Project 8: Stable Diffusion 4
Re: Project 8: Stable Diffusion 4
This week’s assignment was an exercise to find pattern in clusters. As explained in my last week’s assignment, these last few exercises are geared towards my final project for the quarter with MAT255. I am juxtaposing multiple photos of different subjects and objects into a collage to find thematic harmony within dialectical/contradictory relationship of elements within the images. I have used clusters people against, books and other objects to see if they form a cohesive thematic relationship within each other. Below are the examples.
Calcutta Street:
Prompt 5:
desaturated dramatic black and white photograph of a busy, overcrowded, polluted city street in contemporary Calcutta in India during daytime, top-angle, top-shot, helicopter view, Kodak TriX 400 ISO film, high-contrast Negative prompt: watermark, ugly, deformed, ugly, mutilated, disfigured, text, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 150, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Refiner: sd_xl_refiner_1.0 [7440042bbd], Refiner switch at: 0.6, Version: v1.6.0
Saved: 00561-150.png
Prompt 10:
desaturated dramatic black and white photograph of a busy, overcrowded city street in contemporary Calcutta in India during daytime, top-angle, top-shot, helicopter view, Kodak TriX 400 ISO film, high-contrast Negative prompt: watermark, ugly, deformed, ugly, mutilated, disfigured, text, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 50, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Refiner: sd_xl_refiner_1.0 [7440042bbd], Refiner switch at: 0.6, Version: v1.6.0
Saved: 00608-50.png
Prompt 20:
desaturated dramatic black and white photograph of a busy, overcrowded city street in contemporary Calcutta in India during daytime, top-angle, top-shot, helicopter view, Kodak TriX 400 ISO film, high-contrast Negative prompt: watermark, ugly, deformed, ugly, mutilated, disfigured, text, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 250, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Refiner: sd_xl_refiner_1.0 [7440042bbd], Refiner switch at: 0.6, Version: v1.6.0
Saved: 00688-250.png
Book shelf:
Prompt 22:
desaturated dramatic black and white photograph of tall bookshelves in a library full of old and dusty books, extreme low-angle shot, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, strong beam of light, dust particles flying around, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sky, clouds, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 250, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Version: v1.6.0
Saved: 00714-250.png
Prompt 23:
desaturated dramatic black and white photograph of tall towering bookshelves in a library full of old and dusty books, extreme low-angle shot, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, strong beam of light, dust particles flying around, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sky, clouds, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 250, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Version: v1.6.0
Saved: 00719-250.png
Factory Chimney:
Prompt 29:
desaturated dramatic black and white photograph of an smokey chimney's of old factories silhouetted against the skyline, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, smoke clouds, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 200, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Version: v1.6.0
Saved: 00894-200.png
Prompt 30:
desaturated dramatic black and white photograph of an smokey chimney's of old factories silhouetted against the skyline, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, smoke clouds, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 150, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00899-150.png
Covalent Bonds:
Prompt 31:
desaturated dramatic black and white photograph of a model of chemical covalent bonds against a black background, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, clouds, sky, human beings, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 4248109, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00909-4248109.png
Prompt 32:
desaturated dramatic black and white photograph of chemical covalent bonds against a black background, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, clouds, sky, human beings, faces, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 2725146285, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00919-2725146285.png
Prompt 36:
desaturated dramatic black and white photograph of chemical covalent bonds against a black background, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, clouds, sky, human beings, faces, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 300, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00949-300.png
Final montage/collage composition:
This exercised was focused mainly on how to make the prompt as precise as possible to determine the exactness of outcome. I struggled quite a bit with the negative prompts this time. Even though, I had used “ugly, deformed, mutilated, disfigured, text, extra limbs, face cut, head cut” etc in my negative prompt in the end, the results displayed multiple instances of disfigured human subjects. Although, some unwanted elements were omitted successfully through the use of negative prompt. For instance, as soon as I removed the word “polluted” from my prompt, the rubbish piled up along various corners of the streets disappeared.
I also figured out why there were digital Glitch-like patterns present all over my last assignment. The glitches were due to low Denoising strength value. The denoising strength is responsible for finer rendition of the picture as well as accurately following the prompt. The lower it is the closer the final image will be to the given prompt. But a lower denoising vale (I had used as low as 0.25) runs the risk of compromising the resolution and quality of the final image, which in my case, registered as glitch-like pattern all over the images produced. The default value is 0.7. But I have noticed, unless the prompt is extremely detailed and covers every minute aspect of the expected image, such a high denoising value will eventually stray far away from the text prompt, albeit while rendering crisp images. The denoising value must be then tried and tested for each prompt in order to determine the perfect middle ground, where the images are of a considerably high quality and it sticks to the prompt as closely as possible.
Calcutta Street:
Prompt 5:
desaturated dramatic black and white photograph of a busy, overcrowded, polluted city street in contemporary Calcutta in India during daytime, top-angle, top-shot, helicopter view, Kodak TriX 400 ISO film, high-contrast Negative prompt: watermark, ugly, deformed, ugly, mutilated, disfigured, text, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 150, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Refiner: sd_xl_refiner_1.0 [7440042bbd], Refiner switch at: 0.6, Version: v1.6.0
Saved: 00561-150.png
Prompt 10:
desaturated dramatic black and white photograph of a busy, overcrowded city street in contemporary Calcutta in India during daytime, top-angle, top-shot, helicopter view, Kodak TriX 400 ISO film, high-contrast Negative prompt: watermark, ugly, deformed, ugly, mutilated, disfigured, text, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 11, Seed: 50, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Refiner: sd_xl_refiner_1.0 [7440042bbd], Refiner switch at: 0.6, Version: v1.6.0
Saved: 00608-50.png
Prompt 20:
desaturated dramatic black and white photograph of a busy, overcrowded city street in contemporary Calcutta in India during daytime, top-angle, top-shot, helicopter view, Kodak TriX 400 ISO film, high-contrast Negative prompt: watermark, ugly, deformed, ugly, mutilated, disfigured, text, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 250, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Refiner: sd_xl_refiner_1.0 [7440042bbd], Refiner switch at: 0.6, Version: v1.6.0
Saved: 00688-250.png
Book shelf:
Prompt 22:
desaturated dramatic black and white photograph of tall bookshelves in a library full of old and dusty books, extreme low-angle shot, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, strong beam of light, dust particles flying around, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sky, clouds, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 250, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Version: v1.6.0
Saved: 00714-250.png
Prompt 23:
desaturated dramatic black and white photograph of tall towering bookshelves in a library full of old and dusty books, extreme low-angle shot, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, strong beam of light, dust particles flying around, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sky, clouds, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 250, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Version: v1.6.0
Saved: 00719-250.png
Factory Chimney:
Prompt 29:
desaturated dramatic black and white photograph of an smokey chimney's of old factories silhouetted against the skyline, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, smoke clouds, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 200, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Denoising strength: 0.52, Hires upscale: 1.45, Hires upscaler: Latent, Version: v1.6.0
Saved: 00894-200.png
Prompt 30:
desaturated dramatic black and white photograph of an smokey chimney's of old factories silhouetted against the skyline, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, smoke clouds, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 150, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00899-150.png
Covalent Bonds:
Prompt 31:
desaturated dramatic black and white photograph of a model of chemical covalent bonds against a black background, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, clouds, sky, human beings, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 4248109, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00909-4248109.png
Prompt 32:
desaturated dramatic black and white photograph of chemical covalent bonds against a black background, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, clouds, sky, human beings, faces, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 2725146285, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00919-2725146285.png
Prompt 36:
desaturated dramatic black and white photograph of chemical covalent bonds against a black background, Kodak TriX 400 ISO film, high-contrast , dark, Chiaroscuro lighting style, grainy, push-processing Negative prompt: watermark, ugly, deformed, glitchy, sea, ocean, sun, moon, stars, clouds, sky, human beings, faces, mutilated, disfigured, extra limbs, face cut, head cut, extra fingers, extra arms, poorly drawn face, mutation, bad proportions, cropped head, malformed limbs, mutated hands, fused fingers, long neck, light-source Steps: 60, Sampler: DPM++ 2M Karras, CFG scale: 20, Seed: 300, Size: 640x480, Model hash: 7440042bbd, Model: sd_xl_refiner_1.0, Version: v1.6.0
Saved: 00949-300.png
Final montage/collage composition:
This exercised was focused mainly on how to make the prompt as precise as possible to determine the exactness of outcome. I struggled quite a bit with the negative prompts this time. Even though, I had used “ugly, deformed, mutilated, disfigured, text, extra limbs, face cut, head cut” etc in my negative prompt in the end, the results displayed multiple instances of disfigured human subjects. Although, some unwanted elements were omitted successfully through the use of negative prompt. For instance, as soon as I removed the word “polluted” from my prompt, the rubbish piled up along various corners of the streets disappeared.
I also figured out why there were digital Glitch-like patterns present all over my last assignment. The glitches were due to low Denoising strength value. The denoising strength is responsible for finer rendition of the picture as well as accurately following the prompt. The lower it is the closer the final image will be to the given prompt. But a lower denoising vale (I had used as low as 0.25) runs the risk of compromising the resolution and quality of the final image, which in my case, registered as glitch-like pattern all over the images produced. The default value is 0.7. But I have noticed, unless the prompt is extremely detailed and covers every minute aspect of the expected image, such a high denoising value will eventually stray far away from the text prompt, albeit while rendering crisp images. The denoising value must be then tried and tested for each prompt in order to determine the perfect middle ground, where the images are of a considerably high quality and it sticks to the prompt as closely as possible.