HOME | DD

Description

This was picture was made with Photobashing together outputs from Stable Diffusion text2img, img2img with masking and corrective work done with Photoshop.

Repository being utilized is Automatic1111 Stable-Diffusion-webui; all relevant technology and credits can be found from the github page.

The total time spent on doing this was about 6 hours. The workflow involved first using Stable Diffusions text2img to generate a good base image, about 100 generations of which 1 was chosen as the seed, prompt and settings configuration. After choosing the generation setting and space I proceeded to do tuning and exploring the surrounding space. This was done by fine-tuning of the seed variations and adjusting the parameters with fixed seed; about 200 images were generated. After finding a suitable output this was taken to Photoshop to have slight issues corrected and adjusting the colours. After this the correct version was run trough img2img to add further refinement of the overall composition; about 100 fine adjustments to allow the corrections done with Photoshop to blend in to the image; this was also when the subjects hair colour was adjusted from unclear brown-yellow to blond by prompting the img2img system to focus on the hair. After good output was received it was run through once more the img2img, this time to adjust overall style using low noise and low scale settings. The best base image was then taken to Photoshop in which the Diaper was bashed in to the picture and 3 different masks were painted for the img2img in-painting process. This is when the bulk of the work started, where in every generation was taken to back to Photoshop to be adjusted and corrected along with the masks, additional bashing done. This took more generations and adjustments than I cared to keep track of however closer to 50 iterations of this workflow from photoshop to img2img and img2img inpainting was used.

In constantly iterating manner and adjusting of the mask only the cartoon characters on the diaper the tapes, and the fold above the left side tape remained untouched by the img2img system.

Description

This was picture was made with Photobashing together outputs from Stable Diffusion text2img, img2img with masking and corrective work done with Photoshop.

Repository being utilized is Automatic1111 Stable-Diffusion-webui; all relevant technology and credits can be found from the github page.

The total time spent on doing this was about 6 hours. The workflow involved first using Stable Diffusions text2img to generate a good base image, about 100 generations of which 1 was chosen as the seed, prompt and settings configuration. After choosing the generation setting and space I proceeded to do tuning and exploring the surrounding space. This was done by fine-tuning of the seed variations and adjusting the parameters with fixed seed; about 200 images were generated. After finding a suitable output this was taken to Photoshop to have slight issues corrected and adjusting the colours. After this the correct version was run trough img2img to add further refinement of the overall composition; about 100 fine adjustments to allow the corrections done with Photoshop to blend in to the image; this was also when the subjects hair colour was adjusted from unclear brown-yellow to blond by prompting the img2img system to focus on the hair. After good output was received it was run through once more the img2img, this time to adjust overall style using low noise and low scale settings. The best base image was then taken to Photoshop in which the Diaper was bashed in to the picture and 3 different masks were painted for the img2img in-painting process. This is when the bulk of the work started, where in every generation was taken to back to Photoshop to be adjusted and corrected along with the masks, additional bashing done. This took more generations and adjustments than I cared to keep track of however closer to 50 iterations of this workflow from photoshop to img2img and img2img inpainting was used.

In constantly iterating manner and adjusting of the mask only the cartoon characters on the diaper the tapes, and the fold above the left side tape remained untouched by the img2img system.