As of: 2026-10-12 07:22 CEST.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code). Human editor: Adam Bicz.
Canonical page: https://postcutoff.com/v/charliejhills-opus-5-5-vs-astra-vs-fable-video-edit-x/

# Opus 5.5 vs GPT-6 Astra vs Fable 5.1 edit the same raw talking-head video

Charlie Hills (@charliejhills), 1 October 2026, X. 93,197 views as of 11 October 2026.
Kind: Made by AI.
Watch: https://x.com/charliejhills/status/2105682280765350117

## Why it is here

Listed in the awesome-claude-5-5-videos library; ~93k views. Same footage, brief and editing process given to three models.

## Description (written by Gemini from the video)

**Summary**  
Charlie Hills presents a side-by-side video editing comparison between three frontier AI models: Claude Opus 5.5, GPT-6 Astra, and Claude Fable 5.1. Using the exact same raw talking-head footage and identical verbal instructions, each model independently edits and animates the video, showcasing different motion design choices, color correction, and graphic implementations.

**What is shown**  
* **Setup & Workflow Mapping [00:00–00:07]**: A stacked three-way split screen (Opus 5.5 at top, GPT-6 Astra in center, Fable 5.1 at bottom) depicting how each model graphs the initial prompt pipeline ("1 recording", "three editors").
* **Audio & Color Grading [00:08–00:12]**: Each model's visualization of audio cleanup (waveform displays comparing noisy vs. clean audio) and visual color correction with before/after toggles and subtitle insertion.
* **Synchronized Subtitles [00:13–00:17]**: Demonstration of word-level animated subtitles turning yellow in sync with spoken dialogue.
* **Side-by-Side Comparison Layouts [00:18–00:23]**: Layout management comparing two elements simultaneously ("A" vs. "B", "Raw take" vs. "Edit").
* **Dynamic Venn Diagrams [00:24–00:28]**: Animated Venn diagrams illustrating overlapping commonalities between elements.
* **Process Roadmaps [00:29–00:34]**: Generation of visual milestone paths and journey lines that animate progressive nodes ("Record", "Clean", "Edit", "Publish") in real-time as the narrator speaks.
* **Pixel Art Character Animation [00:35–00:40]**: A custom 2D 8-bit pixel-art sprite modeled on the presenter walking across the screen along a guided track.
* **Conclusion & Model Decision [00:41–00:48]**: Recapping summary flowcharts depicting the three models and asking the audience to evaluate which editing style is superior.

**Claims & numbers**  
* The presenter says he gave three AI models—Claude Opus 5.5, GPT-6 Astra, and Claude Fable 5.1—the exact same raw video recording and set of instructions.

**Notable quotes**  
* [00:00] "Which AI edits better? I'm giving Opus 5.5, Astra, and Fable the same recording to see what they do with it."  
* [00:18] "Say I want to compare two things side by side, put them side by side. Now show me what they've got in common: do a Venn diagram."  
* [00:41] "Same video, same instructions, three different AI edits. Which AI would you actually choose?"

**Assessment**  
This is a creator-led comparison demonstration testing multimodal and agentic video-editing capabilities across Claude and OpenAI models. The video highlights stark differences in automated graphic layout and aesthetic choices, though it functions primarily as an observational side-by-side showcase rather than a standardized quantitative benchmark.

**Lyrics & themes**  
The spoken narration acts as a direct command prompt outlining video editing requirements, progressing from basic technical polish to complex graphic motion:
* *Basics & Audio [00:08]*: "So first, I want them to clean up the audio, fix the color, and add subtitles."
* *Interactive Subtitling [00:13]*: "And if I want those subtitles to turn yellow as I'm speaking, they should do that."
* *Data Visualization [00:26]*: "Now show me what they've got in common: do a Venn diagram."
* *Character Animation [00:35]*: "And if I want a little pixel art version of myself walking across the screen, let's add that in as well."

**Lore & references**  
* **Opus 5.5 & Fable 5.1**: Anthropic's flagship reasoning and agentic models, here evaluated on motion graphics composition and script-following capabilities.
* **GPT-6 Astra**: OpenAI's frontier GPT-6 tier model, demonstrating its distinct visual design aesthetic (clean circular nodes, dark-blue minimal cards, and centered typography).
* **AI Video Editing Agents**: Demonstrates the paradigm shift where LLM-based creative agents generate SVG/canvas motion graphics, keyframing, and video layout directly from spoken prompts rather than human NLE editors.

**Visual style & craft**  
The video employs a vertically stacked 9:16 layout on a grid notebook background, with each model assigned its own horizontal canvas. Each AI interprets instructions with different design languages:
* **Opus 5.5**: Prefers rounded diagram cards, modular audio waveform overlays, smooth blue gradient roadmaps, and subtle picture-in-picture frames.
* **GPT-6 Astra**: Uses minimalist stark geometric shapes (large labelled circles, crisp linear typography, high-contrast dark backgrounds).
* **Fable 5.1**: Focuses heavily on linear step-by-step checklists, blue curved sine-wave timelines with labeled milestone beads, and placing the presenter in the lower left corner.

_Described by gemini-3.8-flash on 2026-10-11 from the video's audio and frames._

## Made by AI

- Made with: Claude Opus 5.5, GPT-6 Astra, Fable 5.1
- How we know: X post (2026-10-01): 'I gave Opus 5.5, GPT-6 Astra and Fable 5.1 the same raw video, the same brief and my real YouTube editing process'
- Human role: Human supplied footage and reference, reviewed the first 10 seconds and fixed round by round.
- Pipeline: Raw footage + brief → each model cuts pauses and adds visuals → iterative human review
- Series: Agent video editing (LLM edits real footage)

## Related

- 2026-09-22: [Anthropic releases Claude Opus 5.5](https://postcutoff.com/e/2026-09-22-claude-opus-5-5/)