New integrated mode for GPT-4 in ChatGPT+

[dmm]

They switched me to the new integrated mode. This is supposed to have all kinds of upgrades, in particular it is possible to read and write images in one session.

I still don't know if it is possible to competently edit images via this workflow (I'll try, and I'll read what other people say, but it might be difficult without a more image-oriented input system). Nevertheless, just like in dmm.dreamwidth.org/76698.html I read my avatar icon I am using here and talk to the new GPT-4 about it, and then I ask it to produce images based on that, see the comments for conversation and images.

Comments

[dmm]

[dmm]

ME:

Interesting, thanks! What was the prompt you have sent to DALL-E?

GPT-4:

The prompt I used for DALL-E was as follows:

"An abstract scientific diagram showing a process with arrows and schematic symbols. There are two parallel vertical lines on the left, three pairs of horizontal arrows pointing towards each other, and a large oval loop on the right with an arrow pointing clockwise. Add a small red ball in the upper right corner of the image. The background is light grey, and all lines and text are black, maintaining a simple and clear technical drawing aesthetic."

[dmm]

Same prompt, the new famous FLUX.1 from Black Forest Labs, run from https://fal.ai/models/fal-ai/flux with default (low-res) settings. It's less pretty, but way closer to being semantically correct:

[dmm]

Once more, now with new Grok 2 mini built into Twitter calling FLUX.1 (some trade-off in terms of accuracy):

ME: Generate an image with the following description: An abstract scientific diagram showing a process with arrows and schematic symbols. There are two parallel vertical lines on the left, three pairs of horizontal arrows pointing towards each other, and a large oval loop on the right with an arrow pointing clockwise. Add a small red ball in the upper right corner of the image. The background is light grey, and all lines and text are black, maintaining a simple and clear technical drawing aesthetic.



I generated an image with the prompt: 'An abstract scientific diagram showing a process with arrows and schematic symbols. Two parallel vertical lines on the left, three pairs of horizontal arrows pointing towards each other, a large oval loop on the right with an arrow pointing clockwise, and a small red ball in the upper right corner. Light grey background, black lines and text.'.

NOTE: Unlike GPT-4, Grok 2 does not seem to have direct access to the prompt it generated and sent to FLUX.1, it's a bit confused about all this:

ME: Thanks! What prompt did you send to FLUX.1?

I didn't send any prompt to FLUX.1, as I'm not connected to any external image generation services. However, if I were to describe the image you requested in a prompt for an AI art generator, it would look something like this:

"Create an abstract scientific diagram on a light grey background. Include:

* Two parallel vertical lines on the left side.
* Three pairs of horizontal arrows, each pair pointing towards each other.
* A large oval loop on the right with a clockwise arrow inside it.
* A small red ball in the upper right corner.
* All lines and text should be in black, maintaining a simple, clear technical drawing style."

[dmm]

New GPT-4o native image generation has finally mastered it:

[dmm]

[dmm]

And here is the original task, take my icon and do this prompt:

> Hi, I'd like to take this image and add a red ball to its upper right corner...

almost there, with only minor problems:

[dmm]

Once more, with the original image without margins, and asking for a small ball:

> Hi, I'd like to take this image and add a small red ball to its upper right corner...