[Fillmore]

Originally Posted by PGA

I use ChatGPT to clarify a lot of doubts about agriculture and plant nutrition..

Originally Posted by sagarpadaki

You cannot expect quality value adding content from such mediocre content creators..

Originally Posted by SmartCat

The best way to get more out ChatGPT is to talk to it like you would to a person...

Originally Posted by DigitalOne

Correct. By sheer force of habit, I also include 'please'...

I found these settings pretty useful and of value add so thought of sharing in case you were not aware.
And the kicker is that : These tips were NOT on that paid course I signed up for


Tips
On the Desktop Chat GPT app page click on your Profile Icon bottom left > Custom Instructions.

Setting:1
What would you like ChatGPT to know about you to provide better responses?
Here you can set instructions about your skills and background
eg :

I am a software developer familiar with languages like python, Node JS, Oracle , SQL, Pl/Sql Javascript.
I also like to do a lot of DIY stuff and am handy with a few power tools.

(This way it will tune its replies based on your background and not give you too many unwanted explanations)

Setting:2
How would you like ChatGPT to respond?

Sample answer

Always inform about the confidence level of your answer.
When you answer facts always include a valid URL which is the source of your answer.
When you speculate something please inform me.

(This way you will get a flagging for each response with a
Confidence level: High / Medium / Low
Also when it states facts, it will include the URL from where it is pulling this stuff ..
And will tell you if it is speculating something)

[SmartCat]

We can now get free access to DALL-E via Microsoft Edge browser side pane:



or via the link
https://designer.microsoft.com/

I asked it to create an image of Voldermort sitting in his living room, staring menacingly at you. This is what it created:



I then asked it imagine how Apple's first EV will look like:



and its interiors:



And when I asked it to create new 'racy' TeamBHP logos:





Note that the output primarily depends on 'quality' of your input. Your image request should be as descriptive as possible. For each request, it gives you four image ideas. You can choose one and modify it to your liking using 'Microsoft Designer':



The modification aspect too is "AI enabled".

[sinner]

I've been having quite a lot of fun with Gen AI weather its editing pictures in Photoshop or using other AI Websites.

Recently I've started liking the BING Image Generator quote a bit.

Link

Prompt : An Indian red Color 2023 2 door Mahindra Thar in Hotwheels packaging. image to be in Verticals 1080P dimension

Result:


Prompt : 2020 Mahindra Thar Red Color in Pixar car style with eyes

Result:

[vredesbyrd]

Had been using MidJourney for for than a year with mixed results (mostly good) and then deemed it too expensive.

Bing Image Creator is a game-changer in terms of how close it gets to the prompts.
I am known in my circle for unhinged artwork, so here's some of that:

[nimbleguru]

If you folks are interested in having a open source, standalone & FREE desktop AI creator, try AUTOMATIC1111. For a advanced users, you can try ComfyUI.

You can download all sorts of AI models from huggingface or civitai websites.

Here's a guide to setup:
https://stable-diffusion-art.com/install-windows/

or if you prefer to watch:

https://www.youtube.com/watch?v=kqXpAKVQDNU

[vsathyap]

I just stumbled upon this thread on a lazy Saturday evening - and boy oh boy - TBHP Never disappoints!

Just tried my hand at Generative images and I wasnt disappointed. I tried to see what my Alto K10 would look like from the GPT's perspective and it gave me some really neat designs! Take a cue Maruti! Suzuki ...and stop releasing boring hatches!

Dont ask me why Baby King Kong is in the background!

[Samfromindia]

Originally Posted by vredesbyrd

Had been using MidJourney for for than a year with mixed results (mostly good) and then deemed it too expensive.

Bing Image Creator is a game-changer in terms of how close it gets to the prompts.
I am known in my circle for unhinged artwork, so here's some of that:

Master Chief having paratha made my day Following this thread to discover more innovative ideas.

[SmartCat]

Originally Posted by vredesbyrd

Had been using MidJourney for for than a year with mixed results (mostly good) and then deemed it too expensive. Bing Image Creator is a game-changer in terms of how close it gets to the prompts. I am known in my circle for unhinged artwork, so here's some of that:

From what I understand, each picture is associated with a bunch of words. So to get the best output, our image request should have all the important keywords?

[vredesbyrd]

Originally Posted by SmartCat

From what I understand, each picture is associated with a bunch of words. So to get the best output, our image request should have all the important keywords?

Yes, things like 4k or hyper realistic for extra texture and detail and also specific key phrases like sunset with two suns, green sun and orange sun. The more details you can provide, the better accuracy (though not guaranteed).

I believe it takes negative prompts too, but haven't tried them so far.

From what I have experienced, bing is definitely better at understand the prompts than midjourney and especially stable diffusion (maybe I am running an old model or something).

My next step is to try and generate some liminal spaces with brutalist architecture, something similar to what bolek_3d (instagram) does.

It is a tall order but then again, Rome wasn't built in a day!
Here's a sample:

[cogWheel]

Those of you who have read Agatha Christie's "Five Little Pigs" may recall that a painting of a young woman is an important part of the plot. The key event unfolds as this portrait is being painted. There are multiple references to this scene in the novel and a description of the painting by Hercule Poirot when he sees the painting for the first time. When I was reading this book I have often tried to imagine this painting.

So, I drafted a prompt, based mainly on Poirot's description and fed it to Microsoft's Dall-E based image generator (https://designer.microsoft.com/image-creator) and also to StabilityAI's Stable Diffusion (https://stablediffusionweb.com/#ai-image-generator)

The prompt that I settled on after some tweaks is as follows:
A painting of an early 20th century English girl in canary-yellow shirt and dark-blue slacks, sitting on a grey wall in full sunlight against a background of violent blue sea. The girl is looking at you, over her shoulder, watching you. She is a superb, slim, straight girl, arrogant, her head turned, her eyes insolent with triumph. The girl is full of life, of passionate youth, of sheer blazing vitality. A painting with an old superficial smoothness, which at first sight may look like a poster, with seemingly crude contrasts. Should have subtle distortions with an amazing brilliance and clarity in the light.

Here are the best of the outputs:


Microsoft Designer (no art style specified)


Microsoft Designer (art style: abstract expressionism)


Stable Diffusion (art style: abstract expressionism)


Stable Diffusion (art style: cubism)

I think the Dall-E based Microsoft Designer was definitely better at translating the word portrait into an image. It got the pose, look and scene elements better. Stable Diffusion's images were not so faithful to the word descriptions. However, it was better at applying art styles. .

Agatha Christie fans - was this how you imagined Elsa Green's portrait to be?

Spoiler Alert - Begin
Interestingly, when I tried to to include the sentence "It is a picture of a murderess painted by her victim, a picture of a girl watching her lover die", Microsoft's Designer refused to draw it saying that some words in the prompt may have triggered Microsoft's responsible use of AI guidelines.
Stable diffusion had no such qualms and came up with this slightly more menacing look on the face (or am i imagining it?)


Spoiler Alert - End

[dicor]

Originally Posted by sinner

Prompt : An Indian red Color 2023 2 door Mahindra Thar in Hotwheels packaging. image to be in Verticals 1080P dimension

Result: Attachment 2557100

Biggest pain point of Jeep Wrangler/ Rubicon owners.
AI is no exception. Wrangler is referred to as a Mahindra Thar.

Originally Posted by vredesbyrd

Bing Image Creator is a game-changer in terms of how close it gets to the prompts.
I am known in my circle for unhinged artwork, so here's some of that:

Would you mind sharing the prompt for that Ritz image, please?

Originally Posted by nimbleguru

If you folks are interested in having a open source, standalone & FREE desktop AI creator, try AUTOMATIC1111. For a advanced users, you can try ComfyUI.

You can download all sorts of AI models from huggingface or civitai websites.

Here's a guide to setup:
https://stable-diffusion-art.com/install-windows/

Thanks for sharing. Helpful indeed.

Here is my creation

Tool - Bing Image Creator
Prompt: Create a real-looking DSLR photo of a Black Maruti Suzuki Jimny fast-moving on narrow roads of the Himalayan region. The weather is windy and snow is falling. The time is around noon. The vehicle registration number is MHXXWEXXXX

Output

[vredesbyrd]

Originally Posted by dicor

Would you mind sharing the prompt for that Ritz image, please?

Bing is not displaying past creations for some reason. It keeps deleting them after 10-14 days.

Prompt was something along "Lord suzuki alto monster energy off road in snow and himalyas in the background"

[ValarMorghulis]

I haven't tried this but heard it in one of the clips from the Joe Rogan Show.

He said try prompting AI (Dall-e or MS Designer or Stable Diffusion) to show you a photo of a happy bunny.
Then keep asking it to make it "happier" still in each iteration (happier even more and so on).

What do you get as the final result after 7-8 iterations?

[Amrit@wheels]

Some quick fun I had today-

[DigitalOne]

Originally Posted by ValarMorghulis

What do you get as the final result after 7-8 iterations?

This was the result after 4 iterations