Last updated by Wing Chan on August 20, 2026.
Video Generation creates or edits a video using Generative AI. Use the timeline editor when you want to add text, logos, images, clips, or audio onto an existing generated or uploaded video.
Riverflow subscription grant you access to the leading generative video models like Seedance 2.5, Minimax H3, Kling O3, Google Veo and much more. Some subscriptions include bonus free video generations each month too without using credits.
Step-by-step tutorial:
Before you start
Decide whether you want to start from text, one image, two frames, or multiple references.
Prepare any source images, videos, or references you want to use.
Know the target aspect ratio, duration, and channel if the video is for a specific placement.
Keep the motion idea simple enough for a short generation.
Choose the right video mode
Text to Video
Use Text to Video when you want to describe the full video from scratch.
This is best for exploration, mood, environment, or simple product moments where you do not need to preserve an exact starting frame.
Image to Video
Use Image to Video when you already have a strong still image and want to animate it.
This is usually the best starting point for product work because the product, composition, and style are already established in the source image.
Start and End Frame
Use Start and End Frame when you want the video to move from one image state to another.
This is useful for transitions, reveals, product movement, before/after ideas, or controlled motion between two visual states.
Multi-Reference Video
Use Multi-Reference Video when you want several reference inputs to guide one video. It is available with Seedance 2.0 and works best when each reference has a clear role, such as product identity, character appearance, motion, camera movement, style, setting, or audio pacing.
Generate from one image
Open
Video, or open a supported image and choose the video generation action.Choose Image to Video.
Select or upload the source image.
Choose the available model and output settings.
Add a short motion prompt.
Generate.
Example prompt:
Slow camera push-in. The product remains sharp and centred. Subtle condensation moves naturally on the can. Keep the label readable and do not change the packaging.
Generate from start and end frames
Choose Start and End Frame.
Select the start frame.
Select the end frame.
Write a short prompt describing the movement between them.
Set duration and other available options.
Generate.
Example prompt:
Transition smoothly from the first frame to the second frame. A hand enters naturally and places the product on the surface. Keep the product packaging accurate and preserve the lighting style.
Generate from text
Open
Video.Choose Text to Video.
Write a specific visual prompt.
Choose the available model and output settings.
Generate.
Text to Video works best when the prompt describes the subject, setting, camera movement, and action clearly.
Example prompt:
A short cinematic product video of a matte black coffee tin on a warm wooden kitchen counter. Soft morning light comes from the left. Slow camera push-in. Steam rises gently from a mug beside it. Keep the product label readable.
Use motion presets
If motion presets are available in the workspace, use them to start from a common motion direction. A preset can fill or guide the motion prompt. Review the prompt before generating and adjust it if the product needs specific protection.
Use Multi-Reference Video
Use Multi-Reference Video when one source asset is not enough to explain the video you want. You can combine references so one asset shows the product, another shows the person or style, another shows the motion or camera movement, and another guides the audio or rhythm.
Multi-Reference Video is available with Seedance 2.0.
You can include up to 12 references in total:
up to 9 image references
up to 3 video references
up to 3 audio references
Use image references for product identity, packaging, character appearance, styling, lighting, setting, or composition. Use video references for motion, camera movement, pacing, transitions, or action. Use audio references for rhythm, spoken audio, soundtrack direction, or atmosphere.
Keep each reference focused. A small number of clear references usually works better than many conflicting ones.
Tag human references
If an image or video reference contains a person or human-like subject, treat it as a human reference.
Before generating with Seedance 2.0, Riverflow will ask you to review eligible image and video references. Select any reference that contains one synthetic person. Leave product-only, scene-only, packaging, logo, audio, or non-human references unselected.
Riverflow can currently register one synthetic person per reference. Multi-person references are not supported in this flow.
Fix a synthetic-person reference video pixel error
If Riverflow says that a reference video containing a synthetic person must contain between 409,600 and 927,408 pixels, the video's width multiplied by its height is outside the range Seedance 2.0 accepts for this type of reference.
Resize the video while preserving its aspect ratio.
Keep
width × heightbetween 409,600 and 927,408 pixels.For example,
720 × 1280for a 9:16 video or1280 × 720for a 16:9 video equals 921,600 pixels and meets this requirement.Export the resized video, upload it again, then select it as the synthetic-person reference.
Riverflow does not currently resize these reference videos automatically.
Refer to references in the prompt
Tell Riverflow what each reference should control. Use the reference labels from the control panel, such as Image 1, Video 1, or Audio 1, and give each one a specific job.
Example prompt:
Use Image 1 for the product shape, packaging, and label.
Use Image 2 for the studio lighting and background mood.
Use Video 1 for the slow push-in camera movement.
Use Audio 1 for the rhythm and pace.
Create a 9:16 product video where the product rotates slowly on a clean studio surface. Keep the packaging readable and do not copy unrelated props from the references.
Use this approach when you want reference videos to improve output quality without making the model guess which part of the reference matters.
Model and mode notes
Available models and settings can vary by mode. Choose the model shown in the workspace for the type of video you are creating.
For Seedance access, keep using the dedicated Access to Seedance 2.0 Uncensored article.
Tips for better results
Start with a strong still image when product accuracy matters.
Match the output aspect ratio to the source image or intended placement to avoid awkward framing or black bars.
Keep prompts short and visual.
Describe camera movement and subject movement separately.
Protect important product details in the prompt.
Avoid asking for too many actions in one short video.
Review product shape, text, and logo details before using a video publicly.
Review before use
Check:
product accuracy
label and logo readability
motion quality
camera movement
aspect ratio
duration
unwanted extra objects
whether the clip can be edited or combined with other clips
What to do next
See Use the Video Timeline Editor to add and time text, logos, images, video clips, or audio onto a generated or uploaded video.