Carolletta reports Grok Bot catalogued 20,000 Midjourney images

• Grok Bot was building a catalogue of 20,000 Midjourney images, according to carolletta’s update.
• A request to pick favourites expanded into 366 session ZIPs totalling 46 GB, as carolletta reported.
• Storage problems and constant quality review constrained the workflow; carolletta called herself the bottleneck.

Every Bot on an account shares the same persistent cloud computer, including its files and browser sessions. Alongside another 3 a.m. wake-up, carolletta’s updates included a hefty 75 GB cleanup job on the bots’ computer.

Carolletta described Grok Bot as creating the catalogue, with work still underway. The bulk archive had exceeded her original request to select the best images.

She wanted to stay involved in the bot’s decisions, saying it should disclose its assumptions rather than silently act on them.

Midjourney’s own Organize page supports bulk downloads, including selecting an entire day’s images at once. Carolletta did not describe the catalogue’s indexing method or selection criteria.

After a failed Google Drive attempt, carolletta wondered whether files needed to be smaller than 9 GB. She also floated a hosted database and asked whether Grok Bot could keep accessing Drive.

Grok Bot’s official file documentation describes separate rules for connected storage and chat attachments:

• Connected Google Drive accounts can exchange files with the bot’s computer without passing their contents through chat.
• Saving to Drive creates a new file, even when another file has the same name.
• Documents and other files attached in desktop chat have a 250 MB-per-file limit.

The documented attachment cap applies to desktop chat, rather than establishing a 9 GB limit for Drive transfers.

Carolletta reported mixed results from short instructions:

• Typography: The bot understood a request to maintain good margins between type, shapes and other elements.
• Drawn feedback: It could follow instructions drawn by hand.
• Hotel search: Giving “Faena Casa” as a reference did not communicate what she wanted for a Miami stay; she thought the task needed a longer prompt.

She said the work still required constant quality review.

Carolletta described continually improving prompts before packaging them into skills.

Task management needed separate attention. Instructions such as “park this” and “don’t work on this now” left work needing reminders to resume.

Her two-month planning notes proposed a “Chief of Staff” checkpoint if the bots forgot unfinished work. She was also considering “Grok Templates”, while questioning how much highly particular polish could be standardised.