# Keyframes Are Material Too: Use Frames From a Decomposed Video as Cover and Image References

> Once a video is decomposed in Beav, its keyframes are copied to the decompositions folder in your asset library along with the transcript, shot MP4 clips and M4A audio cues. When you need a cover or an illustration, drag a keyframe from the asset library on the right into a new chat as a reference image, say whether you're after its composition, colors or layout, and swap in your own subject. Results land in Generated, and you star the ones you like to save them to the library. A prompt, a note on rights, and how it compares with screenshots and inspiration boards.

- Source: https://getbeav.com/en/blog/features/keyframes-as-cover-references
- Author: JambaHailar (https://x.com/JambaHailar), Beav
- Published: 2026-11-22
- Last updated: 2026-11-22

> **Key takeaways**
>
> * When a video is decomposed, its keyframes are copied to `decompositions/<title>` in the asset library along with the transcript, shot MP4 clips and M4A audio cues, and the origin panel records the source and author. The screenshot-to-camera-roll step is already done for you.
> * For a cover or an illustration, describe the image you want in a new chat and attach a keyframe as a reference: paste it, drag it in, or drag it straight from the asset library on the right. Say whether you're borrowing its composition, colors or layout.
> * Reference frames are for learning composition, color and layout, not for using someone else's shot as your cover. Generated images land in **Generated** first; star the keepers to save them to your library.

After decomposing a video, most people reach for the transcript and the shot script first. The keyframes tend to get ignored: a folder of a few dozen stills that looks like a by-product.

Yet they're exactly what you're missing when it's time to make a cover. Chances are you saved that video because one frame caught you: the subject standing in the left third with a wide empty space on the right for text, or a muted warm palette that runs through the whole thing. Getting that feeling across to an AI used to mean pausing, screenshotting, digging through your camera roll and uploading to an image tool, with the source long gone. Decomposing now saves those frames for you and remembers where they came from. Saving and decomposing are how you build a material library. Turning a keyframe into a cover of your own is where it pays off.

## Who it is for

* Creators on Xiaohongshu (RedNote) or short-video platforms who have saved a set of benchmark videos and want their covers and images to have that composition and tone.
* Shop owners and personal brands who can shoot and write, but whose covers always end up as "big text over a photo".
* Content leads who need sourced references to tell a designer or freelancer "this is the look we want".

## How to use it

First save the reference video to your knowledge base: click &#x2A;*+** beside Knowledge Base, choose **Add link**, and paste the link or share text. **Add link** goes through the capture service and uses quota according to its rules. Once the video has downloaded to your computer, open it and click **Decompose**, which uses model quota. How that works is covered in [breaking down a viral short video](https://getbeav.com/en/blog/features/video-breakdown-shot-script).

![Four steps to use keyframes as cover references: keyframes land in the asset library after decomposing, pick the frame you want to learn from, describe the image in a new chat and drag the keyframe in, choose results in Generated and save them to the library](https://getbeav.com/blog/features/keyframes-as-cover-references-steps-en.webp)

*The reference frame gives the composition and tone, the content is yours, and the keepers go back into your library.*

1. **Find the keyframes in the asset library.** Results are copied to `decompositions/<title>` in the asset library: the transcript, keyframes, shot MP4 clips and M4A audio cues. Items decomposed earlier keep `拆解/<title>`. The origin panel keeps the source, author and time range, so you can always check whose frame it is.
2. **Pick the frame you actually want to learn from.** Don't just pick the prettiest one; work out why it works. Where is the subject, which side is empty, where does the text sit, what's the overall tone? If you're unsure, send the image to the AI and ask about the subject and text placement. It inspects the area without changing the original.
3. **Describe the image in a new chat, then attach the keyframe.** Start a new chat and say what the image is for: a Xiaohongshu cover or an in-post image, the aspect ratio, the title, and what your subject is. Then attach the reference: paste it, drag it in, or press `⌘K` (`Ctrl+K` on Windows / Linux) with the input box focused to open the right sidebar and drag the keyframe from `decompositions/<title>` in **Assets** onto the input box. For example:

```text
Make a Xiaohongshu cover, 3:4, titled "A balcony vegetable garden for renters".
The attached image is a keyframe from a video I decomposed. Use only its composition and tone:
subject in the left third, a wide empty area on the right for the title, muted warm colors.
The content should be mine: a small balcony, a wooden rack with pots of scallions and lettuce, early evening light.
Leave out the people, objects and text from the reference. Keep the title clear and skip extra decoration.
```

4. **Say what to borrow, and don't let the AI trace it.** Name one or two things: composition, colors, text placement, or the rhythm of the frame. For Xiaohongshu covers, the AI considers 12 types of composition (single image, image with text, comparison, collage and more) and picks one based on the content; that doesn't mean 12 images each time. A reference frame settles that choice for it. To lock in a visual style for the long run, use the creation templates below a new chat, or create a personal template from a finished image of your own that you like. See [turning a reference image into an AI cover template](https://getbeav.com/en/blog/features/create-covers-with-reference-images).
5. **Pick results in Generated and save the keepers.** AI-generated images don't go straight into the asset library. They land in **Generated**, the item below the divider under the asset library categories on the left, where you can search, preview and drag them into a chat. To keep one, hover over the card and click the star in the top-left corner, or right-click and choose **Save to library**; generated image cards in a chat also have a star in the top-right corner. Beav's AI searches only the asset library by default, so to reuse the one you just made, say something like "use the image I just generated".

![What to learn from a reference keyframe and what not to take: on the left composition, colors, text placement and empty space; on the right the original's people, footage, text and brand elements](https://getbeav.com/blog/features/keyframes-as-cover-references-learn-en.webp)

*Learn how the frame is organized, not the frame itself.*

**Same for in-post images.** Pick a few frames from the same video and ask the AI for a set of images in that tone, so the whole post feels consistent. When making a complete Xiaohongshu image-and-text post, if you haven't authorized generating every image at once, the AI makes two first for you to confirm, then continues with the rest.

**What uses quota?** Saving with **Add link** goes through the capture service and uses quota according to its rules. Decompose uses model quota. The chat runs on the chat model. Image generation uses credits or external services you've configured; if nothing comes back right away, check the original task's status instead of submitting it again.

## How it compares with the usual ways

|                                | Screenshot to camera roll, upload to an image tool | Collecting on Pinterest or a similar inspiration board         | Beav: decomposed keyframes as references                                                 |
| ------------------------------ | -------------------------------------------------- | -------------------------------------------------------------- | ---------------------------------------------------------------------------------------- |
| Strength                       | Quick to grab, lots of image tools to choose from  | Huge volume and good categories, great for finding a direction | Frames from the videos you saved are already extracted, next to the transcript and shots |
| Where the reference comes from | You pause and screenshot                           | Images curated by other people                                 | Decompose copies them into `decompositions/<title>`                                      |
| Is the source kept?            | Usually lost after the screenshot                  | Often links back to the original                               | The origin panel records source, author and time range                                   |
| Where images get made          | Open another tool and upload the reference         | Usually not part of it                                         | Describe, attach and generate in the same chat                                           |
| Where results live             | Camera roll or downloads folder                    | Not applicable                                                 | In Generated first; keepers saved to your library                                        |

For broad direction and sheer variety, an inspiration board covers far more ground than a few dozen of your own videos. Image tools each have their strengths, and for a specific art style a dedicated tool may serve you better. Beav suits people who are already collecting benchmark videos: no separate screenshots, the source travels with the frame, and generating and keeping images happen in the same library.

## Limits and notes

* **Reference frames are for composition, color and layout.** Don't use a frame from someone else's video as your cover, and don't ask the AI to trace one. The people, footage, text and brand elements belong to the original creator; get permission before using them.
* **Decompose first, with the video stored locally.** Decompose only runs when you click it; saving never triggers it. A video that hasn't downloaded to your computer can't be split into shots, so it has no keyframes.
* **Check what you generate.** Proofread every character of cover text, and check the aspect ratio and sharpness against the platform's requirements.
* **Pasted images don't go into the asset library.** An image you paste into a chat is saved only as an attachment to that message. Keyframes you drag in from the asset library are already there.
* **Beav doesn't post for you.** It helps you find references, generate and keep images. Publishing is yours.

## Next steps

* How to choose a Xiaohongshu cover composition based on the content: [making a Xiaohongshu cover](https://getbeav.com/en/blog/features/viral-cover-34-generation).
* Where keyframes come from: [breaking down a viral short video](https://getbeav.com/en/blog/features/video-breakdown-shot-script). For why it's worth building up references over time, read the series overview, ["Accumulation Is the Moat"](https://getbeav.com/en/blog/features/accumulation-is-the-moat).
* Full reference: [images and covers](https://getbeav.com/en/docs/workflows/media#images-and-covers), [Generated](https://getbeav.com/en/docs/workflows/media#generated) and [decompose videos and notes](https://getbeav.com/en/docs/workflows/collect-to-create#decompose-videos-and-notes).

*Applies to Beav desktop 2.10 and later.*
