[BLU/Officers] Need help with BLU presenation

[John Abreau]

Hi, guys, I've been playing with AI for a few months, and I've hd some
remarkable
successes that I'd like to share in a presentation at one of our
upcoming BLU meetings.

The problem is, I have no real experience preparing a presentation.

Can we get together sometime soon so I can show you what I've accomplished?

I need help preparing this into a proper presentation.

Last night's effort was particularly notable. I was having trouble
generating
Images for creating a consistent character to be used in AI videos, and in
In a moment of frustration I typed, "Why are you refusing to generate my
video requests?"What am I doing wrong?"

It was just an expression of frustration, but Gemini responded and explained
the technical details. We started working together to figure out what was
wrong,
and by the end of the session, Gemini was generating a report ffor me to
submit
to Gemini's feedback page, explaining in detail a serious buffer overflow
problem.
Along the wy there were several moments when the AI had to invent new
terminology to define what it had learned from me.

For example, characters were anchored ito consistency with a reference
image
and a character description. The AI understood that each image had a unique
id
that can anchor a character's 3D model to  consistent presentation when the
text describing the character used the image as a reference fir the various
bits
of metadata it described. In one of times I seered the AI, I instructed it
to retain
the inputs and results of two generations I was to promt to test a hunch.
I used the original soiurce image and the original text I always used wiih
that image, andthen used the non-functioing text the I had been using as a
set of modifications to the original test. When I ran the first test,it
completety fixed
the problem the AI had been strugging with.

I then repeated the test with the identical test but a diffrent image, and
image t
that I had generated with the first image as its reference image.
The results were picel-by-pixel identcia.

Whn i toldthe AI that they were pixel-by-pixel identical, that's the moment
that
the AI learned that it waasn't the image itself that provided the anchor,
it ws its unique id. Which wasn't uniqie after all, becuse all the differnt
views of the character - front, bck, profile eetc shared the same id.

The I intented the term "morphological id" to replace the previous
the now-proven-inaccurate term unique id.

My orifgial problem occurred when used used the original source image
to generate an image of the character thant ws 10 years older, then used
that older image to anchor a new set of reference images. By
creatingtwocharacters with the same morphological id, evert=y time I trired
to generate either of the two
cjharacters, it caused an infinite loop that, among other messes, caused
the buffer overflow.

Now when the Nano Banana tech guys receive the bug report they can fix the
problem.

--
John Abreau / Executive Director, Boston Linux & Unix
Email: abreauj at gmail.com / WWW http://www.abreau.net / PGP-Key-ID 0x920063C6
PGP-Key-Fingerprint A5AD 6BE1 FEFE 8E4F 5C23  C2D0 E885 E17C 9200 63C6
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.blu.org/pipermail/officers/attachments/20260504/af32cf34/attachment.html>