NEW 클로드 Opus 5.5한테 유튜브 100% 맡김 (촬영, 녹음, 편집 ❌) 오퍼스 5.5 레전드입니다...🙀
AI하쥬 · 2026-09-24 · ai-made · 4,523 views · 한국어
Made by AI
Model: Claude Opus 5.5 · Series: "Made This Entire Video By Itself"
Evidence: Description (Korean): 'This video is one that Claude Opus 5.5 made 100%: planning, script, filming, recording, editing.'
Human role: Gave the task; Higgsfield-sponsored.
Pipeline: Opus 5.5 + Higgsfield MCP → script, AI-generated footage and voice, edit
Lore: made-it-by-itself
What's in the video
Description written by Gemini, which watched and listened to the whole video.
Summary
Korean AI creator channel AI하쥬 (AI Haju) presents an explainer video ostensibly produced end-to-end by Anthropic’s Claude Opus 5.5 connected to Higgsfield via Model Context Protocol (MCP). The avatar presenter outlines the architecture and benchmark improvements of Opus 5.5 over Opus 5 and Fable 5.1, demonstrates how to link Claude with Higgsfield tools to generate multimedia, and breaks down the exact workflow, timeline, and cost required for Claude to write, direct, generate assets for, and edit the video.
What is shown
- [00:04] – [00:12] Montage of autonomous creative tasks: After Effects game creation, car motion tracking, 3D skull reconstruction (Homo longi), botanical simulations, and Blender tower collapse physics.
- [01:18] – [01:26] Graph showing time required for a 200,000-line code audit comparing Opus 5 (20+ hours) against Opus 5.5 (<3 hours).
- [01:46] – [02:29] Direct model comparison charts across Opus 5, Fable 5.1, and Opus 5.5 on Terminal-Bench 4.0, GDPval-AA, and API token pricing.
- [02:42] – [02:51] Anthropic UI thinking effort slider ("생각 강도") illustrating settings from Low to Max, noting token output scaling.
- [03:14] – [04:08] Showcase of Higgsfield x Opus 5.5 multi-step workflows: generating playable After Effects mini-games, complex 3D aircraft exploded-view diagrams ($60 in 40 min vs. $143 in 70 min on previous models), 3D VFX simulations, simulated fly neural evolution, and dynamic WebGL/Three.js sites ("Meet Aiko").
- [04:47] – [05:08] Conceptual papercraft animation demonstrating remote desktop agent control via the Claude mobile app while away from home.
- [05:30] – [05:47] Step-by-step setup in Claude settings connecting the custom MCP server connector (
https://mcp.higgsfield.ai/mcp) exposing 37 tools (generate_image,generate_video, etc.). - [05:48] – [06:18] Live execution in Claude: Opus 5.5 processes a Korean prompt requesting a 5-second papercraft diorama video of a laptop timeline editor, invoking tool calls to generate the base image, queue video animation, and output self-evaluative review text.
- [06:19] – [06:23] Playback of the generated 5-second papercraft diorama animation clip with audio.
- [06:24] – [06:30] Higgsfield web interface showing Opus 5.5 selectable inside the "Supercomputer" mode.
- [06:33] – [06:58] Breakdown of production costs and timeline breakdown for the video.
Claims & numbers
- Release date: Anthropic released Claude Opus 5.5 on September 22, 2026 (the presenter states at [00:58]).
- Code review benchmark: For a 200,000-line codebase audit, Opus 5 took over 20 hours, whereas Opus 5.5 completed it in under 3 hours (Anthropic customer case study cited at [01:18]–[01:26]).
- Terminal-Bench 4.0: Opus 5 scored 52.3%, Fable 5.1 scored 55.8%, and Opus 5.5 reached 66.4% ([01:55]–[02:05]).
- GDPval-AA benchmark: Opus 5 scored 1,708 Elo, Fable 5.1 scored 1,735 Elo, and Opus 5.5 scored 1,846 Elo ([02:06]–[02:10]).
- API pricing (Input / Output per 1M tokens):
- Opus 5: $5 / $25 ($0.225 for a standard benchmark task).
- Fable 5.1: $10 / $50 ($0.45 for the same task).
- Opus 5.5: $4 / $20 ($0.18 for the same task; 20% cheaper than Opus 5 and 60% cheaper than Fable 5.1) ([02:11]–[02:29]).
- Performance specifications: Output speed increased by +30%, subscriber 5-hour usage allowance expanded by +20%, and context window remains 1M tokens ([02:30]–[02:35]).
- Thinking tokens: At maximum thinking intensity, a single task outputs approximately 119,000 tokens on Opus 5.5 compared to 73,000 tokens on Opus 5 ([02:46]–[02:49]).
- Video production cost & time for this video:
- Higgsfield credits used: ~770 credits (approx. 35,000 KRW under Ultra plan pricing).
- Claude subscription: Claude Max (no additional marginal cost).
- Total elapsed time: ~4 hours 30 minutes (Research/scripting: 40m; Asset generation: 35m; Screen recording & editing: 45m; Feedback iteration: 2h 30m) ([00:21], [06:33]–[06:53]).
Notable quotes
- [00:00] "지금 보고 계신 이 영상 제가 만든 게 아닙니다." ("The video you are watching right now was not made by me.")
- [01:01] "한 줄로 요약하면 이거예요. 시키면, 끝까지 한다." ("If summarized in one line, it's this: if you tell it to do something, it finishes it to the end.")
- [07:38] "앞으로는 AI한테 이거 해줘가 아니라 이 프로젝트 맡아줘라고 말하는 시대가 올 거예요." ("In the future, rather than telling AI 'do this task,' the era will come where we say 'take charge of this project.'")
Assessment
This video is a detailed creator review, practical workflow demonstration, and product integration guide exploring Claude Opus 5.5 via Higgsfield's MCP server. While the narrative framing presents the video as fully created and edited autonomously by Opus 5.5, the execution incorporates standard scripted YouTube presentation tropes, curated screen recordings, animated infographics, and a live step-by-step tool invocation that convincingly highlights real MCP tool calling and image-to-video generation capabilities.
Lyrics & themes
The video features spoken narration in Korean structured across thematic sections:
- Intro & Claim [00:00 - 00:54]: Announcement that the video's research, script, assets, recording, and editing were delegated autonomously to Opus 5.5.
- "기획, 대본, 자료 조사, 인포그래픽, 화면 녹화, 그리고 편집까지 처음부터 끝까지, AI가 혼자 해냈어요." ([00:03]–[00:11])
- Core Upgrades [00:56 - 01:42]: Transitioning from question-answering LLMs to multi-step executing agents that exhibit adaptive thinking and concise scriptwriting.
- "질문에 답하는 모델이 아니라 수십 단계짜리 긴 작업을 처음부터 끝까지 굴리는 데 초점을 맞췄어요." ([01:05]–[01:12])
- Benchmark & Pricing Comparison [01:43 - 03:11]: Comparing Opus 5.5 against Opus 5 and Fable 5.1 on Terminal-Bench, GDPval, cost efficiency, and speed.
- "더 똑똑한데, 더 싸고, 더 빠르다. 이게 이번 업데이트의 핵심이에요." ([02:36]–[02:41])
- Agent Workflows & MCP Setup [03:12 - 06:30]: Demonstrating complex end-to-end creative workflows and configuring the Higgsfield MCP tool suite inside Claude.
- Production Audit & Practical Tips [06:31 - 07:47]: Disclosing the project's exact financial cost (770 credits), time investment, and advice for framing prompts with clear guardrails and intermediate checkpoints.
- "한 번에 완벽을 바라지 말고, 결과를 보고 스스로 고치게 하세요." ([07:28]–[07:32])
Lore & references
- Claude Model Hierarchy (Opus 5, Fable 5.1, Opus 5.5): Highlights Anthropic's release cadence spanning Opus 5 (July 2026), Fable 5.1 (early September 2026), and Opus 5.5 (September 22, 2026), contrasting Fable's niche deep-reasoning role against Opus 5.5's cost-effective agentic execution.
- Model Context Protocol (MCP): References Anthropic's open standard for letting Claude seamlessly invoke external developer tool ecosystems, demonstrated here via Higgsfield’s hosted MCP service (
mcp.higgsfield.ai/mcp). - Higgsfield AI Ecosystem: Showcases Higgsfield's tools for multi-modal generation (image generation, video motion generation, and its web-based "Supercomputer" interface).
- Autonomous Project Agent Vision: Refers to the transition of AI from short-horizon prompt-and-response chat assistants to long-horizon autonomous operators capable of error recovery, directory management, and pipeline iteration.
Visual style & craft
- Presenter & Studio: A clean digital avatar presenter in a sunlit modern studio with photorealistic textures and subtle lip-sync motion.
- Motion Graphics & UI Demos: Clean paper-textured 2D motion graphic overlays, benchmark bar graphs, and annotated terminal screenshots with UI callout badges.
- Higgsfield Asset Visuals: Includes stop-motion-style papercraft cutouts, 3D exploded engineering models of fighter jets, Blender node graphs, and procedural cellular/particle simulations.
- Evidence of Craft: Real desktop UI recordings of the Claude MCP connector interface, JSON payloads, and live generation outputs are interspersed with pre-rendered graphical slides and stop-motion animations assembled according to an automated video production pipeline.
Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.