- A Higgsfield Katana reference video defines the movement, pacing, and camera behavior of your final AI edit.
- Pick references with one clear action and readable subject movement for the most stable results.
- Combine the reference with character, product, or clothing images to control the replacement subject.
- Use conversational follow-up prompts to refine results instead of regenerating from scratch.
- Always review the credit estimate before submitting a new generation request.
What a Higgsfield Katana Reference Video Does
In Higgsfield Katana, the reference video is the creative foundation of the entire edit. Instead of generating a clip from text alone, Katana rebuilds the visual direction of your reference—its motion, camera moves, timing, and shot rhythm—while swapping in your own character, product, wardrobe, or style. This is what separates Katana from a basic text-to-video generator or a conventional timeline editor.
A good reference answers three questions before you upload anything:
| Question | What It Controls | Why It Matters |
|---|---|---|
| What is the action? | Motion transfer and performance | One readable action produces more stable generations |
| How does the camera move? | Framing and shot progression | Clear camera direction is easier to preserve |
| What is the pacing? | Shot rhythm and edit timing | Critical for short-form ads and social clips |
Choose a reference that matches the type of content you want to make—a product ad reference for product ads, a fashion walk for fashion clips. Structure transfers best between similar content types.
Reference Video Selection Criteria
Not every clip works well as a Katana reference. Use this checklist of criteria when reviewing candidate footage.
| Criteria | Good Reference | Problem Reference |
|---|---|---|
| Subject visibility | Main action visible throughout | Heavy occlusion or off-screen action |
| Cuts | Minimal, purposeful cuts | Rapid, unnecessary cutting |
| Duration | Focused on the key action | Long footage with filler segments |
| Camera | Readable movement and framing | Shaky or ambiguous camera behavior |
| Rights | Owned or licensed media | Unverified third-party content |
Personalized Edits
- Reference guides structure while your subject takes over
- Best for product ads and character variations
- Requires a clear replacement image
Style Transfer
- Reference keeps motion and composition
- Color, lighting, and atmosphere change
- Best for fashion shorts and brand looks
Motion Transfer
- Reference performance carries over
- New character or product performs the action
- Best for action and performance clips
Make sure your character, product, and clothing images all describe the same intended subject. Mismatched references are a leading cause of results that drift away from the reference video.
Step-by-Step: From Reference to Result
Follow this six-step workflow to turn a reference video into a finished Katana generation.
Choose Your Reference Video
Select footage with clear subject movement and readable camera actions that matches your intended content type, such as a product ad, fashion clip, or action sequence.
Upload Creative Assets
Add character, product, or clothing images showing the replacement subject from useful angles, and describe each asset's role so Katana can distinguish it from the reference footage.
Select the Editing Direction
Decide between Personalized Edits, Style Transfer, or Motion Transfer based on whether the subject, visual identity, or movement is your priority.
Describe the Intended Changes
Use a concise prompt that states what must stay consistent—camera rhythm, main action—and what should change, such as character, product, palette, or wardrobe.
Generate and Review
Submit the request, then check whether the main action, framing, pacing, and replacement subject consistency match your expectations.
Refine with a Follow-Up Edit
Use targeted conversational instructions to adjust one element at a time, explicitly protecting the parts of the result that already work.
Each rerun is a new credit-consuming request. Refine your reference and instructions before regenerating, and check the displayed credit estimate every time.
Prompt Templates for Reference-Based Edits
Structure your prompt by separating what is preserved from what is replaced. These templates cover the most common scenarios.
| Scenario | Prompt Template |
|---|---|
| First generation | Use the reference video as the foundation. Replace the main subject with [subject]. Preserve movement, camera, and pacing. Set the scene in [environment] with [style]. |
| Product replacement | Replace the product with [product]. Keep the shot sequence, camera movement, and timing. Make the product visible with [details] in a polished [ad/social] presentation. |
| Character and wardrobe | Replace the character with [description]. Preserve performance and body movement. Dress in [wardrobe] with [colors, materials]. Keep framing and rhythm consistent. |
| Style transfer | Apply [style] to the reference. Preserve composition, movement, and pacing. Change palette to [colors], lighting to [lighting], atmosphere to [mood]. |
| Follow-up modification | Keep current camera, pacing, and main action unchanged. Change only [element]. Preserve [working elements]. |
Describe characters using visible traits rather than vague labels, and make one focused modification per follow-up request rather than stacking contradictory instructions.
Input Requirements and Pre-Upload Checklist
Upload failures and unstable generations usually trace back to preparation. Verify every input before submitting.
| Input Type | Key Requirements |
|---|---|
| Reference video | Plays correctly, movement visible, within duration and format limits shown in the interface |
| Character image | Recognizable view, no heavy blur or occlusion, permission to use confirmed |
| Product image | Unobstructed product, readable logo and details, consistent angles and lighting |
| Clothing image | Full garment visible, silhouette and material clear, no hiding shadows or folds |
| File links | Direct link, opens without extra login, resolves to the correct file |
Before You Generate:
- Reference video plays and shows one clear action
- All subject images are consistent with each other
- Prompt separates preserved elements from changed elements
- Link or upload tested and accessible
- Credit estimate reviewed before submitting
FAQ
Q: What makes a good Higgsfield Katana reference video?
A good reference shows one readable action with clear subject visibility, purposeful camera movement, and minimal unnecessary cuts. Duration should stay focused on the action you want transferred, and you must own or be licensed to use the footage.
Q: Why doesn't my generated video match the reference?
Reference transfer weakens when the action is unclear, subject images conflict, or the prompt combines contradictory directions. Use a reference with visible movement, keep all assets consistent, and iterate on one change at a time.
Q: Do retries consume extra credits?
Yes. Each new generation, including retries and revised requests, is a separate credit-consuming request. Review the displayed estimate before submitting and refine your inputs first to reduce wasted generations.
Q: Which editing direction should I choose?
Choose Personalized Edits to adapt the reference to your own subject, Style Transfer when the visual identity needs to change, and Motion Transfer when the reference performance is the priority for a new character or product.