THE AI RAP VIDEO WORLD PLAYBOOK
How to choose the best photos for AI rap videos
Choose photos that make each person easy to recognize before thinking about the stage or effects. This checklist separates useful portrait choices from the file checks the current workspace actually performs, so you can prepare a two-person rap brief without guessing what an upload error means.
Selecting a photo previews it in your browser; it does not submit a video job. Public video generation and checkout are not open yet. Passing a photo check is not proof that a model will preserve a face, animate a mouth accurately or produce a finished video.
Start with two portraits or one shared photo
Two separate portraits
Use one photo of each adult. The first portrait represents the left performer; the second represents the right performer. Choose two different files showing the intended people.
One shared photo
Use one image showing both adults with visible, separated faces. The initial left and right positions identify the roles. Avoid a busy group shot with several other people.
Separate portraits can be easier to arrange when the people are in different places. A shared photo can preserve the context of an occasion. Neither setup has been independently demonstrated here to produce better video quality, so choose the one whose faces and role assignments you can review most clearly.
The two-person rap workspace lets you switch between these setups. For a celebration, begin with the birthday rap workspace and keep the same left/right check when assigning lyrics.
A practical portrait checklist
- A visible face. Prefer an unobstructed face that is large enough to inspect. Hair, hands, dark glasses or a microphone covering the mouth make your own review harder.
- Even light. Choose a photo where the face remains readable rather than disappearing into deep shadows or blown-out highlights.
- A natural angle. A front or gentle three-quarter view is easier to assess than an extreme profile. This is preparation advice, not an automatic pose check.
- Useful space. Leave some room around the head and shoulders. Tight crops, clipped hair and faces at the edge reduce the framing choices you can judge.
- Distinct people. In a shared image, keep the two faces separated and recognizable. In separate mode, check that the same person or file was not selected twice.
- An image you may use. Use adult portraits with permission and confirm that the intended use matches that permission.
Lighting, face visibility and consent are not established by a file-format validator. Review those yourself. A photo can be technically accepted yet still be a poor creative reference; a larger file does not automatically make it a better portrait.
Current file limits and checks
These are the workspace’s current acceptance rules. They describe supported inputs, not a quality score.
| What to check | Current requirement |
|---|---|
| Format | JPEG, PNG or static WebP with a matching supported image type. HEIC, GIF and animated WebP are not accepted. |
| File size | At most 6 MiB per photo: 6,291,456 bytes. |
| Encoded dimensions | Each side must be 301–5999 pixels. Exactly 300 or 6000 pixels is outside the accepted range. |
| Shape | Width divided by height must be strictly greater than 0.4 and strictly less than 2.5. |
| Browser preview | The file must pass supported metadata checks and decode as an image in the browser. |
For example, a 1200 × 1600 portrait fits the dimension and ratio limits. A 4000 × 1000 panorama does not: its ratio is 4. A 250 × 250 thumbnail is too small even if its file size is below the limit. These are arithmetic examples, not model-generated sample images.
The shape of your reference photo is separate from the output aspect ratio selected in the studio. A portrait photo does not lock the final video to 9:16. Review the output framing as its own choice rather than stretching the photo to imitate it.
What to do when a photo is rejected
- The file is too large
- Export a smaller JPEG, PNG or static WebP while retaining a readable face. Check both the file size and dimensions again.
- The dimensions or shape are unsupported
- Crop from a larger original when possible. Simply enlarging a tiny thumbnail changes its dimensions without restoring missing detail.
- The file is the wrong format
- Export it to a supported format in your photo app. Renaming a .heic or .gif file to .jpg does not convert its contents.
- The file still will not preview
- Try a fresh export that opens in your browser. A familiar extension alone does not establish that the encoded image is valid or supported.
- The people or roles look wrong
- Remove the photo and choose the intended reference. Then match the left/right lyric labels to the preview again before saving the brief.
The browser checks the selected file locally. Do not treat an accepted preview as identity verification or as a prediction of singing quality. The server also validates photo bytes and metadata before an enabled generation workflow can submit them.
Prepare the words and save your plan
Once both people look correct, write a short topic and assign the lines. The original rap-video prompt examples show how to keep the story, lyrics and settings separate. Use the local lyric starter and line checker to review the draft’s structure.
You can download a text brief or an editable JSON brief. Those files contain creative text and settings, not the selected photos. Keep your source images separately and reselect them when continuing from an imported brief. Importing a brief provides a review step and does not submit a generation request.
Before any future finished video is shared, review the actual output: recognizable faces, which person delivers each line, mouth movement, missing words and audio. Photo preparation and a successful file check are only inputs to that review.