# Guide a video with Reference Images to Video

> Use the Nodespell Reference Images to Video Creative Tool to generate a clip guided by several images of a subject, product or style.

Source: https://docs.nodespell.com/creative-tools/reference-images-to-video/ · Last updated: 2026-11-10

Reference Images to Video is the Nodespell Creative Tool for generating a video clip guided by several images at once. The images describe who or what should appear, the style, or the composition, and the prompt describes the shot; unlike Image to Video, no single image has to become the first frame.

## When to use Reference Images to Video

| If you have…                                              | Use                                               |
| --------------------------------------------------------- | ------------------------------------------------- |
| Several images of a character, product, outfit or setting | **Reference Images to Video**                     |
| One image that must be the opening frame                  | [Image to Video](https://docs.nodespell.com/creative-tools/image-to-video/) |
| No images, only a prompt                                  | [Text to Video](https://docs.nodespell.com/creative-tools/text-to-video/)   |
| A finished clip you want to restyle or edit               | [Modify Video](https://docs.nodespell.com/creative-tools/modify-video/)     |

Reference Images to Video lists models whose image input accepts many images. Many of them also accept reference videos and reference audio.

## Generate a video from reference images

1. Switch the workspace to **Create**. Select **Image to Video** in the tool rail, open the tool switcher at the top of the form, and choose **Reference Images to Video**. It is also listed under **All tools**.
2. Check the model. The default is **Seedance 2.5 Reference to Video**.
3. Add images to the reference-images field (labelled **Reference images**, or **Reference Images** on models that use their own labels). Select the field to open **Choose assets** and pick several Assets or upload files.
4. Add reference videos or reference audio if the model offers those fields and you need motion or timing guidance.
5. Write a prompt that says what each reference is for, for example "the woman in `@Image1` wears the jacket from `@Image2`".
6. Set duration, resolution and aspect ratio.
7. Set **Generations** from 1 to 4, then select **Generate**.

## Refer to images in the prompt

Several models number references so the prompt can point at them. Seedance 2.5 Reference to Video, for example, reads `@Image1`, `@Image2`, `@Video1` and `@Audio1` in the order you added the files. Reorder references if the numbering matters, and name each one's role in the prompt. Models that do not use this syntax treat the images as general guidance.

## Reference Images to Video controls

| Control                             | What it does                                        | Options and defaults                        |
| ----------------------------------- | --------------------------------------------------- | ------------------------------------------- |
| Tool switcher (title)               | Moves to another video Creative Tool                | All video tools                             |
| Model picker                        | Chooses the model                                   | Default: Seedance 2.5 Reference to Video    |
| **Previous model** / **Next model** | Steps through eligible models                       | —                                           |
| Reference-images field              | The ordered set of guide images                     | Multiple images; limits depend on the model |
| **More settings**                   | Less-used model settings                            | Collapsed                                   |
| **Generations**                     | Number of runs submitted together                   | 1 to 4, default 1                           |
| **Generate**                        | Adds or updates the model node and submits the runs | Shows the credit estimate                   |

Reference video, reference audio, duration, resolution, aspect ratio and audio fields appear only when the model has them, under the model's own labels.

## Recommended models for Reference Images to Video

The **Featured** collection lists, in order:

- Seedance 2.5 Reference to Video (default)
- MiniMax H3 Reference to Video
- Veo 3.1

Veo 3.1 appears in both Image to Video and Reference Images to Video because it accepts a start frame, a last frame and reference images. See [Choose a model](https://docs.nodespell.com/creative-tools/choose-a-model/) for the rest of the list.

## What changes with the model?

- **Reference types:** Seedance 2.5 Reference to Video accepts reference images, videos and audio. Its audio references need at least one image or video reference. Other models accept images only.
- **Required references:** some models, such as Gemini Omni Flash Reference To Video, require at least one reference image; on Seedance 2.5 all reference inputs are optional.
- **Duration:** Seedance 2.5 offers `auto` or 4 to 30 seconds; MiniMax H3 offers 5 to 15 seconds.
- **Resolution and audio:** Seedance 2.5 offers 480p to 1080p and a **Generate Audio** toggle that is on by default.
- **Start frame:** a few reference models, such as Kling 4.0 Reference to Video, also take an optional start image alongside the references.

## Get consistent subjects

- Use clean, well-lit images that show the subject clearly. Several angles of the same subject help more than many near-duplicates.
- Keep the references to what the shot needs. Unrelated images pull the result in different directions.
- State in the prompt what must stay the same, such as face, logo or outfit.

For a full walkthrough, see [Keep a character consistent](https://docs.nodespell.com/recipes/consistent-character/).

## Where does the video go?

**Generate** adds a Reference Images to Video model node to the current graph level and runs it. Finished clips appear in **Results** in the Create form and on the node; see [Takes](https://docs.nodespell.com/graph/takes/). Generated media also appears under **Generated** in the [Asset Manager](https://docs.nodespell.com/assets/asset-manager/).

## Common problems

- **"This value is required."** The prompt is empty, or the model requires a reference image.
- **A reference field says "Controlled by an upstream connection".** A node in the graph feeds that input. Change it in the graph, or disconnect it.
- **The subject drifts between shots.** Add clearer references, refer to each image in the prompt, and keep the same model for every shot.

## Frequently asked questions

### How many reference images can I add?

It depends on the model, and the field accepts only what the model allows. Read the model's description in the picker before relying on a large set.

### Is a reference image used as the first frame?

No. Reference images guide the subject and style; they are not placed as the opening frame. To fix the first frame, use [Image to Video](https://docs.nodespell.com/creative-tools/image-to-video/).

### Can Reference Images to Video use video and audio references?

Yes, on models that offer them. Seedance 2.5 Reference to Video, Seedance 2.0 Reference To Video and Wan 3.0 Reference to Video accept images, videos and audio.

### What is the default Reference Images to Video model?

The default is Seedance 2.5 Reference to Video, which accepts mixed image, video and audio references and clips of `auto` or 4 to 30 seconds.

## Related

- [Image to Video](https://docs.nodespell.com/creative-tools/image-to-video/) — anchor the shot on one image
- [Keep a character consistent](https://docs.nodespell.com/recipes/consistent-character/) — references across several shots
- [Choose a model](https://docs.nodespell.com/creative-tools/choose-a-model/) — compare reference support
- [Upload and organize Assets](https://docs.nodespell.com/assets/upload-and-organize/) — prepare reference images
- [Connections](https://docs.nodespell.com/graph/connections/) — feed references from other nodes