Tools
Reference images, video and audio
A reference is a file you attach so the video copies something real: a photo for a face, product or place, a short MP4 clip for movement or camera, or an MP3/WAV file for a voice or music. Each video model (the AI engine that makes the video) takes its own number of each kind, and the studio shows that number on every add button.
Who can use it: any signed-in account in Cinema Studio. Developers sending files by API should read API reference inputs.
Updated · View as Markdown
Steps
- Pick the model first. Press the Model chip and choose. The add buttons below the text box change to match it, and a model may say Needs 1 photo or Needs photo or video.
- Add photos. Press Reference photos ({n} of {cap}) and pick picture files, such as JPG, PNG or WebP. The line under it says "Up to {cap} photos for this model". Each photo can be up to 10 MB. You can drag thumbnails to change their order.
- Add a video clip (if your model takes one). Press Add a reference video and pick an MP4. The line under it says "Up to {cap} MP4 clips under {MB}MB each, {sec} seconds total". A clip longer than the total shows "Will be trimmed to {sec}s automatically".
- Add sound (if your model takes it). Press Add reference audio and pick an MP3 or WAV file ("MP3 or WAV, up to {MB}MB each"). Some models need at least one photo or video next to the sound.
- Point at each file in your words. Type
@in the text box. A list opens ("Add a reference"); pick a file and a tag goes into your prompt:@图片1for photo 1,@视频1for video 1,@音频1for audio 1. Then write what it is for, for example "the woman in @图片1 walks like the dancer in @视频1". - Check the price and press Generate. If the model list showed Ref Video · {x}× price, the button already includes that.
At a glance
| Kind | File types | Size per file | How many |
|---|---|---|---|
| Photo | Picture files such as JPG, PNG, WebP | 10 MB | 1 to 30, by model |
| Video | MP4 | 64 MB on most models, 60 MB on some; between 409,600 and 8,847,360 pixels (for example 640×640 up to 4096×2160) | 0 to 10, by model, with a total-seconds limit |
| Audio | MP3, WAV | 50 MB on most models | 0 to 10, by model |
What each model takes, read on 2026-10-05 from the public model list (photos / videos with total seconds / audio):
| Model | Photos | Videos | Audio |
|---|---|---|---|
| Seedance 2.5 | 30 | 10, 30 s total, 60 MB each | 10 |
| Seedance 2.5 720P VIP | 30 | none | 10 |
| seedance 2.5 720p (30 s only) | 9 | none | none |
| Seedance 2 | 9 | 3, 15 s total | none |
| Wan 3.0 | 10 | 5, 15 s total | 5, needs a photo or video too |
| MiniMax H3 | 9 | 3, 15 s total, 50 MB each | 3, 15 MB each, needs a photo or video too |
| Kling V3 | 4 | none | none |
| Veo 3.1 Fast | 3 | none | none |
| Hailuo 2.3 | 1 | none | none |
Model names are shown exactly as the model list in Cinema Studio prints them.
Two Seedance 2.5 720p models look alike but work differently:
- Seedance 2.5 720P VIP makes videos of any length from 4 to 30 seconds in six frame shapes, takes up to 30 photos and 10 sound files but no video clips, and cost 102 credits for 5 seconds on 2026-10-05.
- seedance 2.5 720p makes only 30-second videos in 9:16, 16:9 or 1:1, takes up to 9 photos and no video or sound files, and cost 354 credits for its 30 seconds on 2026-10-05.
Every model's caps, lengths and prices, side by side, are on AI video models. Developers sending files by API: /docs/api/models.
Use when / Avoid when
| Use references when | Skip them when |
|---|---|
| The same face or product must look right in every shot | You only need a mood or a scene you can describe in words |
| You want to copy a real movement or camera move from a clip | Your model shows "none" for that kind of file — pick another model first |
| A voice or music track should drive the video | The photo shows a real person and the model says it cannot use real people (see below) |
Troubleshooting
| Message on screen | What it means | What to do |
|---|---|---|
| "Only {x} of {y} photos added — this model takes up to {cap}." | You picked more photos than the model takes | Keep the best ones, or switch to a model with a higher cap |
| "Reference images trimmed to {cap} for {model}." | You switched to a model with a lower cap | Check which photos stayed |
| "Reference videos removed because this model does not support them." | The new model takes no video | Switch back, or describe the move in words |
| "Reference audio removed because this model does not support it." | The new model takes no audio | Switch to a model with Add reference audio |
| "That video is {w} x {h}. Use a video between 409,600 and 8,847,360 pixels." | The clip is too small or too large in picture size | Export it at a normal size such as 720p or 1080p |
| "Reference video too big — use one under {MB}MB." | The file is over the size limit | Shorten or compress the clip |
| "Reference videos can be up to {sec} seconds total." | All clips together are too long | Remove or trim a clip |
| "Reference audio too big — use one under {MB}MB." | The sound file is over the limit | Use a shorter or lower-quality MP3 |
| "Add at least one reference photo or video with reference audio." | This model needs a picture next to the sound | Add one photo or clip |
| "Use reference images or reference videos, not both." | This model takes one kind at a time | Remove the photos or the clips |
| "Use reference-photo mode when adding a reference video." | Image upload mode is set to First and last frames | Set it back to Reference in Model parameters |
| "This model can't use photos of real people — try a different photo or model" | The model refused a photo of a real person after you pressed Generate; your credits come back | Use a drawn, anime or 3D character picture, or another model |
| "A reference file was rejected by a copyright check. Retrying will not help — replace that reference, then render again. Your credits are returned." | A file looked like protected material | Replace that file |
More failure messages: Every error message and what to do. See plans and prices.
FAQ
How many reference photos can I add?
It depends on the model: from 1 (Hailuo 2.3) up to 30 (Seedance 2.5). The chip shows the cap, for example Reference photos 0 of 9.
Why did my reference video disappear when I changed model?
The new model does not take video references, so the studio removes them and says so. Switch back to a model that lists Add a reference video.
Does a reference video cost more?
On some models, yes. Those models carry a tag like Ref Video · 1.8× price in the model list, and the Generate button shows the new total.
How do I tell the model which photo is which?
Type @ in the prompt and pick the file. A tag such as @图片1 (photo 1) or @视频1 (video 1) goes into your words, so you can write what each file is for.