Quick tour - New
Here you define a new txt2img command and then hit the Generate! button to place it on the queue. A1111 will process the commands on the queue and when the images are generated, you'll find them back at the ready page.
When you have generated a command you can edit it and press generate again, to create a modified version. Use the Clear button to fill this form with the project defaults.
Let's go through all values:
Width x Height : The dimensions of the image you want to generate. You can trade of resolution for speed and file size, choose dimensions that fit with the sort of image you generate and so on.
Sampling method: Experiment with various sampling methods.
Sampling steps: (range 1..150). The more steps the longer the generation takes but also the more time the AI had to work on it, thus the better the results.
CFG Scale: (range 1..300). The CFG scale determines how strongly the images should confirm to the prompt. 70 is a good starting point. With rather high values the image tends to confirm very well to the prompt but quality goes down. At different models different CFG scales give the best results.
This is the scale A1111 uses multiplied by 10. It is just not handy to work with decimals.
Denoising: (range 0 to 1000) This is not used in a TXT2IMG but can already be set, the default denoising for when you later on start IMG2IMG commands based on the output of this command.
On a IMG2IMG command denoising determines how much to leave the original image in tact. At 0 you will get the same image you started with, at 1000 a completely unrelated image.
This is the scale A1111 uses multiplied by 1000.
It is just not handy to work with decimals.
Batch count: How many times you want to run this command. This determines together with Batch Size the number of generated images.
Batch size: How many images you want to generate per run.
Resize mode: This applies to IMG2IMG commands when the target size is different than the original size.
Restore faces: Do an extra effort to generate faces better.
Tiling: Make tileable images.
Seed: Leave this at -1 to generate a new random seed each run. You can also specify your own seed, to reduce randomness.
Prompt: Your prompt. Here you tell what image you want. Good prompts generate awesome images, prompting is the essence of the craft of image generation.
Negative prompt: Things you don't want in your image. Often I read the advice to start with a very limited or even no negative prompt. I tend to put "Text, watermark" to somewhat avoid that silly texts.
Experiment with prompts and other parameters! That is a big part of the fun.
Above prompts you see this buttons:
You can select a piece of text and then use the (..) or [...] to put that text between brackets, either (emphasize) or [de-empasize) that part of the text. (((Or use multiple brackets to even stronger emphasize))) or [[[de-empasize]]].
Select a position in your text, and then hit the Segments button. This opens a dialog in which you can maintain a collection of segments, pieces of text you commonly use. Just pick one, and it will be inserted.
For the prompt and negative prompt there are 2 different segment lists.
When you have a lot of segments, use the sort button and the filter to easily find them. Also tag segments, e.g. segments that describe hair but don't have the word hair in the text, could have the tag hair. When you specify a filter it is a filter over texts and tags. Thus a filter on "hair" gives all segments with the word hair in either the text or tag.
The button Recent gives a similar list of lines, but then all lines used recently. Thus when you add ta text to several prompts, typing it once will be enough.
Import
Besides a new TXT2IMG you could also start a new Import command. That is a simple import of images from your local disks into the ZTX37 database. You can drag images to the import and those will be imported.
After import one or more commands will appear in the ready tab. From there you can process the import as if it were images generated by Txt2Img.