Leonardo AI Video Generator: Motion 2.0 and Other Models

Leonardo started as an image tool and now covers video too, in two ways. Its own video model, Motion 2.0, comes with Leonardo-style controls: preset motion controls and motion elements you can weight, much like Elements for images. Alongside it sit many third-party models, from Google's Veo 3.1 to Kling, Seedance, Wan and MiniMax's Hailuo, so you can pick the engine per shot without leaving the platform. This guide covers both, using the parameters Leonardo documents in its API documentation; the web app shows the same controls under similar names.

Motion 2.0: Leonardo's own video model

Motion 2.0 (and a Motion 2.0 Fast variant) turns a prompt, or a start frame plus a prompt, into a short clip. What sets it apart are two kinds of add-on:

• Motion controls: one preset per generation, with a weight from −2 to 2 (default 1).
• Motion elements: several presets can be stacked, each with its own weight from −2 to 2.

Leonardo defines them separately even though they share a format, so read each preset's description in the app before combining them. Other settings: a negative prompt of up to 1,000 characters, prompt enhancement (Auto, On or Off), frame interpolation to blend frames smoothly (on by default), a seed and style presets for vibe, lighting and colour.

Motion 1.0 was retired on 17 June 2026; Leonardo recommended Hailuo 2.3 as its replacement in the API, so older tutorials that show Motion 1.0 no longer apply.

Third-party models inside Leonardo

Leonardo's documentation lists video models including Veo 3.1 (with Fast and Lite), Kling (2.1 to 3.0, O1 and O3), Seedance (1.0 Pro to 2.5), Wan (2.6, 2.7 and 3.0), MiniMax Hailuo 03, Gemini Omni Flash, Grok Imagine and others. Each brings its own options. Veo 3.1 is a good example of how they differ:

• Durations of 4, 6 or 8 seconds.
• Start frame, and an end frame when a start frame is set.
• Up to three reference images on the full Veo 3.1 model only, not on Fast or Lite.
• Audio on by default, which you can switch off.
• 720p, 1080p or 4K in 16:9 or 9:16; 4K is not available on Veo 3.1 Lite.

Models are also retired as providers change them: Leonardo removed Veo 3 and Veo 3 Fast in June 2026 in favour of Veo 3.1, and Sora 2 in July 2026. Check the model menu before following any older guide.

Image to video, step by step

1. Make or choose the still. For a product or a character, generate it in Leonardo first so the look is approved before motion is added; see Lucid Origin vs Phoenix.
2. Open video generation and select the model.
3. Add the image as the start frame (and an end frame where the model supports it).
4. Write the prompt about motion only: what moves, how the camera moves, the pace.
5. Choose resolution, aspect ratio and duration, and turn audio on or off where offered.
6. Generate at low resolution first, then rerun the keeper at full resolution with the same seed.

Prompting motion that holds together

• One action, one camera move. "She turns towards the window; slow push in" works; a list of actions does not.
• Describe the motion, not the picture. The start frame already shows the scene.
• Name the pace. "Slow", "gentle" or "sudden" changes the result as much as the action does.
• Use the negative prompt sparingly on models that have one, for faults such as flicker, warping or extra limbs.
• Keep presets subtle. A motion control at full weight on top of a strongly worded prompt often overshoots; lower the weight before rewriting the prompt.

Choosing a model for the shot

Whatever you choose, record the model, seed and settings with the clip; the same prompt behaves differently on each engine.

Managing cost and iterations

Video is the most expensive thing you can generate on any platform, and Leonardo is no exception: each model shows its own cost, and higher resolutions, longer durations and audio raise it. A sensible routine keeps the spend under control. Test the motion idea at the lowest resolution and shortest duration the model allows. Fix the seed once a test looks promising, so the final render differs only in resolution. Change one thing per attempt, whether the prompt, a preset weight or the model, and write down what changed. When a model fails three times on the same idea, try a different engine rather than a fourth wording; some motions simply suit one model better than another.

Try it

The hiker stops at the ridge and looks out over the valley as low cloud drifts across the peaks. Slow push in, gentle and calm, soft morning light.

Use it with a start frame of a hiker on a ridge. Generate at 480p on Motion 2.0 with no presets, then add one motion control at weight 1 and compare. If the motion overshoots, lower the weight to 0.5 rather than rewriting the prompt.