Create Ai images in bulk?

Alf85

Elite Member
Joined
Sep 7, 2021
Messages
2,586
Reaction score
922
I am wanting to create Ai images in bulk. My idea is to give it a prompt like "rooms decorated with red stuff" and I can generate some images. I'm testing with Koala, but it's quite limited. Any idea?
 
You might have to explore DALL-E, which is designed to generate images from text. DALL-E tools can offer more diverse image generation compared to simpler models like Koala.

Thanks!
 
1. your prompt is limited in instruction and that will limit the amount and kind of images you can generate. Use better prompt
2. If you are using a free service, you will also be limited. Use paid service
 
Thanks for the advice I will check it out. Of course the prompt in the post was just an example....

Do the images generated with Ai from Bing, Canva and Koala have any watermarks that a human can't see but a bot can? I also think they can be used on Pinterest.
 
Do the images generated with Ai from Bing, Canva and Koala have any watermarks that a human can't see but a bot can? I also think they can be used on Pinterest.
Absolutely NOT.

Any of these mentioned above AI Tools generate picture without any sort of watermark whatsoever.

Thanks!
 
Ask ChatGPT to write a custom Python script for that. It takes about 30 seconds to get what you need.
 
Try Midjourney and DALL-E Bulk Image Generator. Just that Midjourney can only generate four different images based on a prompt.
 
Try Midjourney and DALL-E Bulk Image Generator. Just that Midjourney can only generate four different images based on a prompt.

There are many scripts that allow you to generate images via Midjourney in bulk and even insert them into imported or AI-generated articles automatically. No manual input is required anymore. Auto-blogging has never been that easy...
 
Thanks for the advice I will check it out. Of course the prompt in the post was just an example....

Do the images generated with Ai from Bing, Canva and Koala have any watermarks that a human can't see but a bot can? I also think they can be used on Pinterest.

Some of AI tools include invisible watermark. For example, there is a watermark in stable diffusion XL.
 
someone to share prompt so that they come out very realistic
 
Some of AI tools include invisible watermark. For example, there is a watermark in stable diffusion XL.
Any tool to detect the watermark?
 
Today, I started testing Foocus 2.3.1 with this configuration:



and this hardware:

AMD Ryzen 5 3600
16 Gb Ram
1 TB SSD
AMD Radeon RX5700 8GB

An image takes about 2.3 minutes.... I think the problem is the video card. I don't think I can improve the time anymore.

Any way to set the size to 1000x1500 or 9:16 for a Pinterest pin?
 
Any tool to detect the watermark?
Well, yeah. I mean they wouldn't watermark it if it were undetectable. Most the sites also embed EXIF data but that's easy enough to remove. When in doubt there's always a watermark in anything you do so law enforcement can come knocking, like you want to know why your color printer refuses to print black and white when it's out of color ink? Your printer puts a watermark of dots that correspond to the printer serial number on the paper using yellow ink. So if you need to be tracked down, there's a good chance they can do it by finding out who bought the printer.

If you don't want a watermark at all then run SDXL (upgrade the GPU, as 8GB is only reliably going to do 512x512 [above that you'll get weird artifacts, if it works at all]; it's not expensive either as a Tesla M40 with some fans zip tied to the thing already gives you SDXL) locally on your computer. Then you can just not enable the watermark, and this is better anyway as you get a lot more options vs these websites. You can run any model from hugging face or civitai and get FAR better results if you put in a few hours of work to get everything set up.
Any way to set the size to 1000x1500 or 9:16 for a Pinterest pin?
Not with that GPU, no. And unless you want to spend big bucks on a GPU don't expect your times to get much better. The cheap M40 route with zip tied fans (or steal the cooler from a GTX 770) lets you make reasonably sized images with much better quality (8GB always felt to me like it has this "anime" effect, skin is far too smooth and unrealistic), but you're still not going to speed much up. SDXL takes about 9-ish minutes on an M40 but it's also only $80 bucks compared to, what like thousands still? for a 4090. Both can make the same end result with 24GB, the M40 just takes longer.
 
Well, yeah. I mean they wouldn't watermark it if it were undetectable. Most the sites also embed EXIF data but that's easy enough to remove. When in doubt there's always a watermark in anything you do so law enforcement can come knocking, like you want to know why your color printer refuses to print black and white when it's out of color ink? Your printer puts a watermark of dots that correspond to the printer serial number on the paper using yellow ink. So if you need to be tracked down, there's a good chance they can do it by finding out who bought the printer.

If you don't want a watermark at all then run SDXL (upgrade the GPU, as 8GB is only reliably going to do 512x512 [above that you'll get weird artifacts, if it works at all]; it's not expensive either as a Tesla M40 with some fans zip tied to the thing already gives you SDXL) locally on your computer. Then you can just not enable the watermark, and this is better anyway as you get a lot more options vs these websites. You can run any model from hugging face or civitai and get FAR better results if you put in a few hours of work to get everything set up.

Not with that GPU, no. And unless you want to spend big bucks on a GPU don't expect your times to get much better. The cheap M40 route with zip tied fans (or steal the cooler from a GTX 770) lets you make reasonably sized images with much better quality (8GB always felt to me like it has this "anime" effect, skin is far too smooth and unrealistic), but you're still not going to speed much up. SDXL takes about 9-ish minutes on an M40 but it's also only $80 bucks compared to, what like thousands still? for a 4090. Both can make the same end result with 24GB, the M40 just takes longer.
I will continue testing with this GPU for now. I don't have the money to upgrade the GPU at the moment, but in the future it might be possible. I understand that the best for this kind of projects is to use Nvidia.
 
Update:

I tried several Lora models for what I'm looking for (gardening), and they don't quite convince me.

Is there any way to train the Ai with real gardening images and use it?
 
someone to share prompt so that they come out very realistic

There is a hint for you. Your AI will analyze your content... to create a realistic prompt for the text you create. No need to do it manually anymore, because everything works fully automatically - AI creates text, AI analyzes it, AI creates prompts for images, AI creates the images, AI publishes the articles...
 
Back
Top