Skip to content

Troubleshooting & common Issues

This guide addresses common problems you might encounter when deploying your Virtual Influencer.

👻 Twitter Account Visibility ("Ghost Account" / Shadowban)

The Problem

You created a new Twitter account, paid for API access, and started posting. * You can see your posts when logged in. * Others (logged out or other users) see "This account does not exist" or cannot find your profile in search. * Your posts get 0 impressions.

Why is this happening?

This is Twitter's Anti-Spam Protection for new accounts, especially those using the API immediately. If a brand new account (< 48 hours old) starts posting via API without "human" behavior first, Twitter flags it as a potential bot farm and hides it from public view (Search Ban / Ghost Ban).

How to Fix

You are likely NOT blocked, just hidden.

  1. Stop the Bot immediately: Do not let the API post for 24-48 hours.
  2. Warm Up Manually:
    • Log in to the account on a phone (official app).
    • Change your profile picture and bio if you haven't.
    • Follow 5-10 verifies accounts (Blue checks).
    • Like and Retweet a few real posts manually.
    • Reply to a post with text (e.g., "Agree!" or "Nice").
  3. Wait: New accounts often take 24-72 hours to be indexed by search.
  4. Appeal: If it persists after 3 days of manual usage, contact Twitter Support via the Help Center.

🛑 "403 Forbidden" or "401 Unauthorized" on Twitter

Cause

  • 401: Your API Keys or Access Tokens are wrong. Did you regenerate them after changing permissions to Read/Write?
  • 403: Your App lacks "Write" permissions.

Solution

  1. Go to Twitter Developer Portal.
  2. Check User Authentication Settings.
  3. Ensure App Permissions is set to Read and Write.
  4. Regenerate the Access Token and Secret (Consumer Keys stay the same).
  5. Update your .env file.

🖼️ Replicate Error: "NSFW Content Detected"

The Problem

The image generator returns an error instead of an image URL.

Cause

The safety filter on Replicate (or the model itself) flagged your prompt. "Seductive", "Hot", or "Lingerie" keywords can trigger this depending on the model's strictness.

Solution

  • Tone down the prompt: In core/image_gen/pipeline.py, ensure the injected keywords aren't too explicit.
  • Use a different model: Uncensored models exist but may be lower quality. We stick to FLUX because it's SOTA (State of the Art) and usually compliant with ToS.

📋 Monitor & Logs

To see what the bot is doing, use docker compose logs.

1. Watch the Scheduler (The Brain)

To see when the next post is scheduled: 

docker compose logs -f scheduler
* Look for lines like: 📅 SCHEDULE FOR 2026-02-18:...

2. Watch the API (The Hands)

To see errors during the actual posting process: 

docker compose logs -f api

3. Check All Services

To make sure everything is running (and not "Exited"): 

docker compose ps