Mission Beach, San Diego California yesterday.

Being that time of year and weather, the beach was littered with many by-the-wind sailors (Velella velella). Almost all of them were transparent, with only a couple blue ones (not shown in these photos).

"Bard just got more 🏞 visual"

See attached screenshots:, an email I received & a Bard search I did (one of their example prompts).

Images really enhance text for most people.

This is a smart move by Google, because they're really good already at using AI to understand images, so Bard can choose good images. And they're not using synthetic images, but rather images in web pages that are in the web search index, so they're sourced, and someone else is on the hook for image licensing, not them.

Harmony Revealed

In the heart of the wood, a hidden pond nestled amongst skeletal trees.

Silhouettes soared skyward toward symphonic color, a dazzling sight indeed.

Twilight descended, the ethereal haven glistened, and ripples mirrored the remarkable sheen.

Travelers paused, entranced, while Earth herself breathed deeply in tranquil repose.

A contrast so true, solemn trees and luminous sky rising in crescendo.

A tender equilibrium, a harmony concealed, was gently revealed.

#Graphviz or #Mermaid #Diagrams for #systems #visualization with #ChatGPT - ask it to create "Mermaid diagram source" or "Graphviz source" for outlines of bullets, paragraphs, etc. GPT-4 works better than 3.5

Render diagrams with https://mermaid.live/ or https://dreampuf.github.io/GraphvizOnline/ - ymmv with complicated diagrams.

Here's someone using Neo4j instead: https://neo4j.com/developer-blog/chatgpt-4-knowledge-graph-from-video-transcripts/

Image shows an example system diagram created by GPT-4 in Graphviz format from human-created bullets describing a system:

#ChatGpt #ChatGPT4 is now available to Plus subscribers!

Current cap is 100 messages every 4 hours.

Face Forward v1

I synthesized 825 images to briefly assess how good current versions of Stable Diffusion and wavymulder's Analog Diffusion model are at faces. I was looking for a diversity of types of faces and photography styles, and after weeding out the duds, I ended up with 144 (17.5%) images that look "95%" good without fixes. Each image is as-generated at 768x512 pixels, no upscaling.

For more info, see blog post:

https://peterkaminski.wiki/face_forward_v1

or Flickr album:

https://www.flickr.com/photos/pixelthesia/sets/72177720305544773