How to Automate YouTube Visuals With Claude Code

TL;DR

Use timestamped audio transcripts to automate the creation and placement of video visuals with Claude Code and the Higgsfield skill. Record the narration, transcribe it with timestamps, then prompt Claude Code to generate a 16:9 image for every timestamp and name each file with its corresponding time so you know where it belongs in the edit.

Transcript

I just pasted my YouTube script into cloud code. Give it one prompt and it helped me edit my entire YouTube video for me. So, instead of spending hours or even days editing your YouTube video, you can now do it within minutes using clot code. So, in this video, I'll show you how to set everything up, what prompt to use, and how to turn a simple You... Read More

Key Insights

• Timestamped transcripts are the foundation of this automation workflow because they connect individual sections of narration with specific moments in the video. A plain script lacks these timing markers, so the creator first records the audio and then transcribes that recording.
• The Higgsfield skill is installed through its CLI by copying commands into Terminal or PowerShell. The setup includes installing the package, signing in, authorizing the account connection, selecting Higgsfield Generate, and choosing project-level installation with the symbolic-link method.
• Claude Code can verify whether the Higgsfield skill is installed by responding to a direct question about the installation. The demonstrated setup then uses the connected skill to generate footage or images, including a test image showing Earth as viewed from space.
• Recorded audio is required before visual generation because the workflow depends on timestamps derived from the spoken narration. Even when a creator already has the original script, the narration must be recorded and processed to obtain the timing information used for image creation.
• TurboScribe provides timestamped transcription after an audio file is uploaded and processed. The demonstrated free option offers three generations, and processing is described as taking roughly 30 seconds to one minute before the completed transcript becomes available.
• Directly copying the transcript preserves its timestamps, while downloading it as a PDF does not preserve the required timing information in the demonstrated workflow. The copied transcript is pasted beneath the generation instructions inside Claude Code.
• The core prompt requests one image for every timestamp and specifies the GPT Image 2 model through Higgsfield. It also requires every visual to use a horizontal 16:9 format, matching the layout requested for the YouTube video.
• Timestamp-based filenames make generated visuals easier to place because each filename identifies its intended position in the video. Examples in the demonstration include names corresponding to zero seconds and three seconds, allowing the creator to map images back to the transcript.

Questions & Answers

Q: How can Claude Code automate visuals for a YouTube video?

Claude Code can automate visual creation by combining a timestamped narration transcript with the Higgsfield generation skill. The creator supplies a prompt requesting one image for every timestamp, specifies a horizontal 16:9 format, and asks that each image use its timestamp as the filename. The resulting names indicate where the generated visuals should be placed in the video.

Q: What tools are required for the demonstrated video workflow?

The demonstrated workflow uses a Claude account, the Claude desktop application and its Code interface, the Higgsfield CLI and Higgsfield Generate skill, Terminal or PowerShell, and TurboScribe. Claude Code coordinates image generation, Higgsfield creates the visuals, and TurboScribe converts recorded narration into a transcript containing the timestamps needed to organize those visuals.

Q: How do you install the Higgsfield skill for Claude Code?

Open the MCP and CLI section of the Higgsfield website, select CLI, and copy its first installation command into Terminal or PowerShell. Next, run the sign-in command and authorize the connection by selecting Connect. Finally, run the skill command, select Higgsfield Generate with the space key, choose project installation, select the symbolic-link method, and confirm the installation.

Q: Why does the workflow require recorded audio instead of only a script?

Recorded audio is necessary because the automation relies on timestamps that correspond to the actual narration. A normal written script does not contain the required timing information. By recording the narration and transcribing that audio, the creator obtains markers for specific seconds, which Claude Code can use to generate and organize visuals for individual moments.

Q: How do you create a timestamped transcript for video automation?

Render or export the audio from the recorded video, open TurboScribe, and upload the audio through its Transcribe Files option. Start the transcription and wait for processing, which the demonstration says usually takes around 30 seconds to one minute. When the completed status appears, open the transcript and copy the version that visibly includes timestamps.

Q: Why should the TurboScribe transcript be copied instead of downloaded as a PDF?

The transcript should be copied directly from the TurboScribe page because that displayed version contains the timestamps required by the automation prompt. In the demonstrated process, downloading the transcript as a PDF removes those timestamps. Without them, Claude Code cannot associate generated images with exact moments or provide timestamp-based filenames for placement during editing.

Q: What should the Claude Code image-generation prompt include?

The prompt should state that the supplied text is a video script with timestamps and request one Higgsfield image for every timestamp using the GPT Image 2 model. It should require horizontal images in a 16:9 format and direct Claude Code to name every image with the exact corresponding timestamp. The timestamped transcript is then pasted after those instructions.

Q: How do timestamped filenames help with video editing?

Timestamped filenames provide a direct reference between each generated image and the narration segment it supports. If an image is named for zero seconds, three seconds, or another transcript marker, the creator can identify its intended position without manually matching every visual to the script. This naming convention organizes the assets and simplifies placing them into the video timeline.

Summary & Key Takeaways

• The workflow begins by creating a Claude account, downloading the Claude desktop application, and selecting its Code interface. Video footage is generated through the Higgsfield skill, which requires installing the Higgsfield CLI, authorizing the account connection, and adding the Higgsfield Generate skill to the chosen Claude Code project.

• Creators need recorded narration rather than only a written script. Uploading that audio to TurboScribe produces a transcript containing timestamps, such as zero, three, and six seconds. Those timing markers allow Claude Code to associate each generated visual with a specific point in the narration and intended video sequence.

• The main prompt instructs Claude Code to use Higgsfield and the GPT Image 2 model to create one horizontal 16:9 image for every transcript timestamp. Each file receives the exact corresponding timestamp as its name, helping the creator identify where every generated image should be placed during editing.