Highlight a word in the prompt and press ctrl and either up or down to easily increase or decrease the weight of a keyword.
Check out the xyz plot script. This allows you to change certain parameters while generating a batch. You can increment the cfg for example to see how it effects the image. Or denoising strength, model, or anything really.
Merge words like this [banana | apples].
Use the above technique to create a unique and consistent look using celebrities. [britney spears | megan fox | ann coulter]
Use the PNG info tab to retrieve the parameters used to create an image.
You can use multiple instances of controlnet, choose how many instances in the settings.
Search for "Quicksettings list" in settings. Add sd_vae and CLIP_stop_at_last_layers seperated by a comma. This adds two new fields to the webui. sd_vae allows you to easily switch out vaes. CLIP_stop_at_last_layers adds the clip skip input, which allows you to stop the clip model earlier than usual. This changes the image in various way.
The Civitai Helper extension can scan your checkpoints, loras, etc and find them on Civitai, then pull meta data and a sample image. Helps keep track when you have many models. It can also look for updates for models you have, or download a model for you (although I prefer downloading outside of the tool and then scanning).
There's another called Model Preset Manager, which looks like it can do similar and more, but I haven't tried it yet.
Some additions to one of yours:
Use the PNG info tab to retrieve the parameters used to create an image.
If an image wasn't generated, or had its meta data stripped, you can pop the image into the img2img tab and click one of the "Interrogate" buttons next to Generate. The first time will be slow, because it has to download some models, but after that it will be quick.
Both buttons analyze the image and populate the positive prompt with terms and phrases that fit the content and style of the image. CLIP uses more natural language to describe the image, while DeepBooru produces comma separated tags. I find the latter more useful. Great for getting prompt ideas.
You can also use this on images you generated, which had "happy little accidents" that you liked, to get ideas for terms to add to the prompt to get more like it.
There shouldn't be anything special you need to do. Just make sure an image is loaded in the img2img tab (and img2img subtab). None of the settings matter for this.
Just be patient the first time. Watch the terminal for signs of progress or errors.
Try the DeepBooru one first, it's much smaller and should download faster.
Install the sd-dynamic-prompts extension from the addons tab. This lets you use wildcards. Wildcards are random inputs. For example you can do "__race__ person doing __verb__with a __environment__ background". Download wildcards from civai and put it in the wildcard folder located in the sd-dynamic-prompts folder located in extensions. Or easily create your own. Can even use chatgpt for easy wildcard ideas.
[ + ] GloryBeckons
[ - ] GloryBeckons 1 point 1.9 yearsMay 27, 2023 21:59:52 ago (+1/-0)
The Civitai Helper extension can scan your checkpoints, loras, etc and find them on Civitai, then pull meta data and a sample image. Helps keep track when you have many models. It can also look for updates for models you have, or download a model for you (although I prefer downloading outside of the tool and then scanning).
There's another called Model Preset Manager, which looks like it can do similar and more, but I haven't tried it yet.
Some additions to one of yours:
If an image wasn't generated, or had its meta data stripped, you can pop the image into the img2img tab and click one of the "Interrogate" buttons next to Generate. The first time will be slow, because it has to download some models, but after that it will be quick.
Both buttons analyze the image and populate the positive prompt with terms and phrases that fit the content and style of the image. CLIP uses more natural language to describe the image, while DeepBooru produces comma separated tags. I find the latter more useful. Great for getting prompt ideas.
You can also use this on images you generated, which had "happy little accidents" that you liked, to get ideas for terms to add to the prompt to get more like it.
[ + ] Niggly_Puff
[ - ] Niggly_Puff [op] 0 points 1.9 yearsMay 27, 2023 22:18:50 ago (+0/-0)
[ + ] GloryBeckons
[ - ] GloryBeckons 0 points 1.9 yearsMay 28, 2023 01:24:42 ago (+0/-0)
Just be patient the first time. Watch the terminal for signs of progress or errors.
Try the DeepBooru one first, it's much smaller and should download faster.
[ + ] Niggly_Puff
[ - ] Niggly_Puff [op] 1 point 1.9 yearsMay 27, 2023 19:07:30 ago (+1/-0)
[ + ] Niggly_Puff
[ - ] Niggly_Puff [op] 0 points 1.9 yearsMay 27, 2023 20:48:32 ago (+0/-0)