Skip to content
trending

Guides

Higgsfield Katana: reference editing in Claude

A practical guide to Higgsfield Katana for reference-based video editing inside Claude. Covers MCP setup, asset limits, current plan pricing, and an original workflow for preparing reference clips and photos.

By 10 min read

Update (9 Nov 2026): Claude Fable 5.2 and Mythos 5.2 are out - prices, access and benchmarks are in my Claude Fable 5.2 guide.

👁 117 reads

Rebuilding an existing video edit with new products or talent usually requires manual timeline cutting, shot matching, and repeated rendering. Higgsfield Katana addresses this bottleneck by operating as a specialized skill inside Claude via the Model Context Protocol (MCP). According to official Higgsfield documentation published on October 12, 2026, Katana takes a reference video - either an uploaded file or a library preset - along with up to five reference images of characters, products, or clothing, and reconstructs the edit structure using the user’s assets.

Try Higgsfield Before buying, check the included models, credits, billing period and total payment, including taxes, on the official pricing page.

Open Higgsfield

Partner link

This guide separates documented vendor mechanics from practical production strategy. Below, readers will find the exact technical specifications for Katana, how it connects to a standard Higgsfield AI account, current subscription credit costs, and an original proposed workflow for preparing assets and troubleshooting shot replacements without wasting credits.

What Is Higgsfield Katana and How Does the Architecture Work?

Documented by the vendor as an MCP-powered skill for Claude, Katana is designed for creators who already have a working visual structure - such as a fast-paced product showcase or a rhythmic fashion montage - and want to replicate its shot order, cut timing, and pacing with different subjects. Rather than generating standalone clips in isolation, the tool reads the temporal structure and audio of the reference edit and places the user’s uploaded subjects into the corresponding shots.

To understand where Katana fits within the broader platform, it helps to compare its inputs and outputs against standard web generation and other MCP skills.

Feature or Parameter Documented Specification Operational Note
Access Method Claude (claude.ai or Desktop) via MCP, or Higgsfield Requires active paid subscription; no separate API key needed
Reference Input 1 uploaded video file, direct media link, or library preset Links to TikTok or Instagram profiles are not accepted
Subject Assets Up to 5 photos (characters, products, clothing) Uploaded via the Higgsfield window inside Claude, not standard chat attachment
Output Format MP4 matching reference duration, aspect ratio, and frame rate Saved simultaneously in the Claude chat widget and Higgsfield Assets
Community Presets 3 to 30 sec motion video, up to 40 images, prompt with @ mentions Published from the Katana page; subject to takedown upon report

A critical technical distinction documented by Higgsfield involves file handling inside the chat interface. Standard image attachments dropped directly into Claude’s native message bar are not passed through to the MCP server as Katana reference assets. Instead, Katana triggers a dedicated Higgsfield upload window inside the conversation, which transfers the files directly to the user’s Higgsfield account. Direct URLs pointing to raw media files are also supported.

Connecting Katana in Claude and Entry Points

Katana relies on the Higgsfield MCP connector. According to the vendor’s setup documentation, connecting the tool does not require generating a developer API key, provided the user holds an active subscription.

  1. Open claude.ai or the Claude Desktop application and navigate to Settings -> Connectors.
  2. Select Add custom connector.
  3. Enter the name Higgsfield and paste the official server endpoint: https://mcp.higgsfield.ai/mcp.
  4. Click Connect to launch the authorization window and sign in to the active Higgsfield account.

Once the connector is active, the documentation notes two distinct ways to initiate a session:

  • Via the Web Form: Navigating to higgsfield.ai/katana presents a three-part form containing fields for Add a reference video, Add your characters, and an optional Instructions toggle. Clicking Send & Generate in Claude transfers the structured brief into a chat session.
  • Directly in Claude: Users can type /katana to browse the preset catalog, /katana/<preset-name> to execute a specific template, or write a natural-language request asking to make an edit using Higgsfield.

Within the preset library, templates are categorized along two axes: Category (Personalized edits, Style transfer, Motion transfer) and Source (Higgsfield or Community).

Pricing, Credits, and Account Rules (October 2026)

Running Katana inside Claude does not incur an MCP surcharge or separate per-call API billing; it draws directly from the credit balance of the user’s primary subscription. Before any video is generated, Claude displays a confirmation card showing the exact credit deduction required for that task. Follow-up conversational edits that require regenerating specific shots will consume additional credits.

Note that while some older Higgsfield blog posts from October 2026 still reference legacy Plus tier figures ($49 for 1,000 credits), the current authoritative pricing page as of October 24, 2026 establishes the following structure:

Plan Tier Monthly Billing (USD) Annual Billing Monthly Equivalent Monthly Credits
Free $0 $0 0 in comparison table (limited trial use on web)
Starter $19/mo $19/mo 270 credits
Plus $59/mo $47/mo 1,200 credits
Ultra 3000 $129/mo $99/mo 3,000 credits

All listed USD prices exclude VAT and local taxes. Subscription credits do not roll over to the next billing period. Furthermore, platform rules specify that “unlimited” or free promotional generations apply only on the main higgsfield.ai web interface and do not extend to MCP, CLI, Canvas, or Supercomputer workflows. Paid plans include commercial usage rights as a pricing feature, though users remain fully responsible for securing rights and permissions for all uploaded reference videos, audio tracks, logos, and likenesses. For a deeper breakdown of credit math across different models, see the Higgsfield AI pricing guide.

Check the current price Plans change often. The official pricing page shows today's price, and taxes are added at checkout.

See Higgsfield plans

Partner link

Original Proposed Workflow: Preparing Assets for Reference Editing

The following section outlines an original proposed workflow and preparation strategy designed to help users organize inputs before launching Katana. These steps are practical recommendations, not vendor-guaranteed system behaviors.

Because Katana inherits the duration, aspect ratio, and frame rate directly from the reference video - and restricts asset uploads to five photos per edit - unplanned uploads can easily lead to mismatched framing or unnecessary credit spend on corrective regenerations. A structured pre-flight workflow minimizes friction.

Step 1: Audit and Trim the Reference Video

Before opening Claude, inspect the reference video locally. Because the output will mirror the reference’s technical container, verify that the source file already matches the target delivery format (for example, 9:16 vertical at 30 frames per second). If only a 10-second segment of a 25-second clip is needed, trim the video file locally before uploading so credits are not spent rebuilding unwanted shots.

Step 2: Map the Five-Photo Slot Allocation

Katana accepts up to five photos to define characters, products, or clothing. Rather than uploading five nearly identical front-facing portraits, allocate the five slots based on the shot variety present in the reference video. Below is a suggested allocation matrix for different production scenarios:

Project Type Slot 1 & 2 Allocation Slot 3 & 4 Allocation Slot 5 Allocation
Apparel / Fashion Edit Full-body front and 3/4 angle of the primary talent Close-up of garment texture and primary silhouette Secondary accessory or footwear detail shown in cutaways
Physical Product Ad Hero product front elevation on clean background Product angled view showing depth and scale Human hand or talent holding the product for scale context
Character Montage Neutral medium shot establishing facial structure Side profile or dynamic posture matching reference action Key prop or environment anchor element

Step 3: Draft Explicit Keep-or-Change Instructions

When using the Instructions (optional) field on the Katana page or writing the initial prompt in Claude, clearly separate structural preservation from asset replacement. Specify which shot index holds the hero product and state which background elements should remain untouched versus replaced.

Step 4: Execute Targeted Single-Shot Corrections

Once the initial MP4 appears in the Claude widget, review it shot by shot. Because vendor documentation confirms that plain-message follow-ups can regenerate only the affected shots, avoid vague complaints like “make it look better.” Instead, reference the specific cut or subject photo that needs adjustment, check the credit card displayed by Claude before confirming, and iterate incrementally.

Suggested Prompt Template for Claude

The prompt below is an original suggested template for initiating a custom product edit inside Claude once the Higgsfield MCP connector is active. It has not been executed in a paid generation test and serves as a structural example of how to organize instructions.

Suggested Prompt (Not Executed):
/katana I am uploading a 12-second 9:16 reference edit via the Higgsfield upload window, along with 4 asset photos:
- Photo 1: Matte black insulated water bottle (front view)
- Photo 2: Matte black insulated water bottle (angled top cap detail)
- Photo 3: Athletic talent in grey hoodie (medium shot)
- Photo 4: Close-up of bottle strap texture

Instructions:
1. Rebuild the reference video keeping its exact cut timing and camera pacing.
2. Place the athletic talent (Photo 3) into the opening and closing lifestyle shots.
3. Use Photos 1, 2, and 4 for the rapid tabletop cutaway shots in the middle section.
4. Keep the high-contrast lighting style of the reference edit, and keep the bottle geometry consistent across all cuts.

Troubleshooting Decision Logic: Katana vs Other Tools

Because the Higgsfield MCP connector exposes multiple skills in the same chat environment, choosing the wrong tool for a task can waste credits. The following original decision logic helps determine when to stay in Katana and when to switch:

  • Problem: Files attached in Claude are ignored by the skill.
    Action: Do not use Claude’s standard paperclip attachment for Katana assets if they fail to register. Wait for or trigger the dedicated Higgsfield upload window inside the conversation, or supply a direct URL to the hosted media file. Remember that TikTok and Instagram profile links will be rejected.
  • Problem: You have no reference edit, only a “What if…” story concept.
    Action: Switch from Katana to Content Factory (accessible via /content-factory in Claude). Content Factory is documented for scripting and assembling faceless narrative videos from 15 seconds to 5 minutes, whereas Katana requires an existing visual structure.
  • Problem: You need precise multi-keyframe interpolation rather than edit reconstruction.
    Action: Use the web interface for Kling 4.0 under Create Video -> Frames, which supports up to 10 keyframes across a 3 to 30 second clip. Katana is built to match shot cuts and pacing from a finished video, not to interpolate between arbitrary still keyframes without a motion reference.
  • Problem: Only one object in a single continuous clip needs swapping.
    Action: Consider invoking Genjutsu (Object Swap or Motion Transfer) inside the same Claude MCP session. As noted in the documentation, clips modified in Genjutsu are saved to the shared Higgsfield Assets library and can subsequently be fed into Katana as reference material.

Pre-Flight and Output Review Checklist

Use this practical checklist before confirming the credit card in Claude and after receiving the rendered MP4:

  • Rights Verification: Confirm ownership or explicit usage permission for the reference video, including its background audio, music track, choreography, and any visible trademarks.
  • Container Check: Verify the reference file’s aspect ratio, duration, and frame rate, as Claude does not provide separate settings to override these parameters during a Katana run.
  • Upload Channel: Ensure all character, clothing, and product images (maximum 5) are uploaded through the Higgsfield window or via direct file URL.
  • Credit Card Inspection: Read the pre-generation cost card in Claude before approving the run, as well as before approving any follow-up shot corrections.
  • Cut and Asset Review: In the returned MP4 preview widget, check whether cut points align with the reference audio and verify that product proportions remain consistent across fast transitions before downloading from Assets.

If you want to explore the platform and set up the connector, you can check current plans on Higgsfield AI. Affiliate disclosure: Some links in this guide are sponsored affiliate links. If you purchase a subscription through them, the publisher may earn a commission at no extra cost to you.

Frequently Asked Questions (FAQ)

Does Higgsfield Katana require a separate paid API key to work in Claude?

No. According to official documentation, Katana connects through the custom connector URL (https://mcp.higgsfield.ai/mcp) in Claude Settings and authenticates via your existing paid Higgsfield account login.

How many photos can I upload to a Katana edit?

For standard reference editing in Claude, you can upload up to five photos of characters, products, or clothing. If you are publishing a shared preset to the Community library from the Katana web page, the preset format supports a 3 to 30 second motion video and up to 40 images referenced with @ mentions.

No. Katana does not accept links to TikTok or Instagram profiles. You must upload the reference edit as a video file through the Higgsfield upload window, provide a direct link to a raw media file, or select a preset from the Katana library.

How can I change the aspect ratio or frame rate of a Katana output inside Claude?

You cannot set duration, aspect ratio, or frame rate separately in the Claude chat for Katana. The output MP4 automatically inherits those parameters from your reference video, so any format changes must be made to the reference file before uploading.

Are Katana generations in Claude covered by Higgsfield’s unlimited generation promos?

No. Higgsfield pricing rules specify that unlimited or free promotional generations apply strictly on higgsfield.ai and do not apply to MCP/CLI connections, Canvas, or Supercomputer. All Katana generations and shot regenerations in Claude deduct credits from your plan balance after showing a cost preview card.

Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

Sources

The link to Higgsfield is a partner link: if you buy through it, QUANTSLANT earns a commission and you pay the same price. It has no influence on the content, including where the coverage is critical.

// read next

More on this from QUANTSLANT

Piotr Olszewski

Editor and publisher · Quantslant

I edit and publish Quantslant. AI tools help with research, drafts and translation; choosing topics, running the tests, checking facts and the decision to publish are mine, and so is the responsibility for what appears here.

read nextI tested Kling 4.0 on 10 clipsnewsI tested Kling 4.0 on 10 clipsTen clips reveal strong Polish speech and useful character continuity, alongside an incomplete camera orbit and stubborn reference props.

Most read

  1. 1Best AI Video Generator Apps for Your Phone (2026)
  2. 2Claude Opus 5.5 Motion Graphics: 60s Showreel from One Prompt
  3. 3Wan 2.2 on RTX 4090: timings, VRAM and four test clips
// site map
Quantslant in numbers 1,427posts in archive26news from the last 7 days18video models in the observatory
🎬 AI Video Benchmarks: Higgsfield DoP vs OpenAI Sora • Higgsfield vs Runway Gen-3 • AI Video Credit Calculator
× ‹ enlarge ›