Claude and Higgsfield: combine Seedance footage, images and music

A person appears, speaks, and the video ends with an animated service interface. That result may involve several different tools.

Showhey shared a setup combining Claude Code, Higgsfield’s Seedance, image generation and Lyria through Gemini, and Remotion. Creator’s breakdown

This guide explains the roles, then provides a setup for connecting Higgsfield, obtaining a short character clip and adding music and graphics. It uses official instructions checked September 27, 2026, rather than reconstructing undisclosed creator settings.

Reference video using this combination

{{tweet:https://x.com/showheyohtaki/status/2103552919249563691}}

The steps below are a suggested workflow based on public information, not the creator’s undisclosed prompts or production settings.

The Claude and Higgsfield production workflow

Claude coordinates the work; Higgsfield generates images and footage. With separate music and interface tools, the process becomes three stages: assets, moving scenes and editing.

Stage	Tools named in the public example	Output to pass onward
Character and set images	Nano Banana Pro through Gemini API	Images
Footage and voice	Seedance 2.5 through Higgsfield	Video
Background music	Lyria 3.5 through Gemini API	Audio
Interface motion	Remotion	Composition code or footage

This table describes the creator’s reported setup. A first trial does not need every service automatically connected. Here we start with one Higgsfield scene, save approved assets and move into editing.

Prepare Claude Code, Node.js and an active paid Higgsfield account with usage available. Remotion is used later. The following route uses the officially documented CLI; it does not assume the original creator used the same connection.

Run these commands in a terminal and complete sign-in in the browser that opens. Official CLI instructions

npm i -g @higgsfield/cli
higgsfield auth login
npx skills add higgsfield-ai/skills

Start Claude Code and ask it to retrieve your Higgsfield balance and recent generations without generating anything new. A real tool call returning account information confirms the connection.

Turn a character image into a Seedance clip with sound

Approving a character image first lets you check appearance and clothing before generating motion. Seedance 2.5 is documented as supporting audio synchronized with video. Official model guide

Give Claude this proposed image request. If the requested model is unavailable, have it present alternatives instead of silently generating with another model.

I want one Higgsfield image of a
fictional adult holding a plain white
mug in a cafe.
Landscape 16:9, natural light, navy
clothing, no text or logos.
Check whether Nano Banana Pro is
available and show the credit
conditions.
Wait for confirmation, then generate
exactly one image.

After approving the image, identify that asset explicitly by attachment or the returned asset ID:

Use the approved image as a reference
for one Seedance 2.5 video.
The person slightly raises the cup and
smiles. Fixed camera.
Requested format: 16:9, 5 seconds.
Spoken line: “Time for a little break.”
No background music. First check
supported duration, resolution and audio
settings
for this connection and show the actual
generation conditions.
After approval, generate once and check
that job until it finishes.
Return its generation ID and download
link.

Five seconds and the spoken line are requested trial settings. If the connection supports different options, adjust before generation. Play the result independently and inspect the face, hands, cup, lip movement and spoken words.

Save the clip as public/scene.mp4 in the Remotion project. If a generation is still processing, check the same job ID before requesting another one; another request may start another generation.

Combine Lyria music and Remotion graphics

Once footage and music are ready, Remotion aligns them and adds final text or graphics. This separates asset generation from editing the explanation.

Follow the Lyria MP3 setup to create music in Gemini and save it as public/bgm.mp3. Install Remotion using this guide.

Open Claude Code in the project containing those assets and request:

/remotion-best-practices
Use public/scene.mp4 and public/bgm.mp3
for a 10-second introduction.
Composition ID: CafeDemo. Size:
1920×1080. Rate: 30fps.
Use the character clip for the first 5
seconds. Preserve its voice
and keep the background music quieter
than the speech.
For the final 5 seconds, show “Time for
a break.” with simple circular graphics.
Do not present the graphics as a real
app screen or customer result.
No new asset generation. Plan first,
then build and preview after approval.
Return the entry file and render
command.

If the project uses src/index.ts, render after review with:

npx remotion render src/index.ts CafeDemo out/cafe-demo.mp4

Play out/cafe-demo.mp4 and check speech audibility and the transition between halves. When adapting this to a real product, compare the generated product’s appearance with the actual item too.

Higgsfield MCP setup and generation credits

In browser-based Claude, use a custom MCP connector instead of the CLI. You do not need to install both routes just to make one work.

In Claude, open Settings → Connectors → Add custom connector. Name it Higgsfield, enter the URL below, connect and complete Higgsfield authentication. Test a balance request in a new chat. Official connection guide

https://mcp.higgsfield.ai/mcp

Generation consumes Higgsfield credits. Not typing an API key does not mean free generation. Model, resolution and duration affect the conditions, so check them before the first clip.

Symptom	What to check
No account information	Authentication and connector activation
Requested model unavailable	Availability through this connection route
Generation appears unfinished	The existing job ID and status
Different person or product	Reference asset, result and specific scene to revise

Save approved images and footage, then distinguish changes that only need editing from those needing regeneration. If text is the main subject, a code-based workflow may fit the job better.