>>>Enter the prompt in the WebUI
Turn on the WebUI and enter the prompts and negative prompts you have written in the sections so far.
Prompt masterpiece, best quality, ultra detailed, fantasy, vivid color, 1girl, witch, black robe, hat, long silver hair,sitting, smile, looking at viewer, flower garden, blue sky, castle, noon, sunny Negative Prompt worst quality, low quality, normal quality, easynegative, |
Next, select the SDXL base model Animagine-xl-3.0.sagetensors and VAE Automatic as the model and VAE, respectively. If you want to use the SD 1.5 base, we recommend selecting blue-pencil-v10.sagetensors as the model and ClearVAE as the VAE.
Next, change the size of the image to be generated. The larger the image, the better the image quality, but the larger the VRAM capacity required for generation, and there is a risk that the layout of the image or the shape of the object will be blurred. Also, if the height is long, the result is likely to be the size of the person's entire body, and if the width is long, the result is likely to be the size of the upper body.
Generation tab → Width and Height initial state is 512x512, so adjust the values of Generation tab → Width and Height to match the model you are using.
If the base of the selected model is SD 1.5 or SD 2.1, it is recommended to start with a maximum of 512x512 pixels, and if it is SDXL, it is recommended to start with 1024x1024 pixels, but if the VRAM capacity is small, select a model of the SD 1.5 series and start with 256x256 pixels. It is recommended to combine the size of the image to be specified here with a power of 2 (such as 128, 256, 512) or a multiple of it.
>>> Create an image and check it.
Let's actually create an image. The goal is a picture of a witch sitting in a flower field with a castle in the distance, smiling at this person. Let's click the Generate button to create the image.
>>>Let's specify the number of images to be generated with batch count
First, let's find out what parts of the prompt need improvement. Since the results will be different even with the same prompt if the seed changes, let's try generating several times without changing the conditions. When you want to generate more than two images with the same prompt, use the Batch count function to generate an arbitrary number of images.
You can set the number of images to be generated by dragging the Batch count bar or directly entering a number. You can select up to 100 images, and it will take that much more time, so it is recommended to set it to 10 or less and look at the results. Since the purpose this time is to find out the influence of the prompt, do not generate in large quantities but set it to about 4 images.
Check the generated images to see if they were generated according to the prompt you specified. If there are any shortcomings, you can highlight the phrases to make them more similar to your thoughts.
>>>Highlighted Phrase
In the image above, if you want to print the flower garden with more emphasis, you can use a prompt called emphasis syntax. The emphasis syntax is a writing method that emphasizes the prompt, and it is expressed by enclosing (flower garden) in parentheses in the prompt. () is emphasized 1.1 times, (()) is emphasized 1.21 times, and the intensity increases as you increase the number of parentheses. And if you put a colon in the parentheses and specify a number, like (flower garden:1.4), you can emphasize it by that ratio.
(Prompt), | Enclose the prompt in parentheses for 1.1x emphasis |
((Prompt)), | Increasing the parentheses can increase the intensity |
(Prompt:1.4), | Enclosing in parentheses and adding a multiplier can increase the emphasis by that multiplier |
(Prompt_A Prompt_B), | Emphasis the entire range within the parentheses. However, it must be separated by commas |
(Prompt_A) Prompt_B, | Enclosing a portion in parentheses for emphasis |
Let's use this feature to change Prompt flower garden to (flower garden) to emphasize the flower garden. Let's skip other images that were created through trial and error for now, and change Prompt sitting to sitting on flower garden to give more specific instructions and specify the composition of the witch sitting in the flower garden. Also, add Prompt boots, which is necessary to create a witch wearing boots and a long robe, and modify Prompt black robe to long black robe by adding long. And to prevent short robes from being created, add Prompt short robe to the negative prompt.
Prompt masterpiece, best quality, ultra detailed, fantasy, vivid color, 1girl, witch, long black robe, boots, hat, long silver hair, sitting on flower garden, smile, looking at viewer, (flower garden), blue sky, castle, noon, sunny, Negative Prompt worst quality, low quality, normal quality, easynegative, short robe |
The flower field is emphasized more than before, but she is also wearing a short robe instead of a long one, and she is also wearing boots. Let's continue adding prompts. To express a kneeling posture with her knees pressed against the floor, add Prompt holding knees. And to make the long robe more appropriately created, emphasize Prompt long black robe as (long) black robe.
Prompt masterpiece, best quality, ultra detailed, fantasy, vivid color, 1girl, witch, (long) black robe, hat, boots, long silver hair, sitting on flower garden, holding knees, smile, looking at viewer, (flower garden), blue sky, castle, noon, sunny, Negative Prompt worst quality, low quality, normal quality, easynegative, short robe |
If you like the image, save it as a PNG. If you click the PNG info button or tab, you can view the model name and parameters used when creating the image with AUTOMATIC1111. In the next section, I will explain how to improve the image quality using that information.
