Sora: frames
On start and end frames, Sora reads “First frame only”. An input image anchors the first frame; the prompt says what happens next. As of 2026-09-25.
The model uses the image as an anchor for the first frame, while your text prompt defines what happens next.
OpenAI Cookbook, Sora 2 prompting guide
- Image inputAn input image anchors the first frame; the prompt says what happens next
1Beside the other models
6 other models read the same way: Grok Imagine, HunyuanVideo 1.5, Leonardo Motion 2.0, LTX Studio, Open-Sora 2.0 and others.
Nothing past the docs is assumed about how the model behaves.
2About the field
A last frame pins where a shot ends, which is what lets it hand over cleanly to the next one.
| Model | Reading |
|---|---|
| Adobe Firefly | First and last frame |
| Google Veo | First and last frame |
| Grok Imagine | First frame only |
| Higgsfield | None in the docs |
| HunyuanVideo 1.5 | First frame only |
| Kling AI | First and last frame |
| Leonardo Motion 2.0 | First frame only |
| LTX | Several keyframes |
| LTX Studio | First frame only |
| Luma Dream Machine | First and last frame |
| Midjourney | First and last frame |
| MiniMax Hailuo | First and last frame |
| Moonvalley Marey | Several keyframes |
| Open-Sora 2.0 | First frame only |
| Pika | First and last frame |
| PixVerse | First and last frame |
| Runway | Several keyframes |
| SceneMixer | None in the docs |
| Seedance | N/A, docs unreadable |
| SkyReels | First and last frame |
| Sora | First frame only |
| Step-Video-TI2V | First frame only |
| Vidu | First and last frame |
| Wan | First and last frame |
Inclusion rule. Every model on the register, whether or not it has a page of its own on this field. Order. Alphabetical by model name.