The reference is not respected
Make the prompt simpler and explicitly ask the model to preserve the subject, style, or product from the reference.
模型
REFERENCE IMAGES
JPEG/JPG/PNG/WEBP · up to 6 images · max 30MB each
REFERENCE VIDEOS
MP4/MOV · up to 3 videos · max 100MB each
REFERENCE AUDIO
MP3/WAV · up to 3 audio files · max 50MB each
Prompt (Optional)
画面比例
时长
分辨率
所需积分
275 积分
Reference to Video 演示视频
开始生成后，最新视频会显示在这里。
REFERENCE TO VIDEO AI
Quick answer
Reference to video uses an image as the visual guide and a prompt as the motion guide. It is best when the subject, style, product look, or character identity should come from a reference image.
Input requirements
Match the source type to a conservative first motion target. That improves the keeper rate before spending credits on longer clips.
| Input | Best for | Avoid |
|---|---|---|
| Product reference | Preserving product shape while adding camera or light motion | Requesting a different product design |
| Character reference | Keeping character look while changing pose or scene mood | Full redesigns that fight the reference |
| Style reference | Matching color, mood, or art direction | Conflicting style prompts |
Failure reasons
Make the prompt simpler and explicitly ask the model to preserve the subject, style, or product from the reference.
Use smaller motion and avoid asking for major transformations from one reference image.
Remove style words that contradict the reference and keep only the desired motion.
FAQ
It is a workflow where an image guides the look and a prompt guides the motion or scene direction.
Image to video focuses on animating the uploaded image itself. Reference to video can use the image as style or subject guidance for a generated clip.
Clear images with one subject, strong shape, and readable style work best.
Related pages
Animate the uploaded image directly when it should be the first frame.
Open image to videoPROMPTGenerate from text alone when no reference image is needed.
Open text to videoCONTROLUse an existing clip when timing and motion should come from video.
Open video to video