In this video, I demonstrate a real ChatGPT Codex automation workflow on Windows—sorting local files and photos, detecting duplicate/near-duplicate pictures, organizing them into folders, and then using Chrome to upload selected files into Google Drive.
Codex creates and runs its own helper scripts, analyzes image content instead of relying only on filenames, and can work across the local Windows environment and browser workflow. I also show the permission prompts, how Codex decides between similar photos, and how to make those decisions explainable.
⬥ Template 1 — Main automation prompt: This tells Codex to organize your local pictures, isolate duplicates, identify a specific category, and visibly upload only that category to your chosen Google Drive folder.
------------------------------------------------------------
DO NOT ask unnecessary follow-up questions.
Start immediately using the permissions already granted.
LOCAL SOURCE:
[PASTE YOUR OWN LOCAL PICTURE FOLDER PATH]
GOOGLE DRIVE TARGET:
[PASTE YOUR OWN GOOGLE DRIVE TARGET FOLDER LINK]
TASK 1 — LOCAL FILE ORGANIZATION
Analyze the ACTUAL visual content of every image.
Automatically create logical folders based on recurring subjects.
Examples:
Family
Travel
Car
Documents
Medicine
Miscellaneous_Unidentified
Create additional categories when clearly appropriate.
Move each image into its best category.
If uncertain:
Move it to Miscellaneous_Unidentified.
Do NOT permanently delete originals.
Do NOT reduce image quality.
TASK 2 — DUPLICATE REVIEW
Create:
Duplicates_Redundant_Review
Exact_Duplicates
Near_Duplicate_Inferior
Detect duplicates using actual image content, NOT filenames.
For exact duplicates:
Keep one copy and move redundant copies to Exact_Duplicates.
For near-duplicates:
Keep the better image and move the weaker image to Near_Duplicate_Inferior.
If uncertain which is better:
KEEP BOTH.
TASK 3 — SELECTED GOOGLE DRIVE UPLOAD
Identify only this requested category:
[ENTER CATEGORY TO UPLOAD]
Use my EXISTING Google Chrome tab containing:
[PASTE YOUR GOOGLE DRIVE TARGET FOLDER LINK]
Do not use unrelated Chrome tabs.
Create this destination folder:
[ENTER DESTINATION FOLDER NAME | DATE]
Upload ONLY files belonging to the requested category.
Do NOT upload unrelated categories or duplicate-review copies.
Perform the Google Drive portion visibly through Chrome so the workflow
can be screen-recorded.
After upload:
Compare the local selected-file count with the Google Drive uploaded count.
Do not ask questions unless a mandatory Windows, Chrome, Google,
or Codex security approval cannot be bypassed.
Continue automatically until finished.
---------------------------------------------------------------------------
⬥ Template 2 — Best-photo / near-duplicate prompt: Use this when you want Codex to prove why it kept one similar photo and moved another instead of making an unexplained choice.
-------------------------------------------------------------------------
Continue the near-duplicate photo analysis.
Do NOT use filenames to decide which image is better.
Compare the actual visual content.
For each near-duplicate group, evaluate:
Sharpness
Blur
Exposure
Useful resolution
Face visibility
Eyes open, when reliably detectable
Subject visibility
Framing
Obstructions
Overall usable quality
Keep the better photo.
Move the weaker photo into:
Near_Duplicate_Inferior
If you cannot confidently determine which photo is better:
KEEP BOTH.
Do not claim that you measured faces, eyes, or another quality
unless your actual analysis tools measured it.
Create a report showing:
Kept_File
Moved_File
Similarity
Quality difference
Decision confidence
Reason kept
Example:
"Both images show the same scene.
Photo A was retained because it is sharper and all visible faces
have open eyes. Photo B has slight motion blur."
Do NOT permanently delete either image.
Continue automatically without unnecessary questions.
Codex creates and runs its own helper scripts, analyzes image content instead of relying only on filenames, and can work across the local Windows environment and browser workflow. I also show the permission prompts, how Codex decides between similar photos, and how to make those decisions explainable.
⬥ Template 1 — Main automation prompt: This tells Codex to organize your local pictures, isolate duplicates, identify a specific category, and visibly upload only that category to your chosen Google Drive folder.
------------------------------------------------------------
DO NOT ask unnecessary follow-up questions.
Start immediately using the permissions already granted.
LOCAL SOURCE:
[PASTE YOUR OWN LOCAL PICTURE FOLDER PATH]
GOOGLE DRIVE TARGET:
[PASTE YOUR OWN GOOGLE DRIVE TARGET FOLDER LINK]
TASK 1 — LOCAL FILE ORGANIZATION
Analyze the ACTUAL visual content of every image.
Automatically create logical folders based on recurring subjects.
Examples:
Family
Travel
Car
Documents
Medicine
Miscellaneous_Unidentified
Create additional categories when clearly appropriate.
Move each image into its best category.
If uncertain:
Move it to Miscellaneous_Unidentified.
Do NOT permanently delete originals.
Do NOT reduce image quality.
TASK 2 — DUPLICATE REVIEW
Create:
Duplicates_Redundant_Review
Exact_Duplicates
Near_Duplicate_Inferior
Detect duplicates using actual image content, NOT filenames.
For exact duplicates:
Keep one copy and move redundant copies to Exact_Duplicates.
For near-duplicates:
Keep the better image and move the weaker image to Near_Duplicate_Inferior.
If uncertain which is better:
KEEP BOTH.
TASK 3 — SELECTED GOOGLE DRIVE UPLOAD
Identify only this requested category:
[ENTER CATEGORY TO UPLOAD]
Use my EXISTING Google Chrome tab containing:
[PASTE YOUR GOOGLE DRIVE TARGET FOLDER LINK]
Do not use unrelated Chrome tabs.
Create this destination folder:
[ENTER DESTINATION FOLDER NAME | DATE]
Upload ONLY files belonging to the requested category.
Do NOT upload unrelated categories or duplicate-review copies.
Perform the Google Drive portion visibly through Chrome so the workflow
can be screen-recorded.
After upload:
Compare the local selected-file count with the Google Drive uploaded count.
Do not ask questions unless a mandatory Windows, Chrome, Google,
or Codex security approval cannot be bypassed.
Continue automatically until finished.
---------------------------------------------------------------------------
⬥ Template 2 — Best-photo / near-duplicate prompt: Use this when you want Codex to prove why it kept one similar photo and moved another instead of making an unexplained choice.
-------------------------------------------------------------------------
Continue the near-duplicate photo analysis.
Do NOT use filenames to decide which image is better.
Compare the actual visual content.
For each near-duplicate group, evaluate:
Sharpness
Blur
Exposure
Useful resolution
Face visibility
Eyes open, when reliably detectable
Subject visibility
Framing
Obstructions
Overall usable quality
Keep the better photo.
Move the weaker photo into:
Near_Duplicate_Inferior
If you cannot confidently determine which photo is better:
KEEP BOTH.
Do not claim that you measured faces, eyes, or another quality
unless your actual analysis tools measured it.
Create a report showing:
Kept_File
Moved_File
Similarity
Quality difference
Decision confidence
Reason kept
Example:
"Both images show the same scene.
Photo A was retained because it is sharper and all visible faces
have open eyes. Photo B has slight motion blur."
Do NOT permanently delete either image.
Continue automatically without unnecessary questions.
- Категория
- AI Инструменты
Комментариев нет.
