read (2026-09-25)
- Whole pageThe API reference asked for a sign-in and showed no text on the day read
- NewsStart and end frame control is part of the diffusion forcing model's inference codeMay 16, 2025
- Image inputAn input image anchors the first frame; the prompt says what happens next
- Model summaryThe image condition becomes the first frame of the generated video
- images parameterTwo images set the start frame and the end frame
- Related API referencesA separate image-to-video mode takes a first and a last frame