Text-to-video (T2V) is a generative AI technique that turns a written description — a prompt — into a moving video clip, complete with motion, lighting, and (on modern models) synchronized audio. You describe the scene; the model films it.
("T2V" is simply shorthand for text-to-video — you'll also see it written as text2video, txt2vid, or text 2 video.)
How it works
Under the hood, most text-to-video models are diffusion models trained on enormous libraries of video paired with captions. During generation, the model starts from pure noise and progressively "denoises" it into frames, guided at every step by your prompt. Because the model generates all frames within a shared temporal context, the output stays coherent: objects persist, motion flows, and the camera behaves like a camera rather than a slideshow.
T2V vs I2V vs R2V at a glance
| Mode | Your input | The model treats it as | Best for |
|---|---|---|---|
| Text-to-video (T2V) | a written prompt only | the full creative brief | net-new scenes from scratch |
| Image-to-video (I2V) | one image (plus an optional last frame) | the literal opening frame | animating a still you already have |
| Reference-to-video (R2V) | a set of images, clips, and audio | identity and style anchors | consistent characters, products, and series |
What it's best at
Text-to-video shines when you're starting from nothing but an idea:
- Concept exploration — visualize a scene before committing to a shoot or a storyboard.
- Short-form content — hooks, B-roll, product teasers, and social clips in vertical or widescreen formats.
- Impossible shots — drone passes through a keyhole, a city made of glass, a fox in a spacesuit. If you can describe it, the model can attempt it.
What a good prompt includes
The strongest T2V prompts read like a director's shot note, not a wish. Cover four things: the subject (who or what), the action (what happens over time), the camera (framing and movement), and the atmosphere (lighting, palette, mood). For a deeper walkthrough, see our guide on how to write AI video prompts.
Text-to-video on Molyin
Text-to-video is the baseline mode across Molyin's video lineup — every live model accepts a text prompt. What differs is how long a single take can run and how high it renders:
| Model | Longest single clip | Top resolution |
|---|---|---|
| Seedance 2.5 | 30s | 1080p |
| Wan 3.0 | 30s | 1080p |
| LTX 2.5 | 20s | 4K |
| FLUX 3 | 20s | 1080p |
| Seedance 2.0 | 15s | 4K |
| Kling 3.0 | 15s | 4K |
| MiniMax H3 | 15s | 2K |
| HappyHorse 1.1 | 15s | 1080p |
| Wan 2.7 | 15s | 1080p |
| Veo 3.1 | 8s | 4K |
Faster, cheaper tiers of the same families — Seedance 2.0 Fast and Mini, LTX 2.5 Pro, Veo 3.1 Fast — bring the generator to 14 live text-to-video models. Pick one, write the shot, and the cost per second is shown before you generate.