Footage over prompts
The reference clip is the direction — camera moves and cut timing carry over exactly.
Katana AI
Reference in. Your edit out. Upload a clip, add your subjects, and the rebuild keeps the motion, framing and rhythm.
KI-Modell 4 Credits
GPT Image 2Vielseitige Bildgenerierung bis 4K Anweisungen0/2000
Vermeide urheberrechtlich geschützte Figuren, Logos und Musik sowie sensible oder explizite Inhalte — sie können blockiert werden.
Auflösung
Seitenverhältnis
Stapelgenerierung1
Geschätzte Kosten: 4 Credits
Dein Edit erscheint hier
Wähle eine Referenz, füge deine Motive hinzu und tippe auf Generieren.
Why Katana AI
Why creators choose Katana AI
Katana AI is a reference-to-video studio: the clip supplies the direction, your photos supply the cast.
Subjects keep identity
Photos anchor faces, products and outfits through the whole rebuild.
Preset library included
Eight proven reference clips ship in the studio — no footage needed to start.
Optional audio
Attach the track your edit should keep, or export silent and score it later.
Katana AI guide
How a reference clip becomes your edit
Katana AI does one thing: it takes a short video you provide and rebuilds it around your own subjects. The reference anchors the motion — camera path, cuts, rhythm — while your photos decide who or what appears in the frame. This guide covers what to upload, how the rebuild works, and how to get a take worth posting.
Why the reference does the heavy lifting
The pipeline is reference-to-video: your clip becomes the blueprint, so the generated edit moves exactly the way the source did. That is why clip choice matters more than instruction writing here — four to fifteen seconds with one clear move gives the engine choreography it can hold.
The same trick works across subjects. People, products, outfits and characters all rebuild as long as the subject photos are clear and you own the footage or have permission to use it.
One clear move
A walk, an orbit, a reveal — stacked choreography confuses the rebuild.
Stable framing wins
The camera path carries over, so smooth reference footage produces smooth edits.
Readable subject
When the source subject is easy to parse, the swap lands cleaner.
What subject photos actually do
Subject photos are the casting call. One to eight images tell the engine who or what should appear — and unlike a text prompt, they carry identity: a face, a logo, a fabric texture. More angles of the same subject give the rebuild more to hold onto.
One subject per run works best. The studio accepts up to eight photos so you can cover angles — front, side, detail — not so you can cast a crowd.
Getting a take worth posting
AI video varies run to run — the same reference and photos produce slightly different edits. The usable workflow is to review like an editor: check subject identity first, then the timing against the reference. A take where both hold is the keeper.
Draft at 480p to check the idea, finish at 720p or 1080p for posting. Downloads are plain MP4s — vertical, landscape or square — ready for whatever app you publish from.
Rights, credits, and cost
Every render asks you to confirm that you own or have permission to use the reference and the photos — including consent from anyone shown. That checkbox is the deal: the format is powerful precisely because it can rebuild real footage, so the rights have to be real too.
New accounts start with free credits. After that, edits bill per second of video at the rate of your chosen resolution, and the studio shows the exact estimate before every run. Failed generations refund automatically.
How to use Katana AI
How to use Katana AI in three steps
Upload the reference clip
4–15 seconds of footage — yours or a preset — sets the motion and rhythm.
Add subject photos
1–8 images of the person, product or outfit that should appear.
Generate and review
Check identity and timing, then download the MP4.
Who uses Katana AI
Who uses Katana AI for image work
Creators
Rebuild trending formats with your own cast — no reshoot needed.
Brands
Product photos step into proven commercial motion.
Social teams
Feed-sized edits that match formats already working.
Fans
Put yourself inside the edits you love — with rights confirmed.
More models
Katana AI and the rest of the studio in one place
Katana Edit handles the video pipeline end to end; companion image models cover thumbnails, covers and stills.
Reference-to-videoMotion transferSubject photosOptional audio4–15s editsImage generationMP4 exportPreset library
FAQ
Katana AI FAQ
01What does Katana AI do?
It rebuilds a short reference video around your own subjects — the motion, camera work and cut timing carry over while the people, products or outfits come from your photos.
02What are the upload limits?
One reference clip of 4–15 seconds, 1–8 subject photos, and optionally one audio file. Output length matches the reference.
03Is Katana AI made by Higgsfield?
No — Katana AI is an independent product inspired by the reference-to-video format. It is not affiliated with Higgsfield or Anthropic.
04Do I need to write a prompt?
No. The reference clip carries the direction. A short instruction field is there if you want to steer emphasis, but it is optional.