Tutorials
Image-to-Video vs Text-to-Video: When to Use Each
Use text-to-video to explore an idea quickly, and image-to-video when the shot has to match a frame, a character or the shot before it. Most finished films use image-to-video or references for the majority of their shots.
Use text-to-video when you're exploring an idea and any good-looking version of the shot will do. Use image-to-video when the shot has to match something specific: a storyboard frame, a character's face, a product, or the shot that comes before it.
In practice, most shots in a finished AI film start from an image or from reference images, and text-to-video is where you sketch. The rest of this guide covers what each mode is good at, what it costs, and two other modes, first/last frame and reference, that sit between them.
What is the difference between text-to-video and image-to-video?
Text-to-video generates a clip from a written prompt alone. The model decides what everything looks like: the people, the room, the framing, the light.
Image-to-video starts from a picture you supply, usually as the first frame, and animates it. The model decides what moves and how, but the composition, faces and set are already fixed by your image.
| Text-to-video | Image-to-video | |
|---|---|---|
| You provide | A prompt | A starting image and a prompt |
| The model decides | Everything | Motion, and what's revealed as the shot moves |
| Composition control | Low | High |
| Character consistency | Low, unless references are used | High in the first frame, then depends on the motion |
| Speed of iteration | Fastest | Slower: you need a good image first |
| Best for | Exploring, mood, b-roll, abstract shots | Planned shots, recurring characters, products, matching a storyboard |
When should you use text-to-video?
Text-to-video is the right tool when you don't yet know exactly what the shot looks like, or when it doesn't need to match anything.
- Exploring an idea. Ten quick takes of "a lighthouse in a storm at dusk" will show you options you wouldn't have drawn.
- Establishing shots and b-roll. A skyline, a forest, traffic at night. Nothing in them has to match another shot exactly.
- Abstract or stylised material. Title backgrounds, textures, dream sequences.
- Testing the plan. Rendering a whole shot list as text-to-video on a cheap model is a quick way to check pacing before you spend on finals.
Write the prompt like a shot-list row: subject, one action, setting, light. Set the camera with camera controls where the tool has them, rather than describing it in prose. The course lesson Writing shot prompts goes into this.
When should you use image-to-video?
Use image-to-video whenever the shot is planned and has to match something.
- A recurring character. Generate a still of the character in the shot, from their reference images, then animate it. The face in the first frame is the face you approved.
- Matching a storyboard. If you've agreed a composition with a collaborator or a client, animating that frame gets you that composition.
- Continuity between shots. Using a frame from the end of the previous shot as the start of the next keeps the set, light and position consistent across the cut.
- Products, logos and props. Anything that must look exactly like the real thing should start from a real image of it.
- Your own photos. A location you shot, a family photo (with permission), concept art.
The trade-off is that the image has to be right first. A bad first frame produces a bad clip, and the model can't fix a composition you've locked in.
What are first/last frame and reference mode?
Two more modes give you different kinds of control:
- First/last frame. You supply the opening and closing frames, and the model fills in the motion between them. Useful for a move between two exact compositions, a transformation, or a camera move that has to land on a specific framing.
- Reference-to-video. You supply reference images (and on some models, reference video and audio) without fixing the first frame. The model keeps the characters, places or style from the references but composes the shot itself. It's the most flexible way to keep a cast consistent across many different shots. See how to keep characters consistent across AI video shots.
Which AI video models support which modes?
Every model takes text. The others vary. These are the models in MoviePro as of October 2026:
| Model | Text | Image | First/last | Reference |
|---|---|---|---|---|
| Seedance 2.5 | Yes | Yes | Yes | Yes (up to 30 images) |
| Seedance 1 Lite | Yes | Yes | Yes | Yes (up to 4 images) |
| Seedance 1 Pro | Yes | Yes | Yes | No |
| Kling 3.0 | Yes | Yes | Yes | No |
| Veo 3.1 and Veo 3.1 Fast | Yes | Yes | Yes | No |
| Runway Gen-4.5 | Yes | Yes | No | No |
| Wan 3.0 | Yes | Yes | No | No |
| Hailuo 2.3 | Yes | Yes | No | No |
| LTX-2.3 Pro | Yes | Yes | No | No |
| MiniMax Director | Yes | Yes | No | No |
| Cinema Studio 4.0 | Yes | No | No | Yes (up to 30 images) |
Does image-to-video cost more than text-to-video?
Not per second of video. On MoviePro a model charges the same rate per second whether you start from text or an image. The only extra cost is the image itself: 2 credits for a generated still, or 3 for a storyboard frame, as of October 2026. Since that image usually saves you a retake, image-to-video tends to be cheaper per usable shot. Full rates are in how much AI video costs per second.
The one mode-related price difference on MoviePro: on Seedance 2.5, sending reference videos raises the rate to about 4.2 times normal. Reference images cost nothing extra.
How do you get a good first frame?
The first frame decides most of the shot, so treat it like a storyboard panel:
- Match the aspect ratio of the film. A square image animated into 16:9 gets cropped or padded.
- Leave room for the motion. If the character will walk left, don't frame them against the left edge.
- Use an image model that takes references when a character is in frame, so the face matches their sheet. Gemini 2.5 Flash Image takes reference images in MoviePro.
- Keep the look consistent. Apply the same style prompt and palette to frames as to video.
In MoviePro, storyboard frames can serve as these first frames. Image frames can become the clip's first frame when you animate from them; Hand-drawn frames go to Seedance as a staging guide for framing and blocking, never as the look (Storyboard it). In Generate Shots, each shot with a frame can either Animate from storyboard frame or Generate from references (no frame), so you can mix both approaches in one run (Render every shot).
A simple rule for choosing
Ask one question about the shot: does it have to match anything?
- No: text-to-video. Iterate quickly and pick the best take.
- Yes, an exact composition: image-to-video from a storyboard frame.
- Yes, a start and an end: first/last frame.
- Yes, the same characters in a new composition: reference mode.
Most films use all four. A common pattern: text-to-video for the opening wide and b-roll, reference mode for dialogue coverage of the cast, and image-to-video for the handful of shots that have to look exactly as planned. The Generate a shot in Video Studio lesson shows each mode in the app, and how to make an AI short film shows where they fit in the whole process.
