Thank you for using Geek Squirrel Grayscale 0.1 version.

Trigger word: GeekS

Base model: Basic algorithm 1.5 (It can also be used with other 1.5 versions of the base model, but if the base model tends to generate colored images, you can manually increase the weight of prompt words like “monochrome” or “grayscale.” Recommended base models include Perfect World and other 2.5D CG rendering styles. Models that are too unusual, inclined towards painting or anime, may perform worse.)

When using controllnet, you can use only depth, but the image will undergo more changes. Combining it with line drawing can better constrain the content, and you can freely test with various constraints. The conditions shown in the video may not necessarily be the optimal configuration.

The base image is best similar to white mold photos or images generated in a similar style using other methods.

If using line art as the base image, it is best to first convert the line art into a 3D effect using 3D style lora or other base models. Generally, the depth map generated directly from line art will have issues (so during the line art to 3D conversion process, do not use depth maps in controllnet). PS: If the base image is line art, it can be directly placed as a line map in controllnet.

Prompt words: When using controllnet for image generation, you can appropriately write some prompt words related to the content, which may significantly improve the result.

liblib does not have a depth plugin, but you can generate it using Photoshop and then upload it. For specific methods, refer to my Bilibili video.

If you need to download this lora, you can follow me on Bilibili and private message “lora” to get it.

https://space.bilibili.com/49882074?spm_id_from=333.1296.0.0