How to Use Gemini Flash for Agentic Video Work

TL;DR

Gemini 3.8 Flash combines much faster generation, low pricing, stronger long-task execution, and especially capable video understanding. Its practical uses include producing designed slide decks, turning narrated recordings into illustrated manuals, editing and searching video, generating short clips, and supporting live-stream workflows through Gemini 3.7 Agentic Video.

Transcript

Hello, everyone. This is Keto. This channel introduces the latest information and use cases of Seisei. Please subscribe to the channel. The theme of this video is the new models Gemini 3.8 Flash and Geni 3.7 Agentic Video, which have been added, so I would like to explain them. Well, the 3.8 has evolved to some extent and its performance has improv... Read More

Key Insights

• Gemini 3.8 Flash is described as approaching Claude Opus 5 for some capabilities, although its overall evaluation is said to remain below Claude Fable and GPT-5.6 Sol. Its strongest advantage is understanding video, making selective use more appropriate than treating it as universally superior.
• Gemini 3.8 Flash is available through the Gemini web interface and Google's AntiGravity agent tool. AntiGravity supports low, medium, and high thinking settings, and the presenter reports especially strong results when the model is used with the high setting.
• Long-running task execution is a major improvement in Gemini 3.8 Flash. Earlier Flash models appeared to think without consistently finishing work, while the newer model is described as more capable of reading skill context carefully and completing multi-step production tasks.
• Generation speed is a defining strength of Gemini 3.8 Flash. The presenter cites a comparison showing 305 for Gemini against 72 for GPT-5.6 and describes Gemini as roughly four times faster than Claude Opus, while acknowledging uncertainty about the graph's exact measurement.
• A five-page slide deck was completed in about two minutes through AntiGravity. The model researched Gemini 3.7 Agentic Video, followed a prepared PowerPoint skill, incorporated diagrams, and reproduced the design system used in the presenter's existing materials.
• Video-understanding performance is the model's clearest benchmark advantage. The cited long-and-short-video benchmark gives Gemini 3.8 a score of 87.8%, compared with 75.4% for Claude Opus 5, 68% for Claude Sonnet, and 82% for GPT-5.6.
• Video recordings can be converted into illustrated operating manuals. The demonstrated workflow analyzes narration and screen activity, extracts relevant frames, converts the explanation into text, and can add red frames or arrows to clarify precisely where the user clicked.
• Gemini 3.7 Agentic Video supports more than offline analysis. The presented applications include video editing, searching within video, automatically generating short-form clips, and using real-time video analysis in live streams while consuming fewer AI tokens than previous approaches.

Questions & Answers

Q: What is Gemini 3.8 Flash best used for?

Gemini 3.8 Flash is best suited to fast, affordable production tasks and workflows requiring strong video understanding. The examples include researching a topic, following an embedded design system, creating PowerPoint materials, analyzing recorded demonstrations, and converting footage into manuals. The presenter recommends selective use because other models may remain stronger overall, while Gemini performs particularly well for video-centered work.

Q: How fast is Gemini 3.8 Flash for document creation?

Gemini 3.8 Flash created a five-page explanatory slide deck in about two minutes during the AntiGravity demonstration. The task included researching Gemini 3.7 Agentic Video, following a prepared PowerPoint skill, applying an embedded design system, and adding structured visual content. The presenter contrasts this with a comparable Claude Code workflow that could take approximately five to ten minutes.

Q: How does Gemini 3.8 Flash perform on video understanding?

Gemini 3.8 Flash scored 87.8% on the cited benchmark for understanding long and short videos. The comparison shown in the presentation placed Claude Opus 5 at 75.4%, Claude Sonnet at 68%, and GPT-5.6 at 82%. The presenter treats this category as Gemini's most important advantage and recommends the model particularly for workflows centered on video analysis.

Q: How can Gemini turn a video into a manual?

Gemini can analyze a narrated recording, identify the important moments, extract useful frames, and convert the spoken explanation into a text-and-image manual. In the demonstrated workflow, the user records the steps for creating PowerPoint slides in AntiGravity, supplies the recording's file path, and asks a prepared skill to add annotations such as red frames and arrows around clicked controls.

Q: What does AntiGravity do with Gemini 3.8 Flash?

AntiGravity serves as Google's agent tool and text-editing environment for running Gemini 3.8 Flash with adjustable thinking levels of low, medium, or high. It can load prepared skills, accept natural-language instructions, read project context, and complete production work. The demonstrated uses include generating PowerPoint and PDF files, following a design system, and creating manuals from recorded software walkthroughs.

Q: Why is Gemini 3.8 Flash useful for long tasks?

Gemini 3.8 Flash is described as better at sustaining and completing long-running tasks than earlier Flash models. The presenter says previous versions could appear to think without fully working through or finishing the assignment. The newer model more reliably reads the context contained in agent skills, follows embedded procedures, reproduces specified designs, and carries multi-step document-production work through to completion.

Q: What can Gemini 3.7 Agentic Video do?

Gemini 3.7 Agentic Video can analyze video in real time while using fewer AI tokens than earlier approaches. The applications presented include video editing, adding information to footage in real time, searching within video, generating short-form clips automatically, and supporting live-stream tools. It can also contribute to manual creation by interpreting footage, narration, screen actions, and relevant captured frames.

Q: How should Gemini 3.8 Flash fit into an AI workflow?

Gemini 3.8 Flash can be assigned fast, lower-cost tasks, especially those involving videos, slide generation, or established agent skills. The presenter suggests using Claude Fable as the main tool and Gemini 3.8 for easier work because Gemini can feel less capable in some broader areas. This allocation takes advantage of Gemini's low price, high speed, and specialized video-understanding performance.

Summary & Key Takeaways

• Gemini 3.8 Flash is presented as a practical model with stronger overall performance, better sustained reasoning, low pricing, and exceptional generation speed. The presenter recommends allocating it to lighter or video-centered work while retaining another model as the primary option when Gemini still feels less capable on broader tasks.

• A five-page explanatory slide deck was generated through AntiGravity using a previously prepared PowerPoint skill and embedded design system. The process took about two minutes and reproduced the intended visual style, suggesting that Gemini 3.8 Flash can follow skill context reliably while completing practical document-production workflows at high speed.

• Gemini's strongest area is video understanding. The demonstrated workflow turns narrated screen recordings into manuals by selecting useful frames, transcribing explanations, and adding visual indicators such as red boxes and arrows. Agentic Video is also positioned for editing, searching within footage, producing short videos, and supporting live-stream applications.