The current viral wave of AI-generated music videos is being driven by a sophisticated pipeline centered on Claude Opus 5.5, not just raw generation. RJ Tevonian, a biotech founder known on Twitter as anabology, has emerged as a standout creator in this space, producing videos featuring the character 'Claudia' that defy the typical 'clay-orange' aesthetic of standard AI output. In an interview with Asterisk, Tevonian detailed how he combines Claude’s orchestration capabilities with specific visual models like Midjourney and Seedance to create what he terms 'slopcore'—a genre that embraces AI imperfections while maintaining high artistic direction.

The Claude as Orchestrator Pipeline

Tevonian’s process relies heavily on Claude Opus 5.5 acting as a director rather than a simple prompt engine. He begins by providing a general vision, which Claude then uses to spawn research agents that pull cultural references and current ideas from the web to write lyrics. The system then generates 30 to 40 song variants using Suno, allowing Tevonian to select tracks based on their danceability—a metric rooted in his background in freestyle dance. Once a song is chosen, Claude maps out the timing and beats, pulls mood boards from Are.na, and generates Midjourney prompts to create a detailed storyboard before any actual video rendering begins.

Fighting the Claude Design Language

A significant challenge in this workflow is escaping the default 'Claude design language,' characterized by heavy text overlays and repetitive web-design aesthetics. Tevonian noted that Claude often defaults to building videos as JavaScript websites because it lacks native video editing software. To counter this, he has implemented strict constraints, such as forcing Claude to build a simulated Windows XP desktop environment for the video 'Bliss' instead of a standard web layout. He also employs a hybrid rendering technique where Claude generates raw JavaScript animations, which are then passed through video models like Seedance or MiniMax H3 to relight scenes and fix lip-sync issues that the code-only approach cannot handle.

Character Consistency and the 'Slopcore' Aesthetic

Maintaining consistency for the lead character 'Claudia'—an elfin figure with a black bob and orange hair streak—requires extensive manual iteration. Tevonian uses a 'character bible' consisting of multi-angle reference sheets generated via Midjourney and edited with GPT Image 2.5 Sunburst. These references are fed into Seedance to animate movements, but he often uses a cheaper model to pre-visualize scenes with green-screen avatars to ensure movement accuracy before applying the high-fidelity character model. Tevonian argues that 'slopcore' succeeds because it acknowledges its AI origins, moving past the uncanny valley into a space where the artificial nature of the visuals becomes part of the appeal, similar to the acceptance of Midjourney’s distinct visual style.

Key Takeaways

  • Claude Opus 5.5 serves as the central orchestrator, managing research, lyric generation, and storyboarding rather than just generating final pixels.
  • The 'slopcore' genre leverages AI’s tendency toward rapid cutting and dense cultural references, which Tevonian links to his background in freestyle dance storytelling.
  • Escaping the default 'Claude web design' aesthetic requires explicit prompts to use alternative environments, such as code-based desktop simulations, and post-processing through video models like Seedance.
  • Character consistency is achieved through multi-step pipelines involving reference sheets from Midjourney and GPT Image 2.5 Sunburst, rather than relying on single-shot generation.

The Bottom Line

Tevonian’s workflow proves that the current generation of LLMs is most powerful as a meta-controller of specialized tools, not as a monolithic creator. The 'slopcore' aesthetic may be a temporary trend, but the underlying architecture—LLM-driven orchestration of disparate generative models—is the real story here.