Seven Ways Peter Yang Used Sonnet 5.5 to Make Videos

Peter Yang demonstrates seven videos he made while testing Claude Sonnet 5.5: motion graphics, a mascot animation, product promotions, a talking-head edit and two animated music videos. He presents the results as his own experiments, including projects that required multiple rounds of feedback and outside tools.

Yang says he has been impressed by Sonnet 5.5 after a week of testing. His comparison with Opus is subjective: he describes Sonnet as a less expensive alternative and asks viewers to judge the videos for themselves.

For a 20-second motion-graphics reel, he prompts the model to make a creative designer showreel using code. The first result has generic copy, so he asks for text introducing Sonnet. The screen describes a canvas-based video encoded with FFmpeg, with a soundtrack also written in code.

For the second example, Yang supplies a reference video and asks Claude Code to animate the Claude mascot through the history of tools, from the Stone Age to AI. He requests a final AGI scene and more engaging transitions, then shows the resulting animation with music and sound effects.

For a launch film promoting his own AI course, Yang uses the open-source HyperFrames project. He supplies brand assets, fonts and colors and reviews key frames before animation. He asks for illustrations and adjusts the pace when scenes feel too fast.

The launch film is an iterative process, not a one-prompt result. Yang gives voice-dictated feedback, reviews music choices and asks the model to synchronize animation with the selected track. He recommends checking key frames, timing, sound and brand consistency.

A vertical short uses a similar prompt with a product website and reference material. The result includes narration and music; Yang says he requested corrections to dark frames and added a voice-over. The examples promote his own course, so its specific offer and prices should be understood as part of the demonstration at that time.

Yang then asks for overlays and effects on a talking-head recording without using HyperFrames. An initial template looks too plain, so he asks for more creative editing. The resulting zooms, cutouts and animated elements are more aggressive than his own style, and he says they need further iteration.

For an animated music video, Yang links an inspiring example and asks for a more optimistic story about young people living with AI and robots. He says the model produces a playful code-based animation with sound in one pass, though he considers its shapes simpler than professional anime.

For a more anime-like opening, he combines several tools. Claude helps develop lyrics, Yang generates and downloads a song with Suno, then provides it and a fal.ai API key for video generation. He says he added about $15 in credits. The final clip is creative in his view but still falls short of professional animation.

His practical advice is to provide a reference, describe the desired story, supply brand fonts and logos for launch videos, and iterate. He recommends generating music separately when needed and using a video model for imagery that code alone cannot provide. External services and model usage can have separate costs.

Yang concludes that these tools make video useful for more everyday communication, including explaining topics to children. He expects to share prompts, but emphasizes that concrete references and feedback can be more useful than a lengthy fixed template.