Anthropic’s Sonnet 5.5 Generates Marketing Videos via Code

Peter Yang

Code-driven video generation is emerging as the breakthrough application that reverses Anthropic's decline after Opus 5, enabling users to automate motion graphics, voiceovers, and post-production edits.

AI-assisted summary of Sonnet 5.5 is Here! It's Insane at Making Videos (7 Incredible Examples) — original source. Check the original for full context.

Source published: 2026-09-28

Key ideas

Anthropic's Sonnet 5.5 operates as a lower-cost, highly capable counterpart to Opus 5.5, with code-driven video generation emerging as its breakthrough application across motion graphics, launch films, and animations. (0:00)
Pairing Sonnet 5.5 with HeyGen's open-source HyperFrames library enforces strict brand kits—including fonts, hex codes, and vector logos—while generating static keyframe previews to verify storyboard pacing before video rendering. (5:49)
During automated vertical video production, Sonnet 5.5 independently sourced and integrated the Kokoro voice engine to deliver synchronized speech, sound effects, and motion graphics with minimal human intervention. (11:48)
Cinematic animation workflows push beyond vanilla code generation by combining Claude for lyric drafting and timeline orchestration, Suno for full-track audio, and fal.ai diffusion models for visual rendering at approximately $15 in GPU credits. (16:40)

Detailed notes

## Background Published by Peter Yang on YouTube on September 28, 2026, this 22-minute tutorial explores the video generation capabilities of Anthropic's newly released Claude Sonnet 5.5. In the description, Yang outlines how he built seven distinct videos using code and external integrations—ranging from a motion graphics showreel and product launch film to an anime music video—and shares the exact prompts and tools used.

## TL;DR Following a slump with earlier model releases, Anthropic has made a major comeback with Opus 5.5 and the more affordable Sonnet 5.5, with code-driven video generation emerging as their standout capability. Through seven hands-on experiments, Peter Yang demonstrates how Sonnet 5.5 can code motion graphics from scratch, animate characters with custom procedural sound, build brand-compliant product launch videos using HeyGen's open-source HyperFrames library, and even edit talking-head footage with dynamic overlays. For higher cinematic fidelity, Sonnet 5.5 can also orchestrate end-to-end pipelines by writing lyrics for Suno and syncing the audio to generative video diffusion models via fal.ai. Ultimately, getting the best results comes down to ditching rigid templates, granting the model creative freedom, and feeding it direct style references from X.

## Atmosphere & style The video feels like an energetic, fast-paced, and practical workshop led by an enthusiastic builder who is genuinely impressed by what the technology can do. Rather than presenting a polished, theoretical lecture, Yang keeps the mood grounded and conversational—sharing his real-time trial and error, including initial prompt failures, blacked-out frames, and playful tactics like flattering the AI's taste or voice-dictating feedback. The emotional arc moves from candid industry commentary about Anthropic's rocky history into a playful, eye-opening showcase of creative experiments that leaves the viewer feeling inspired to immediately try the workflows themselves.

## Bulletin - [00:00](https://www.youtube.com/watch?v=MLnsMIbibZY&t=0s) Anthropic's Sonnet 5.5 operates as a lower-cost, highly capable counterpart to Opus 5.5, with code-driven video generation emerging as its breakthrough application across motion graphics, launch films, and animations. - [05:49](https://www.youtube.com/watch?v=MLnsMIbibZY&t=349s) Pairing Sonnet 5.5 with HeyGen's open-source HyperFrames library enforces strict brand kits—including fonts, hex codes, and vector logos—while generating static keyframe previews to verify storyboard pacing before video rendering. - [11:48](https://www.youtube.com/watch?v=MLnsMIbibZY&t=708s) During automated vertical video production, Sonnet 5.5 independently sourced and integrated the Kokoro voice engine to deliver synchronized speech, sound effects, and motion graphics with minimal human intervention. - [16:40](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1000s) Cinematic animation workflows push beyond vanilla code generation by combining Claude for lyric drafting and timeline orchestration, Suno for full-track audio, and fal.ai diffusion models for visual rendering at approximately $15 in GPU credits.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=MLnsMIbibZY&t=0s) Sonnet 5.5 solidifies Anthropic's recovery after Opus 5, matching Opus 5.5's capabilities at a reduced price point with specialized strength in programmatic video generation. - [02:22](https://www.youtube.com/watch?v=MLnsMIbibZY&t=142s) Pure code-based generation can render dynamic motion design showreels from scratch with synchronized typography and audio, completely bypassing third-party animation frameworks. - [03:17](https://www.youtube.com/watch?v=MLnsMIbibZY&t=197s) Using Claude Code's `/goal` directive alongside style references from social media allows the model to produce multi-scene narrative mascot animations with custom sound effects autonomously. - [05:49](https://www.youtube.com/watch?v=MLnsMIbibZY&t=349s) HeyGen's open-source HyperFrames tool enables programmatic brand governance and storyboard keyframe inspection prior to compiling full video renders. - [05:49](https://www.youtube.com/watch?v=MLnsMIbibZY&t=349s) Voice dictating feedback in bulk and praising the model's creative sensibilities consistently pushes Claude Code to deliver higher-effort visual refinements. - [05:49](https://www.youtube.com/watch?v=MLnsMIbibZY&t=349s) Sonnet 5.5 can independently search for, retrieve, and offer multiple royalty-free audio tracks, subsequently synchronizing visual scene cuts to the tempo of the selected file. - [11:48](https://www.youtube.com/watch?v=MLnsMIbibZY&t=708s) Sonnet 5.5 autonomously integrated the Kokoro voice engine to pair mobile-first 9:16 vertical shorts with synchronized voiceovers, sound design, and kinetic typography. - [12:10](https://www.youtube.com/watch?v=MLnsMIbibZY&t=730s) Imposing rigid templates restricts modern reasoning models, whereas granting open creative freedom unlocks automated editing capabilities like background removal, dynamic zooms, and B-roll overlays. - [14:23](https://www.youtube.com/watch?v=MLnsMIbibZY&t=863s) Creative visual storytelling can be achieved through zero-shot prompts referencing social media posts, with Sonnet synthesizing both procedural visual art and original musical accompaniment. - [16:40](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1000s) High-fidelity cinematic anime requires a modular pipeline where Claude orchestrates timing, Suno produces music, and fal.ai diffusion APIs render the video clips for around $15 in cloud credits. - [20:05](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1205s) Linking high-quality reference videos directly from platforms like X serves as a practical prompting shortcut to steer visual style and motion pacing. - [20:05](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1205s) High token efficiency ensures that complex multi-step video creation workflows remain fully accessible on standard $20 subscription tiers rather than requiring high-tier enterprise plans. - [20:05](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1205s) Automated code-driven video tools lower production barriers beyond marketing, opening practical possibilities for on-demand educational content like custom animated science and math lessons.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=MLnsMIbibZY&t=0s) Sonnet 5.5 and the Anthropic Comeback

Anthropic is on what feels like an aggressive comeback tour. After a period where users had grown disillusioned with the Claude lineup—to the point of switching over to ChatGPT because Claude's writing style and personality degraded from Opus 4.7 onwards, culminating in what Anthropic internally admitted was their worst release yet with Opus 5—the release of Opus 5.5 dramatically reversed fortunes by delivering raw speed, usefulness, and capability. Following closely behind, Sonnet 5.5 continues that winning streak as an essentially cheaper, equally capable version of Opus 5.5.

```mermaid flowchart LR accTitle: Claude Model Trajectory accDescr: Progression showing Claude's dip with Opus 4.7 and Opus 5 leading to a resurgence with Opus 5.5 and Sonnet 5.5. subgraph Decline["Decline Phase"] A["Opus 4.7 Degraded personality"] --> B["Opus 5 Anthropic admits worst release"] end subgraph Comeback["Comeback Phase"] C["Opus 5.5 Fast, useful, high capability"] --> D["Sonnet 5.5 Cheaper, code-driven video powerhouse"] end B -.->|"Turnaround"| C ```

Just as 3D model generation turned out to be the magical killer use case for GPT Astra, code-generated video is emerging as the breakthrough application for both Opus and Sonnet 5.5. Across a full week of testing, Sonnet 5.5 demonstrated that it can generate production-grade video directly via code execution—powering everything from social media memes and dynamic motion design showreels to illustrated mascots, product launch films, and anime music videos.

[02:22](https://www.youtube.com/watch?v=MLnsMIbibZY&t=142s) Video 1: Motion Graphics Reel

The first test was designing a high-energy motion design reel built entirely from scratch with code, bypassing external frameworks like HyperFrames. The prompt framed Sonnet as an elite designer building an audition piece:

> "Make a dynamic 20-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. Go all out and be extremely creative. Don't use HyperFrames, just make it from scratch with code."

The initial output generated polished animations but relied on somewhat generic placeholder copy. Rather than restarting, a quick revision instructed Sonnet to replace the placeholder text with self-referential promotional messaging: *"Guess what, I can make videos now. Introducing Sonnet 5.5."* The final result rendered seamlessly with synchronized music and typographic transitions, proving that simple adjustments and website links can instantly transform code-based animations into customized sizzle reels for any product.

[03:17](https://www.youtube.com/watch?v=MLnsMIbibZY&t=197s) Video 2: Claude Mascot Animation

The second experiment took inspiration from an animated clip posted by Drew on X (formerly Twitter), which featured the Claude avatar surrounded by lightning, energy effects, and intense zooms into its eye. Using Claude Code, the model was assigned a explicit `/goal` directive to reproduce that visual caliber while expanding the scope into an evolutionary narrative:

- **Style Reference**: Linked directly to the tweet demonstrating high-end anime particle effects and dynamic camera work. - **Narrative Scope**: A 30- to 40-second sequence tracking the Claude mascot through human tool evolution, starting in the Stone Age with rudimentary rocks, advancing to the invention of the wheel, continuing to modern AI, and culminating in an added final scene depicting AGI. - **Audio & Sound**: Sonnet authored the code that generated all visual layers alongside customized sound effects and accompanying background music entirely autonomously, without third-party tool pipelines. - **Iterative Feedback**: Applying modern prompting tactics—specifically asking the model to *"make all the scenes more engaging and better"*—pushed the output to generate stylized anime action sequences that match professional video references.

[05:49](https://www.youtube.com/watch?v=MLnsMIbibZY&t=349s) Video 3: Product Launch Video with HyperFrames

To build a formal product announcement film for an AI curriculum hosted at `behindthecraft.com`, Sonnet was paired with HeyGen's open-source **HyperFrames** repository. Installing the library was straightforward: copying the GitHub repo URL directly into Claude Code and directing the model to handle the setup.

```mermaid flowchart TD accTitle: HyperFrames Workflow in Claude Code accDescr: Step-by-step pipeline from HyperFrames installation through keyframe review, iteration, and final render. subgraph Setup["1. Initialization"] A["Clone HyperFrames repo into Claude Code"] --> B["Inject Brand Kit Crimson Pro font, red circle, logos"] end subgraph Review["2. Keyframe & Audio Iteration"] B --> C["Review static keyframes before rendering"] C --> D["Voice-dictated feedback Pacing, timing, tasteful illustrations"] D --> E["Model sources 3 audio options Select favorite track"] end subgraph Delivery["3. Final Output"] E --> F["Compile code & render Beat-synced animation"] end ```

HyperFrames provides two distinct advantages over blind code generation: it strictly adheres to provided branding guidelines (such as Crimson Pro typography, specific brand color hex codes, and vector logos like a signature red circle), and it renders an array of static keyframes upfront. This preview allows creators to evaluate the visual storyboard and critique pacing before committing computing cycles to full video rendering.

The creation process relied heavily on conversational collaboration rather than one-shot generation: - **Research & Storyboarding**: Sonnet researched video structures generated by Opus 5.5, proposing complementary vector illustrations to elevate the aesthetic quality. - **Voice Dictation Feedback**: Feedback was dictated in bulk using Claude Code's voice dictation feature, adjusting scene lengths where transitions felt rushed. - **Model Flattery**: As a practical working tactic, praising the LLM's taste—telling it that it possesses great creativity—consistently encouraged higher-effort programmatic refinements. - **Autonomous Music Selection**: When prompted for background music, Sonnet autonomously retrieved three separate royalty-free tracks, allowing the user to audition the choices before programmatically syncing scene cuts to the rhythm of the chosen track.

[09:54](https://www.youtube.com/watch?v=MLnsMIbibZY&t=594s) Video 4: Vertical Short with Voiceover

The final demonstration in this workflow adapted the product promotional campaign for social platforms using a vertical 9:16 aspect ratio. The prompt directed Claude Code to turn a promotional tweet into an asset-ready vertical reel:

> "Make a punchy and modern vertical short video for behindthecraft.com that promotes the subscription and the course. Needs to be production grade with sound. Here's a relevant tweet: [link]. Make it incredible and be creative."

The output paired dynamic, kinetically timed on-screen typography and background audio with a synthesized voiceover script targeted directly at audience pain points:

*“Every week, a new AI tool, a new model, a new workflow, and you're supposedly behind. Stop collecting AI tools. Build one system. An AI chief of staff to run your inbox and calendar. An AI advisor for the big decisions. Reusable skills that run themselves. Your morning brief is ready before you ask. Over 25 lessons, over 40 prompts, 16 ready to install skills, $600 in credits, monthly live workshops, and 400 builders. Join by October 7th to lock in $150 a year. After that, it's 200. 30-day money-back guarantee. Copy the AI system Peter uses to run his work and life. behindthecraft.com.”*

While the initial run produced a solid, market-ready ad, it still required targeted refinements—specifically addressing a few blacked-out video frames occurring roughly 11 seconds into the timeline.

[11:48](https://www.youtube.com/watch?v=MLnsMIbibZY&t=708s) Wrapping Up Vertical Shorts and Audio Integration

Wrapping up the fourth showcase—the vertical promotional short—the model demonstrated an unexpected ability to handle full audio integration without requiring manual sound design. After being prompted to supply a British voiceover, Sonnet 5.5 located and implemented a voice engine identified as Kokoro on its own. With minimal human direction and a single feedback cycle, the model produced a production-ready vertical asset that included synchronized speech, sound effects, and tailored motion graphics. This confirmed that Sonnet 5.5 can handle both standard horizontal promotional films and mobile-first vertical shorts right out of the box.

[12:10](https://www.youtube.com/watch?v=MLnsMIbibZY&t=730s) Video 5: Talking Head Overlays and Dynamic Editing

The fifth demonstration tackled a common production headache: enhancing a standard talking-head video for a landing page. The goal was to take raw footage of a presenter speaking, keep the face on camera, and generate animated overlays, kinetic typography, dynamic zooms, and custom B-roll entirely from scratch using code rather than third-party frameworks like HyperFrames. When Sonnet 5.5 was first instructed to follow a structured template, the initial keyframe results were underwhelming and pedestrian—featuring generic captions, basic logos, and sterile layout choices.

Recognizing that strict templates stifle the latest generation of reasoning models, the approach was deliberately flipped. Sonnet was instructed: *"This template is super boring. Forget about it. Use your creativity and make it as interesting as possible like from a professional editing studio."* Unshackled from rigid guidelines, Sonnet produced an aggressive, high-energy cut reminiscent of a fast-paced YouTube production. The model successfully keyed out the presenter's background, introduced punchy zoom cuts, dropped in expressive character assets, and layered frantic motion graphics across the screen. While the resulting aesthetic leaned slightly too far into over-the-top "MrBeast style" for a professional corporate video, it proved how effectively AI can automate tedious post-production tasks.

- **Initial prompt mistake:** Restricting the model to rigid templates yields conservative, dull keyframes. - **Unlocking model capability:** Granting creative freedom allows Sonnet 5.5 and Opus 5.5 to behave like an autonomous post-production suite. - **Automated editing capabilities:** Features background removal/rotoscoping, dynamic scale/zoom adjustments, kinetic text overlays, and contextual graphics. - **Cost implications:** While an elite human editor still provides superior nuanced taste, iterating with Sonnet offers a viable, low-cost alternative for solo creators.

[14:23](https://www.youtube.com/watch?v=MLnsMIbibZY&t=863s) Video 6: Animated Music Video

Turning away from commercial marketing formats, the sixth experiment explored creative storytelling through an animated music video. Drawing direct inspiration from a viral X post featuring a frantic, techno-dystopian track about impending AGI doom, the prompt sought to invert that narrative. Rather than leaning into apocalyptic anxieties, Claude was directed toward an optimistic, energetic vision exploring what a young person's daily life might look like living alongside benevolent robots and advanced AI systems.

The generation was nearly a pure "one-shot" output. Pointed at an extensive reference prompt from the original tweet, Sonnet 5.5 synthesized code that generated whimsical geometric character art, fluid kinetic transitions, and its own synchronized procedural soundtrack. While the aesthetic relied on simplified shapes rather than hyper-detailed cel animation, the final piece delivered a surprisingly cohesive rhythm, emotional charm, and upbeat tempo that captured the intended vision without requiring iterative corrections.

- **Inspiration:** An apocalyptic AGI music video shared on X, adapted into an optimistic retro-futuristic narrative. - **Execution:** Zero-shot generation driven by a linked tweet reference and a prompt for hopeful technological themes. - **Multimodal output:** Procedural code generated both the visual animation sequences and the accompanying musical score.

[16:40](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1000s) Video 7: Japanese Anime Opening Sequence

To push beyond lightweight code animations toward a genuine anime opening sequence, Sonnet 5.5 was asked to outline an end-to-end production pipeline. The model explained that high-fidelity cinematic animation cannot yet be coded purely from scratch in vanilla browser scripts; it requires external generative models for custom music and dedicated video diffusion. Following Sonnet's technical roadmap, the workflow combined custom lyric writing in Claude, AI track generation in Suno, and neural video rendering hosted on fal.ai.

```mermaid flowchart LR accTitle: Generative Anime Video Workflow accDescr: Pipeline showing lyric generation in Claude, music rendering in Suno, and video generation via fal.ai orchestrated by Sonnet.

subgraph Audio ["1. Music Production"] A["Claude Sonnet 5.5 Lyric Generation"] --> B["Suno Advanced Mode Full Track Rendering"] end

subgraph Video ["2. Video Generation"] C["fal.ai Platform API Credits (~$15)"] --> D["Video Diffusion Model Cinematic Generation"] end

subgraph Assembly ["3. Final Composition"] B --> E["Claude Sonnet 5.5 Orchestration Audio-Visual Synchronization"] D --> E end ```

The process began with iterative songwriting inside Claude to draft high-energy anime theme lyrics. These lyrics and stylistic tags were pasted into the advanced mode of Suno, which generated multiple polished song options for download. Next, an account was set up on fal.ai with approximately $15 in credits to access advanced video diffusion models. Sonnet 5.5 took the resulting audio file, hooked into the fal.ai API, and orchestrated the scene cadence, cutting dramatic visual sequences against lyrical peaks like *"Stars are falling, catch the light / Shatter every after it tonight."* While missing the fine-grained visual taste of master human illustrators, the resulting music video delivered a compelling proof of concept for automated cinematic production.

- **Audio generation platform:** Suno (advanced mode for custom genre tags and multi-track generation). - **Video infrastructure:** fal.ai running dedicated generative video diffusion APIs. - **Approximate tool cost:** ~$15 in cloud GPU credits to generate the raw visual clips. - **Orchestration role:** Sonnet 5.5 writes the lyrics, structures the timing, and writes the code that binds video clips to musical markers.

[20:05](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1205s) Four Practical Tips for AI Video Generation

Reflecting across all seven video experiments, four core best practices emerged for anyone looking to build complex video assets using Sonnet 5.5 and Opus 5.5. Furthermore, while these tests utilized the $200 Claude Max plan, the models demonstrated exceptional token efficiency, meaning users on standard $20 subscription tiers can comfortably recreate multiple custom video workflows without immediate context exhaustion.

1. **Leverage reference posts from X:** When encountering an appealing visual format or dynamic animation on social media, paste the direct link to the tweet into the prompt. Instruct the model to replicate the visual style while inserting your own subject matter. 2. **Standardize branding with HyperFrames:** For commercial and corporate launch videos, install the free HyperFrames tool. Supply explicit hex color codes, typography files, and brand asset URLs to keep the output faithful to brand guidelines. 3. **Outsource audio to dedicated tools like Suno:** Although Claude can generate primitive audio waveforms purely from programmatic code, higher production values require creating dedicated tracks in Suno, downloading the audio, and prompting Claude to synchronize the cuts to musical beat drops. 4. **Tap into API-driven video models for cinematic fidelity:** Code-only animations are best suited for motion graphics and UI demos; genuine cinematic fidelity requires connecting Claude to models hosted on providers like fal.ai.

Beyond promotional films and creative storytelling, the accessibility of automated video tools points toward practical everyday communication. Democratizing video creation makes it possible to rapidly build interactive visual lessons—such as generating custom animated science or math tutorials for children—transforming passive media consumption into an accessible creative exercise.

[22:20](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1340s) Final Prompting Advice and Sign-Off

Peter wraps up his video-making tips with a straightforward shortcut for getting great results: rather than overcomplicating the instructions, you can "just link an incredible video, even one that you see in this tutorial, and tell it to make something similar." He closes on a warm, familiar note, inviting viewers who enjoyed this practical tutorial and want to see more like it to like and subscribe before signing off until next time.

## Notable quotes > [02:22](https://www.youtube.com/watch?v=MLnsMIbibZY&t=142s) Make a dynamic 20-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. Go all out and be extremely creative. Don't use HyperFrames, just make it from scratch with code. — Peter Yang

> [02:22](https://www.youtube.com/watch?v=MLnsMIbibZY&t=142s) Guess what, I can make videos now. Introducing Sonnet 5.5. — Peter Yang

> [03:17](https://www.youtube.com/watch?v=MLnsMIbibZY&t=197s) make all the scenes more engaging and better — Peter Yang

> [09:54](https://www.youtube.com/watch?v=MLnsMIbibZY&t=594s) Make a punchy and modern vertical short video for behindthecraft.com that promotes the subscription and the course. Needs to be production grade with sound. Here's a relevant tweet: [link]. Make it incredible and be creative. — Peter Yang

> [09:54](https://www.youtube.com/watch?v=MLnsMIbibZY&t=594s) Every week, a new AI tool, a new model, a new workflow, and you're supposedly behind. Stop collecting AI tools. Build one system. An AI chief of staff to run your inbox and calendar. An AI advisor for the big decisions. Reusable skills that run themselves. Your morning brief is ready before you ask.

> [09:54](https://www.youtube.com/watch?v=MLnsMIbibZY&t=594s) Over 25 lessons, over 40 prompts, 16 ready to install skills, $600 in credits, monthly live workshops, and 400 builders. Join by October 7th to lock in $150 a year. After that, it's 200. 30-day money-back guarantee. Copy the AI system Peter uses to run his work and life. behindthecraft.com.

> [12:10](https://www.youtube.com/watch?v=MLnsMIbibZY&t=730s) This template is super boring. Forget about it. Use your creativity and make it as interesting as possible like from a professional editing studio. — Peter Yang

> [16:40](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1000s) Stars are falling, catch the light / Shatter every after it tonight.

> [22:20](https://www.youtube.com/watch?v=MLnsMIbibZY&t=1340s) just link an incredible video, even one that you see in this tutorial, and tell it to make something similar. — Peter Yang

## Mentioned Peter Yang, Anthropic, Claude, Sonnet 5.5, Opus 5.5, Opus 5, Opus 4.7, ChatGPT, GPT Astra, Claude Code, Claude Max, HeyGen, HyperFrames, Suno, fal.ai, Kokoro, behindthecraft.com, creatoreconomy.so, Crimson Pro, X (Twitter), YouTube, GitHub, LinkedIn, Drew, MrBeast

## Video description Copy my best AI workflows to save time and automate busywork: https://www.behindthecraft.com 🤝 Subscribe to my practical AI newsletter read by 150K+ professionals: https://creatoreconomy.so

Sonnet 5.5 is finally here and it's incredible at making videos. In this tutorial, I'll show you how I made 7 videos with Sonnet 5.5, from a reel drawn in code to a product launch film and an anime music video, and more.

What I cover: (00:00) This video was made with code by Sonnet 5.5 (01:16) Why Claude is back (02:22) Video 1: Motion graphics reel (03:17) Video 2: Claude mascot animation (05:49) Video 3: Product launch video (09:54) Video 4: Vertical short with voiceover (12:10) Video 5: Talking head overlays (14:23) Video 6: Animated music video (16:40) Video 7: Anime opening (20:05) 4 tips for making videos

Prompts I used: Motion graphics reel: "Make a dynamic 20-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. Go all out and be extremely creative. Don't use HyperFrames, just make it from scratch with code."

Vertical short: "Make a punchy and modern vertical short video for [your website] that promotes [your product]. Needs to be production grade with sound. Here's a relevant tweet: [link]. Make it incredible and be creative."

Links: HyperFrames (free): https://github.com/heygen-com/hyperframes Suno: https://suno.com fal: https://fal.ai

Where to find me: X: https://x.com/petergyang LinkedIn: https://www.linkedin.com/in/petergyang/

Subscribe to this channel - more tutorials coming soon!

[Source](https://www.youtube.com/watch?v=MLnsMIbibZY)

中文简报

Anthropic的Sonnet 5.5通过代码生成营销视频代码驱动的视频生成正成为扭转Anthropic在Opus 5之后颓势的突破性应用，使用户能够实现动态图形、配音和后期剪辑的自动化。在经历此前模型发布的低谷后，Anthropic凭借Opus 5.5以及成本更低的Sonnet 5.5实现强势回归，而代码驱动的视频生成成为其最亮眼的能力。通过七项实操实验，Peter Yang展示了Sonnet 5.5如何从零编写动态图形代码、利用自定义程序化音效制作角色动画、使用HeyGen的开源HyperFrames库构建符合品牌规范的产品发布视频，甚至为口播视频添加动态叠加层剪辑。为了获得更高的电影级保真度，Sonnet 5.5还可以通过为Suno编写歌词，并借助fal.ai将音频与生成式视频扩散模型进行同步，从而编排端到端的工作流。最终，获得最佳效果的关键在于摒弃僵化的模板、赋予模型创作自由，并直接向其提供来自X的风格参考。 Anthropic的Sonnet 5.5作为Opus 5.5低成本且高能力的分支，其代码驱动的视频生成正成为横跨动态图形、发布短片和动画领域的突破性应用。 (0:00) 将Sonnet 5.5与HeyGen的开源HyperFrames库搭配使用，可严格执行包括字体、十六进制色号和矢量Logo在内的品牌规范包，同时生成静态关键帧预览以便在视频渲染前验证分镜节奏。 (5:49) 在自动化竖屏视频制作中，Sonnet 5.5自主寻找并集成了Kokoro语音引擎，在极少人工干预的情况下实现了同步语音、音效与动态图形的输出。 (11:48) 电影级动画工作流超越了纯代码生成，它结合了用于歌词起草和时间轴编排的Claude、用于全曲音频的Suno，以及用于视觉渲染的fal.ai扩散模型，GPU算力花费约为15美元。 (16:40)

Reading summary

Following internal dissatisfaction with earlier Opus releases, Anthropic's Sonnet 5.5 demonstrates that code-executed video generation has emerged as a practical powerhouse for multimedia workflows, according to Peter Yang. Direct Code Generation and Creative Latitude (2:22) Sonnet 5.5 can generate dynamic motion design reels, character animations, and post-production video edits directly through code execution without third-party frameworks. In early tests, prompt directives that instructed the model to act as an auditioning designer yielded synchronized typography and animations, while mascot animation tests generated multi-scene historical narratives complete with procedural sound effects and background music. Yang emphasizes that constraining the model with strict formatting templates degrades the output into sterile layouts. When tasked with enhancing raw talking-head footage, restricting Sonnet to structured templates produced pedestrian results. Removing those boundaries allowed the model to act as an automated editing suite—applying background removal, kinetic zoom cuts, dynamic text overlays, and expressive graphic assets reminiscent of fast-paced YouTube production. Brand Consistency and Multi-Format Social Campaigns (5:49) For commercial applications requiring strict visual standards, Sonnet 5.5 pairs effectively with HeyGen's open-source HyperFrames repository via Claude Code. This setup enforces brand identity rules—such as Crimson Pro typography, precise hex color codes, and vector logos—while generating static keyframes upfront so creators can evaluate visual storyboards and pacing before rendering full video files. The workflow proved capable of translating written marketing collateral into multi-format campaigns, including a 9:16 vertical short. By autonomously integrating an external speech engine identified as Kokoro, Sonnet supplied a British voiceover, synchronized sound effects, and kinetic typography with minimal human intervention. Iteration was streamlined through Claude Code's voice dictation feature, where Yang noted that offering positive conversational reinforcement consistently encouraged more creative programmatic adjustments. Hybrid Pipelines for Cinematic Video Diffusion (16:40) Because pure in-browser code cannot produce photorealistic or complex anime sequences from scratch, Sonnet 5.5 can act as an orchestrator across specialized generative services. In an anime opening sequence test, Sonnet structured an end-to-end pipeline that combined custom lyric writing in Claude with track generation in Suno's advanced mode. To handle visual rendering, the model connected to generative video diffusion models hosted on fal.ai, which required approximately $15 in cloud GPU credits. Sonnet then wrote the integration code that programmatically aligned the generated visual scenes to the rhythmic peaks of the downloaded audio track. While the output lacked the refined taste of human illustrators, it demonstrated that an LLM can direct multi-platform production pipelines. Practical Execution Tactics and Resource Efficiency (20:05) Synthesizing the experiments, Yang outlines four core best practices: linking directly to reference posts on X to replicate visual structures, installing HyperFrames to maintain brand compliance, offloading soundtrack production to dedicated audio tools like Suno, and connecting to API-driven platforms like fal.ai when cinematic fidelity is required. From an operational standpoint, these multi-step workflows do not necessarily require enterprise tier access. Although the demonstrations were executed on a $200 Claude Max plan, Sonnet 5.5 exhibited high token efficiency, meaning users on standard $20 monthly subscription tiers can run several iterative video generation projects without immediately exhausting context limits. While Sonnet 5.5 automates tedious production tasks from rotoscoping to beat-synced editing, Yang observes that AI still falls short of elite human editors and illustrators in nuanced taste. Nevertheless, its low cost and rapid generation turn video production into an accessible medium for solo creators, social marketers, and educators.

Source quotes

[2:22] Peter Yang: Make a dynamic 20-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. Go all out and be extremely creative. Don't use HyperFrames, just make it from scratch with code. [2:22] Peter Yang: Guess what, I can make videos now. Introducing Sonnet 5.5. [3:17] Peter Yang: make all the scenes more engaging and better [9:54] Peter Yang: Make a punchy and modern vertical short video for behindthecraft.com that promotes the subscription and the course. Needs to be production grade with sound. Here's a relevant tweet: [link]. Make it incredible and be creative. [9:54] Every week, a new AI tool, a new model, a new workflow, and you're supposedly behind. Stop collecting AI tools. Build one system. An AI chief of staff to run your inbox and calendar. An AI advisor for the big decisions. Reusable skills that run themselves. Your morning brief is ready before you ask. [9:54] Over 25 lessons, over 40 prompts, 16 ready to install skills, $600 in credits, monthly live workshops, and 400 builders. Join by October 7th to lock in $150 a year. After that, it's 200. 30-day money-back guarantee. Copy the AI system Peter uses to run his work and life. behindthecraft.com. [12:10] Peter Yang: This template is super boring. Forget about it. Use your creativity and make it as interesting as possible like from a professional editing studio. [16:40] Stars are falling, catch the light / Shatter every after it tonight. [22:20] Peter Yang: just link an incredible video, even one that you see in this tutorial, and tell it to make something similar.

Original transcript

[0:00] The video that I'm about to show you [0:01] next was created entirely using code by [0:05] the new Sonnet 5.5. Let's play it. [0:17] >> [music] [0:23] [music] [0:30] [music] [0:35] >> All right, we're back. You know, [0:36] Anthropic really is on a comeback tour [0:39] right now. Opus 5.5 blew me away by just [0:42] how useful, fast, and capable it is. And [0:46] Sonnet 5.5, based on my testing for a [0:49] week, is going to continue that win [0:51] streak. [0:52] In this tutorial, I'm going to walk [0:53] through how I made seven incredible [0:55] videos with the new Sonnet, from memes [0:58] to motion graphics, to animated [1:00] illustrations, to product launch videos, [1:03] and even anime music videos. [1:06] Now, most of these videos were generated [1:08] with code, but there are also some free [1:10] open-source libraries and tools that can [1:13] take your video to the next level that [1:15] I'll show you. [1:16] Let's get started with this meme. As I [1:18] mentioned in my Opus video, I had kind [1:21] of given up on Claude and switched over [1:23] to ChatGPT. From Opus 4.7 onwards, [1:28] Claude's writing and personality just [1:29] got increasingly worse, and even [1:31] Anthropic admits that Opus 5 was the [1:34] worst model released yet. [1:37] But I and many others have been [1:38] completely blown away by Opus 5.5. [1:42] And now that we have Sonnet 5.5, [1:45] it's essentially a cheaper version of [1:48] Opus 5.5, and still really incredible. [1:53] It's basically the unicorn horn that [1:55] makes this entire meme horse come to [1:58] life. And by the way, I also made this [2:00] meme using Sonnet's video capabilities. [2:03] And just like how GPT Astra's magic use [2:06] case was 3D models, [2:08] Opus and Sonnet's magic use case is [2:11] video. Now, of course, the biggest [2:13] question here is that can Sonnet make [2:16] videos that are just as good as Opus? [2:19] Let's get into our demos and judge for [2:21] ourselves. [2:23] Okay, let's go back to the motion [2:26] graphics reel that I demoed at the [2:27] beginning of this video. Here's how I [2:29] made it. It's a very simple prompt. Make [2:31] a dynamic 20-second motion graphics [2:32] video that shows what an incredible [2:34] designer you are. Like it should show [2:36] reel for a resume. Go all out and be [2:39] extremely creative, right? You can [2:41] easily adjust the prompt. For example, [2:42] you can include this prompt and also [2:44] include a link to the website or [2:46] whatever product you're trying to [2:46] promote so that it can make a really [2:48] awesome sizzle reel for that. [2:50] So, the first version of what I made is [2:52] something like this. [2:54] Pretty amazing. But, it had pretty [2:56] generic copy, right? And I wanted to use [2:57] this as a trailer for Claude itself. So, [3:00] I basically told it to change the copy [3:02] to guess what, I can make videos now. [3:05] Introducing Sonnet 5.5. And the result [3:07] is basically this incredible video that [3:10] you saw at the beginning of this [3:11] tutorial. [3:13] So, that's the motion graphics and [3:14] already I'm kind of blown away. [3:17] Now, let's move on to our next use case, [3:19] which is to make an animated video of [3:21] the Claude avatar. So, I came across [3:24] this amazing video from Drew. [3:26] Uh you can see here. [3:29] Right? It's got the Claude avatar and [3:30] it's got some crazy effects. It zooms [3:31] into the eye. It shows, I don't know, [3:33] thunder and light effects. And it it [3:36] just it it kind of blew me away, right? [3:38] So, I wanted to make a similar video [3:41] that shows the Claude avatar [3:43] basically use tools from the Stone Age [3:47] all the way to AGI. [3:48] All right? [3:49] So, you know, again, I went to Claude [3:52] code. And this time I set a goal, slash [3:54] goal. I wanted to make a 30 to 40-second [3:57] video exactly like this style, and I [3:59] linked the tweet. And the idea is the [4:01] Claude mascot walking through the [4:03] evolution of human tools from the Stone [4:05] Age all the way to us using AI right [4:07] now. Make it as creative as possible [4:10] until the video quality level matches [4:12] the video I linked above. [4:14] And then I kind of worked through [4:15] everything. [4:16] And then I gave it some feedback. I [4:18] said, "Hey, can you actually add a scene [4:20] at the end for AGI and make all the [4:23] scenes more engaging and better?" You [4:25] know, this is kind of sophisticated [4:27] prompting that you have to do now with [4:28] these latest models. Just make it make [4:30] it better better. And then it came up [4:32] with this video. So, I'm going to go [4:34] ahead and play this video now for you [4:35] guys to watch. [4:45] >> [music] [4:56] [music] [5:01] [music] [5:05] [music] [5:16] >> All right. So, as you just saw, it made [5:18] this awesome anime clip from using [5:21] stones to the wheel to everything else, [5:24] eventually ending up with AGI. And [5:26] again, it created all the sound effects [5:27] and music by itself, which is pretty [5:29] mind-blowing. It did not use any tools [5:31] as far as I can tell. It just generated [5:33] all the stuff using code. [5:35] All right. So, this is pretty [5:36] mind-blowing. You can easily make a [5:37] video like this. You can put the Claude [5:38] mascot [5:40] in any kind of scene you want with just [5:42] simple prompting and linking to the [5:44] right video asset that you want it to be [5:46] inspired by. [5:48] Okay, next I want to see if Sonic could [5:51] generate a product launch video for my [5:53] new AI course at behindthecraft.com. [5:57] And in this case, I used the Hyperframes [5:59] repo that you can find on GitHub and [6:02] also link in the description. [6:03] All you have to do to install it just [6:05] like copy this link and paste it into [6:06] Clock code and tell it to install it. [6:09] Now, I like using Hyperframes for a [6:10] couple of reasons. Number one, it lets [6:12] you easily customize the video to match [6:14] your style guide like your fonts and [6:16] colors. [6:17] And also, it shows you a bunch of key [6:18] frames before making a video so you can [6:20] give it feedback along the way. [6:23] All right, so let me first play the [6:25] product launch video that I made with [6:27] Hyperframes and then I'm going to break [6:28] down how I made it. Let's play the video [6:30] now. [6:35] >> [music] [6:53] [music] [7:09] [music] [7:21] >> Awesome, so that was the video. Now, [7:22] let's take a look at how I made this [7:24] video in Clock code. So, as I mentioned, [7:26] if you use Hyperframes, it's going to [7:28] give you a couple of key frames [7:31] like this of the video that it wants to [7:33] make so they can give it feedback. [7:35] Right? And this looks pretty good and [7:36] like, you know, my brand is basically [7:37] the red circle and, you know, this kind [7:39] of font. And it's using this stuff [7:41] because I told it to. I told it to use [7:42] Crimson Pole front. I told it to use the [7:45] recycle and gave you some example images [7:47] and logos on my brand. [7:49] All right. So, it made the same. [7:52] And then I also do some research on like [7:54] really amazing videos made with Opus 5.5 [7:57] to get inspiration. [7:59] And it said, you know, okay, I did some [8:01] research and based on this, I'm going to [8:03] folding some illustrations and some [8:05] other stuff here. [8:07] And I told it to animate it and show me [8:09] what it looks like. [8:10] Okay, as it's working through it. [8:12] And then what I told it was some voice [8:15] dictated feedback. Okay. So, you know, I [8:18] like to use voice dictation. You can use [8:19] it here in Cloud Code. And just give it [8:21] a ton of feedback about what I want to [8:23] make it better. [8:25] And I say, hey, let's also add some [8:27] music. [8:29] And then do some more iterations. And [8:31] here's what it's come up with. So, now I [8:32] added these little illustrations that [8:35] make it a little bit more tasteful. [8:37] Then, let's see, I give it even more [8:39] feedback, right? I told it to give a few [8:41] extra seconds because some of the scenes [8:43] were too fast. [8:44] And then I said, hey, happy to discuss [8:46] all of the above if you want. You want [8:48] to flatter these models. You want to [8:49] tell them that they have good taste and [8:52] creativity [8:53] to make them work even harder. I'm not [8:54] sure if this actually works, by the way, [8:55] but I find it useful to flatter the [8:57] models. [8:58] All right. So, then it started, you [9:00] know, making the stuff and looks pretty [9:01] good. And then I told it to render it [9:02] and play it. And then I told it to add [9:05] music. It went off by itself and it [9:07] found three different music choices. I [9:09] have no idea where it found this stuff. [9:11] So, I play each music and I told which [9:13] one to add. [9:15] And then eventually it made the video. [9:16] So, that's pretty much it, right? So, [9:19] for this one, it was not one shot. I [9:21] went through a whole process, right? [9:22] Again, install Hyper Frames. [9:24] Tell it to make the key frames for the [9:25] video. Tell it to follow your brand [9:27] style guidelines. Give it some examples. [9:30] And then iterate with it. Give it [9:31] feedback. Tell it which scenes are too [9:33] fast. And tell it to add sound and [9:35] music. And also show you a few different [9:37] options, so that you can pick. And by [9:39] the way, it also basically actually [9:41] syncs its animations to the music, so [9:43] that it's extra engaging. You know, like [9:45] this would take a human editor just like [9:47] a lot of work to do, but AI is somehow [9:49] to write code to do all of this. [9:51] All right, so that is the product launch [9:53] video. And next, I wanted to see if [9:57] Sonic can make a vertical short in very [10:00] much the same format. [10:02] And the prompt that I used to make the [10:03] vertical short is "Make a punchy and [10:06] modern vertical short video for [10:08] behindthecraft.com." [10:10] And you can replace that link with any [10:11] website you want to do. [10:13] That promotes the subscription and the [10:14] course or whatever product you want to [10:16] promote, right? Make a production grade [10:18] with sound. And here is the relevant [10:20] tweet, and the tweet is basically just [10:21] like me walking through what's on the [10:24] course and on the website. [10:26] Okay, so let's go back to Claude. [10:29] And basically it's going through this. [10:32] And voila, it made the vertical short. [10:35] So, let's quickly play the video now to [10:36] see what it looks like. [10:38] >> Every week, a new AI tool, a new model, [10:42] a new workflow, and you're supposedly [10:44] behind. [10:46] Stop collecting AI tools. Build one [10:50] system. [10:51] An AI chief of staff [10:53] >> [music] [10:53] >> to run your inbox and calendar. [10:56] An AI advisor for the big [music] [10:59] decisions. [11:01] Reusable skills that run themselves. [11:03] [music] [11:04] Your morning brief is ready before you [11:06] ask. [11:08] Over 25 lessons, over 40 prompts, 16 [11:12] [music] [11:13] ready to install skills, $600 in [11:16] credits, monthly live workshops, and 400 [11:20] builders. [11:21] Join by October 7th to lock in $150 [11:26] a year. After that, it's 200. 30-day [11:30] money-back guarantee. [11:33] Copy the AI system Peter uses [music] to [11:36] run his work and life. [11:38] behindthecraft.com [11:42] >> Okay, so I did have to give it some [11:44] feedback. There was some dark screens at [11:47] 11 seconds in. [11:48] And I also told it to add the British [11:50] voice over that you heard. And it [11:52] somehow found the voice over and just [11:54] added it. It found it from voice engine [11:57] Cockerel. I have no idea what the hell [11:58] Cockerel is, but you know, it was able [12:00] to add it. So, this one was pretty much [12:02] one shot, right? Like with some feedback [12:04] from me. [12:05] So, it can make both horizontal product [12:07] launch videos and also vertical shorts. [12:10] All right, one last use case I want to [12:12] demonstrate for launching a product is [12:15] talking head video. And [12:17] by the way, even this tutorial, like [12:18] some of the animations that you saw with [12:20] my face on them, were basically made [12:22] with Sonix, right? But I want to show [12:24] you how to do that. [12:26] So, basically I told it, "Hey, there's a [12:27] landing page video that I recorded and I [12:30] want you to add a bunch of overlays and [12:32] effects to it to make it more engaging [12:34] while keeping my talking head. And you [12:35] can also add custom B-roll if you want." [12:38] And I told it not to use HyperFrames to [12:40] generate it from code. [12:41] And then it made this. [12:43] So, these are the key frames that it [12:44] showed me. [12:46] And the key frames are kind of boring, [12:48] right? Like there's like one B-roll, [12:49] there's some logos here, there's some [12:51] captions. And it's basically following [12:53] my template, but with Sonix 5.5 and Opus [12:57] 5.5, you actually want to tell it to be [12:59] creative and not follow your template [13:01] because it's constantly surprising at [13:03] what it's capable of and it can kind of [13:04] blow you away. So, that's exactly what I [13:06] did here. I told it, "Hey, [13:08] this template is super boring. Forget [13:10] about it. Use your creativity and make [13:13] it as interesting as possible like from [13:15] a professional editing studio." Okay, [13:17] then it came up with something crazy [13:18] like this. [13:19] All right. Now, let me go ahead and play [13:22] the talking head video and you can judge [13:25] if it makes sense or not. [13:27] So, this is talking head video made [13:28] added zoom, added some crazy overlays, [13:31] is able to cut out my talking head from [13:33] the background. And honestly, this is [13:36] pretty exciting. There's some crazy [13:37] characters in here. But this is a little [13:39] bit too in your face for my style. This [13:40] is more like, I don't know, Mr. Beast [13:42] style or someone like, you know, more in [13:43] your face. [13:45] But yeah, it really went all out with [13:46] the creative effects. Some of this stuff [13:49] isn't quite what I want, right? So, this [13:50] needs to be [snorts] more iterations. [13:52] But you can see how far it can go, [13:54] right? Okay. So, uh [13:57] I want to show you this even though I'm [13:58] not going to use this because I want to [14:00] show you that uh Sonnet and Opus now [14:03] can basically edit your talking head [14:05] videos for you, can add all these [14:06] animations or overlays. [14:08] Uh it would cost a lot money to hire a [14:09] video editor to do all this stuff. And, [14:12] you know, a human editor who's good will [14:14] do a much better job than this, [14:15] obviously. But with a few iterations of [14:17] AI, you can get to a pretty good place [14:20] if you want to save money. [14:22] Okay. So, for the last two videos that [14:24] I'll show you, I want to go back to [14:25] something more fun. At first, I want to [14:28] show you an animated music video. [14:31] So, I was really inspired by Donald's [14:33] animated music video here. [14:35] It's basically some crazy song about how [14:38] probably doom is here and AGI is here. [14:40] And yeah, it kind of blew me away. So, I [14:43] want to make something very similar. [14:44] So, I basically told Claude, "Hey, can [14:46] you make a music video similar to this [14:47] one?" [14:48] And I linked the tweet. [14:50] And there's also a prompt that Donald [14:52] provided in the second tweet down here. [14:54] For reference, it's super long. I didn't [14:56] even read through the whole thing, to be [14:57] honest with you. But I basically pointed [14:59] Claude at it. And I want to make a more [15:02] optimistic video about what a young [15:04] person's life would look like with AI [15:07] and robots in the future because there's [15:08] just too much doom and gloom about AI [15:10] right now, you know? So, I want to make [15:11] it optimistic and fun. [15:14] So, let me go ahead and play this video [15:15] for you now and you can see what it [15:17] looks like. [15:32] >> [music] [15:40] [music] [16:04] [music] [16:16] [music] [16:18] >> All right, so this isn't quite, you [16:19] know, Japanese anime quality. Some of [16:21] the shapes are a little bit basic, but [16:23] it's still like incredibly creative, [16:25] right? And like really engaging and [16:27] really fun to watch. And of course, [16:29] Sonic also added sounds and music, too. [16:33] So, this is pretty much one shot, you [16:34] know, I didn't really give it any [16:35] feedback at all. And it basically one [16:37] shot this whole thing. [16:39] All right, so now for the final video, [16:43] I want to show you what it takes to [16:46] actually make a real Japanese animated [16:49] video, or at least something that's [16:50] somewhat close to that. [16:52] Okay, so once again, I was inspired by a [16:54] tweet of someone using Open as 5.5 to [16:56] make this. And this looks like, you [16:58] know, [16:59] kind of more Japanese anime style, [17:01] right? The art. [17:03] All right, so I wanted to figure out how [17:04] to make this. So, I basically pointed [17:06] Sonic at this video and I said, [17:09] "How do you make something like this? [17:10] How do you How do you even do this, [17:11] right?" And it told me that it needs to [17:13] have a song. [17:15] It needs to have some sort of image and [17:16] video generation API because, you know, [17:18] something like this is probably a little [17:19] bit too hard to write from scratch using [17:21] code, at least for it for now. [17:24] And it told me that I should go to this [17:26] website called FAL to get the API key. [17:29] So, okay, the first thing I did was I [17:30] asked it to come up with lyrics [17:32] for the song that we're going to use. [17:35] So, it came up with some lyrics. [17:37] And then we went back and forth on the [17:38] lyrics. And eventually we landed on some [17:41] lyrics here. [17:42] And next what I did is I went to [17:45] suno.com, okay? So, Suno is this amazing [17:48] website where you can make AI-generated [17:49] music for free. So, basically you can go [17:52] to advanced mode in Suno and paste the [17:54] style prompt here. And then you can also [17:56] come here and paste in the lyrics, [17:59] right? [18:00] I paste it here and it just tells you to [18:02] create songs. And what it's going to do [18:05] is it's going to create two or three [18:08] different songs [18:09] of the music that you want. So, then I [18:11] basically played the songs [18:13] and then I can essentially download [18:15] songs right here after it's done. [18:18] Okay, so that's what I did for the music [18:19] and I basically downloaded song and gave [18:21] it to Claude. [18:22] And next I followed his instructions and [18:25] I went to fal.ai, [18:28] okay? So, fal.ai is a platform where you [18:31] can basically use CSDAS, which is a [18:33] video model, to generate video. And then [18:36] you have to go here and add your API [18:38] key. I'm not going to show you here cuz, [18:40] you know, it's my actual API key. I [18:42] think I added around $15 in credits. [18:46] And now I basically told Claude, "Hey, [18:47] you have the music, you have the API key [18:50] to the video model. Can you go ahead and [18:53] generate the video now?" And let's go [18:55] ahead and play what Claude came up with. [18:57] >> Okay, here's the sign. [19:00] No. [19:01] Let me just [music] say it. [19:04] I can finally breathe [19:07] Stars are falling, catch the light. [19:10] Shatter every after it tonight. [music] [19:12] Empires and asteroids [19:18] >> [music] [19:27] >> Okay, so again, this isn't quite [19:28] professional anime quality. [19:30] It doesn't have as much taste, [19:31] obviously, as real human animators, but [19:33] I think it's pretty creative. [19:35] And you know, I believe that if I kept [19:37] on iterating on this a little bit more, [19:39] I could probably make this into a pretty [19:40] good short animated clip. [19:42] Okay, so we just did seven rapid-fire [19:45] demos of how good Suno 5.5 is at making [19:49] videos. [19:50] And by the way, I am on the Max plan, [19:52] which is $200, but I also found both [19:54] Opus and especially Suno 5.5 to be [19:56] really token efficient. So even if [19:58] you're on a $20 plan, I think you can at [20:00] least make a few of these clips without [20:02] worrying about running out of tokens. [20:04] Now, I want to close with a couple tips [20:06] for you to make uh videos similar to [20:07] this. So number one is if you come [20:10] across an incredible video on X, [20:13] just give Opus or Suno the link to the [20:16] tweet, [20:17] and just tell her to make a video in the [20:18] style with your story or with your [20:20] prompt, right? That's That's probably [20:22] the easiest way to copy whatever style [20:23] that you like. Number two, for product [20:26] launch videos or videos that you want to [20:28] match your brand style and guidelines, [20:30] I recommend installing Hyperframes. [20:32] Again, it's free, [20:33] and give it your logos, colors, and [20:36] fonts. [20:38] The more information you give it, the [20:39] more likely it is to match the brand [20:41] guidelines that you actually have. [20:43] Another tip is for videos where you have [20:46] to sync it to music, you know, Claude is [20:48] capable of generating music directly [20:50] from code, but I think it's better for [20:52] you to use Suno to, you know, prompt [20:54] some amazing tracks, [20:56] and just download the music and give it [20:58] to Claude, and tell her to sync the [21:00] video to the song. And also pick the [21:02] best moments from the song to use. [21:04] And uh finally, if you want to make [21:06] really high-quality videos, you need a [21:08] video model like C dance. So, go check [21:11] out fal.ai [21:13] or some other sites. [21:14] And you know, set up some API keys. And [21:17] this stuff does cost money to use these [21:18] video models, but you can make some [21:20] pretty cool anime clips or other clips [21:23] that I showed you. [21:24] Okay. So, to recap, I think Anthropic [21:27] really hit it out of the park with both [21:29] Opus and Sonnet 5.5. I only demo video [21:32] use cases here, but they're also [21:33] incredible for writing, personality [21:36] design, and making slides, which I'll [21:38] show you in a future video. [21:40] You know, in my opinion, video is one of [21:42] the most effective ways, other than in [21:43] person, for people to communicate with [21:45] each other, because it's just like super [21:47] engaging, it's got music, it's got [21:48] animations, and so on, right? And now [21:50] that videos are so easy to make with AI [21:52] and with these models, you can make them [21:54] for pretty much anything. [21:56] You know, for example, I plan to make [21:59] videos to teach my kid about new topics. [22:02] In fact, I think even making the videos [22:03] together with my kid will help her learn [22:06] math or science, or whatever we're [22:08] trying to cover. [22:10] All right. So, go out there and make [22:11] amazing videos with Sonnet 5.5 and Opus [22:15] 5.5. I'll include some of the prompts in [22:17] the video description and the comments, [22:19] but honestly, you really don't need [22:20] them. Just link an incredible video, [22:22] even one that you see in this tutorial, [22:24] and tell it to make something similar. [22:26] As always, if you enjoyed this practical [22:28] tutorial and want more like this, please [22:31] like and subscribe. [22:32] And I'll see you next time.