Why this matters
Fashion films are expensive and slow to produce, and when they go wrong the result looks cheap. AI video can cut both cost and time, but only if you control what the model does.
That control starts with how you give the model its input. The leading video models accept two kinds of input, and each one hands the model a different amount of creative freedom. Before you build a storyboard or write a single motion prompt, you need to know what each one does and when to use it.
The series at a glance
Over the next few lessons you’ll build a complete fashion film:
Two ways to generate video (this lesson): starting frame vs reference images
Building a storyboard: using a system prompt to lock camera language, pacing and color grade so every clip looks like part of the same film
Prompting for motion: bringing the static frames in your storyboard to life
The final cut: stitching clips together, adding music and finishing
What you’ll build
Two quick tests of the same idea:
a short clip generated from a starting frame, run through three video models so you can compare them
an 8-second clip generated from a single reference image (a character sheet), with the model making every creative decision
The two methods
Starting frame. You attach an image that becomes the very first frame of your clip. Your prompt then describes the motion, and the model builds the rest of the clip forward from that frame. This is the more common method, and most video models support it.
Reference images to video. You give the model a set of visual ingredients, such as a character, a garment, a location or a prop, as separate reference images. The model combines them with your text prompt to build a world around them. Fewer models support this: Seedance 2.5, Google Omni Flash and, more recently, Kling V4.
Step-by-step: Starting frame
1. Bring in your frame
Upload the image you want to animate. Treat it like a frame from your storyboard. The example frame here was generated with Nano Banana Pro.
2. Add a video node
Add a video node, place it next to your image node, and connect the image to the node’s start frame input.
3. Set your properties
In the properties panel on the right, set:
Duration: the length of the clip
Resolution: the output quality
Model: choose from the dropdown
4. Write a motion prompt
Your image already sets the look of the clip, so the prompt only needs to describe motion: what moves and how the camera behaves. For example:
Smooth and slow camera motion as the woman laughs and twirls around freely.
Keep it simple. You can also use the Copilot node to write motion prompts for you.
5. Run it
Click Run Selected.
6. Duplicate and compare models
Duplicate the video node and run the copy with a different model. Do this twice to get three results from identical inputs. In this lesson the comparison uses Seedance 2.5, Kling and Google Omni Flash.
Even with the same frame and the same prompt, each model produces a noticeably different result. Comparing them side by side is the fastest way to find the right model for a shot.
Tip: Duplicating a node keeps all of its connections. Copying and pasting a node does not. When you want to rerun the same inputs through a different model, always duplicate.
7. Choose your best moments
For a full film, you animate several frames from your storyboard this way as 5- to 8-second clips. You then take the best parts of each clip and assemble them in your editor. You’ll do that later in the series.
Step-by-step: Reference images to video
1. Prepare your references
Gather the visual ingredients for the clip. Here that’s a single character sheet showing the model wearing the dress being promoted. You could also add images of a location, props or anything else you want in the scene.
2. Connect them to a video node
Add a new video node (Seedance 2.5 is selected by default) and connect your image node to its reference images input.
3. Describe the world
Your prompt describes the scene the model should build around your references. Keep it short for now:
The woman in the character sheet on a beach near sunset, having fun. Fashion film style camera motion.
4. Set your properties
Set the duration (8 seconds), the resolution (720p) and the aspect ratio (16:9, horizontal), then click Run.
5. Review the result
From a one-line prompt and a single character sheet, the model produced a multi-shot clip. It decided the shots, the framing, the color grade and everything else on its own.
Which method to use
Starting frame | Reference images | |
|---|---|---|
What you provide | The exact first frame | Ingredients (character, garment, location) |
Who decides the shot | You | The model |
Effort | More: you build every frame | Less: one prompt, a few references |
Creative control | High | Low |
Best for | Directed films, campaigns, anything on-brand | Fast exploration, mood tests, quick social content |
The reference images method is easier and the results can be impressive, but the model makes the creative calls. For a fashion film, where every shot needs to look consistent and on-brand, you need to control each frame. That’s why the rest of this series uses the starting frame method.
Takeaway
Starting frames give you control. Reference images give you speed. Use reference images to explore an idea quickly, and starting frames when the brand needs every shot exactly right. In the next lesson you’ll build the storyboard those starting frames come from, with the color grade, lighting and overall look locked across every shot.