ChatGPT Image Tailnet

Use the remote Camoufox browser over its tailnet IP instead of changing the current machine's exit-node routing.

Default assumptions for this workspace:

• Preferred remote browser API: http://100.89.48.48:9377
• This is the residential/Türkiye browser path on Inspiron-gnmd-1
• ChatGPT image generation should happen inside that remote browser session
• Downloads should be captured from inside the browser context, not fetched directly from the local host

Why this skill exists#

• ChatGPT/Cloudflare-sensitive flows work better on the residential remote browser than on datacenter/server IPs.
• The reliable path is remote browser over tailnet IP, not reconfiguring a Tailscale exit node for every task.
• Generated image URLs may 403 when fetched from outside the logged-in browser context; trigger the download from inside the page and collect it through Camoufox downloads.

Core workflow#

1. Open https://chatgpt.com/ on the remote Camoufox browser.
2. Switch to Create an image mode.
3. Type the user's custom prompt.
4. Submit and poll snapshots until a generated image appears.
5. Trigger a browser-context fetch/download from the page itself.
6. Read the downloaded file from /tabs/:tabId/downloads and save/send it.

Script#

Use the bundled helper:

bash

python3 skills/chatgpt-image-tailnet/scripts/chatgpt_image_tailnet.py "your prompt here"

Optional flags:

bash

python3 skills/chatgpt-image-tailnet/scripts/chatgpt_image_tailnet.py \
  "your prompt here" \
  --base http://100.89.48.48:9377 \
  --user lotfi \
  --session chatgpt-image-helper \
  --output /tmp/result.png \
  --timeout 180

Important rules#

• Prefer the remote tailnet Camoufox base over any local browser by default for this flow.
• Do not depend on Tailscale exit-node switching for normal operation.
• Keep the prompt as a script input, not as a baked-in macro.
• If direct local curl to the generated image URL returns 403, that is expected; use the in-browser fetch/download path.
• If the remote browser is unavailable, only then consider the local/fallback browser path.

When to inspect manually#

Read scripts/chatgpt_image_tailnet.py when you need to:

• add aspect-ratio controls
• upload an input image before prompting
• send the final file directly over messaging
• adapt selectors if ChatGPT changes the UI