Split a video into multiple videos for uploading to YouTube.
👉 Goal
Instead of manually cutting a long video into multiple shorter clips using video editing software, use Ishi to:
Automatically split one long video into multiple short videos and generate descriptions for each video
Create an agent that generates thumbnail images and video descriptions for uploading to YouTube
⚙️ Overall Process
(Start) ↓
Step 1: Prepare Video File on Your Computer ↓
Step 2: Open ISHI and Start a New Session ↓
Step 3: Prompt ISHI to Analyze & Split the Video ↓
Step 4: Review the Split Results ↓
Step 5: Generate YouTube Metadata (Titles, Descriptions, Hashtags) ↓
Step 6: Create a YouTube Thumbnail Generator Agent ↓
Step 7: Generate Thumbnails for Each Video ↓
(Finish)
🧰 Required Tools
ISHI (Local AI Assistant)
ISHI application running on your local machine (Download here)
Connected to an AI provider that supports tool use (e.g., Claude 3.5 Sonnet, Claude 4.6 Opus, or Gemini 2.5 Pro)
Agentic Flow Platform
Agentic Flow account (agenticflow.ai) — needed only for Step 6 & 7 (Thumbnail Agent)
📝 HOW TO USE
Step 1: Prepare Video File on Your Computer
What you need to do
Save your long video file to a folder on your computer. Make sure you know the full file path.
Process
Create a dedicated folder for this project (e.g.,
~/Desktop/video/my-youtube-project)Place your long video file (e.g., .mp4) inside this folder
Note down the full file path — you will need to provide this to ISHI
Example
TIP
Keep the file path simple — avoid special characters or very long folder names. This makes it easier for ISHI to locate the file.
Step 2: Open ISHI and Start a New Session
What you need to do
Open the ISHI application on your computer and start a new conversation session.
Process
Launch ISHI from your Applications
Click "+ New session" to start a fresh conversation
Verify your AI model is set to a model with tool use capability (check the model selector at the bottom of the chat)
Confirmation
✅ ISHI is open and a new session is ready
✅ Model is set to Claude 3.5 Sonnet, Claude 4.6 Opus, or another tool-use model
WARNING
If you use a model that does NOT support tool use, ISHI will not be able to run commands (ffmpeg, Python) on your machine. The video splitting will fail. Always double-check the model selection before proceeding.
Step 3: Prompt ISHI to Analyze & Split the Video
What you need to do
Provide ISHI with a prompt describing what you want to do, along with the video file path.
Example Prompt
What ISHI will do automatically
ISHI will perform the following steps autonomously:
1
Extract audio from the video
ffmpeg
2
Transcribe the audio to text
Python + OpenAI Whisper
3
Take periodic screenshots to understand visual content
ffmpeg
4
Analyze transcript + screenshots to identify logical segments
AI reasoning
5
Split the video into separate files
ffmpeg
NOTE
This process takes approximately 5–15 minutes depending on the length of your video and your machine's performance. ISHI handles everything automatically — just wait for it to finish.
Troubleshooting: Dependency Issues
If ISHI encounters errors during transcription (e.g., numpy version conflicts), it will automatically attempt to fix them by:
Downgrading incompatible packages
Installing missing dependencies
You can simply wait — ISHI resolves these issues on its own.
TIP
If ISHI gets stuck or produces an error, you can paste the error message back into the chat and ask: "Can you try again?" — ISHI will review and fix the issue.
Step 4: Review the Split Results
What you need to do
After ISHI finishes splitting, verify the output clips are correct.
Process
ISHI will display a summary of all generated clips, including:
Clip number
Start time → End time
Brief description of content
Navigate to the output folder on your computer
Play each clip to verify the cuts are clean and the content makes sense
Example Output
Confirmation
✅ All clips play correctly without errors
✅ Each clip starts and ends at natural transition points
✅ No content is missing or duplicated between clips
Step 5: Generate YouTube Metadata (Titles, Descriptions, Hashtags)
What you need to do
Ask ISHI to generate YouTube-ready metadata for each clip.
Example Prompt
What ISHI will do
ISHI analyzes each clip's transcript and generates optimized metadata:
Title: Catchy, keyword-rich title for YouTube discoverability
Description: Detailed summary with "In this video..." format, relevant links, and chapter timestamps
Hashtags: Topic-relevant hashtags (e.g.,
#AgenticFlow #AI #Automation)
Result
You will receive a structured output for each clip that you can directly copy-paste into YouTube's upload form.
TIP
You can ask ISHI to adjust the tone or style of the titles and descriptions. For example: "Make the titles more click-worthy" or "Add a call-to-action at the end of each description."
Step 6: Create a YouTube Thumbnail Generator Agent
What you need to do
Ask ISHI to create a dedicated Agent on Agentic Flow that automatically generates YouTube thumbnails.
Example Prompt
What ISHI will do
Research available image generation plugins in Agentic Flow
Create a new Agent called "YouTube Thumbnail Generator"
Configure it with the Generate Image plugin
Set the system prompt to analyze video descriptions and craft optimized image prompts
Agent Details
Name
YouTube Thumbnail Generator
Model
GPT-4o Mini (or Gemini 2.5 Pro)
Plugin
Generate Image
Output
16:9 HD thumbnail image
What the Agent Does
Paste any YouTube video description into the chat
The agent analyzes the content to identify the core topic, emotional hook, and target audience
It crafts an optimized image prompt based on YouTube thumbnail best practices (bold, high contrast, clean composition)
It generates a thumbnail image in 16:9 HD aspect ratio automatically
It suggests text overlay ideas you can add manually in an image editor
NOTE
After ISHI creates the agent, you may want to switch the model to Gemini 2.5 Pro for better image generation results. You can do this in the Agent editor on Agentic Flow.
Step 7: Generate Thumbnails for Each Video
What you need to do
Use the Thumbnail Generator Agent to create thumbnails for each of your video clips.
Process
Open the YouTube Thumbnail Generator agent on Agentic Flow
For each clip, copy the description generated in Step 5
Paste it into the agent chat
The agent will generate a thumbnail image
Download the generated thumbnail
Repeat for all clips
Result
You now have:
✅ Multiple short video clips ready for upload
✅ YouTube titles, descriptions, and hashtags for each clip
✅ AI-generated thumbnails for each clip
🎬 Final Result
After completing all steps, you will have:
✅ Multiple short video clips split from your original long video — ready for YouTube upload
✅ YouTube-optimized metadata for each clip:
Catchy titles
Detailed descriptions with chapter timestamps
Relevant hashtags
✅ AI-generated thumbnails in 16:9 HD format for each clip
✅ Complete YouTube upload package — everything you need to publish multiple videos from a single recording
📌 Summary of IMPORTANT NOTES
Use a model with tool use capability
ISHI needs to run commands (ffmpeg, Python) on your machine. Models like Claude 3.5 Sonnet or Claude 4.6 Opus support this. Models without tool use will not work.
ISHI handles dependencies automatically
If there are missing packages or version conflicts (e.g., numpy), ISHI will fix them on its own. Just wait for it to complete.
Verify clips before uploading
Always play back the generated clips to ensure the cuts are clean and no content is missing.
Switch models for better thumbnails
After creating the Thumbnail Agent, consider switching the model to Gemini 2.5 Pro for higher quality image generation.
Builder Mindset
Watching how ISHI works will help you understand how to build agents and automate more tasks. It gets easier with practice!
Last updated