StableDiffusion experiments with empty prompts

Starting settings and a base for comparison: empty prompts (both positive and negative), seeds from 0 to 5, steps: 30, sampler: DDIM, CFG scale: 10, size: 512x512, model: sd-v1-4.ckpt 7460a6fa

Steps will increase details, sharp image and make colors more vivid, but (since there are no prompts) it doesn't make images easier to recognize.
Generations with 10 steps VS 100 steps
Steps: 10, Sampler: DDIM, CFG scale: 10

Steps: 100, Sampler: DDIM, CFG scale: 10

Steps in between:
Steps 10-100 plot, Sampler: DDIM, CFG scale: 10

CFG scale almost doesn't make any difference with empty prompts
I put steps back to 30, and make CFG=5 and CFG=25 in addition to default 10
Steps: 30, Sampler: DDIM, CFG scale: 5

Steps: 30, Sampler: DDIM, CFG scale: 25

Changing sampling method from original DDIM to others, sometimes just changes little details……
Steps: 30, Sampler: Heun, CFG scale: 10

Steps: 30, Sampler: DPM2, CFG scale: 10

Steps: 30, Sampler: PLMS, CFG scale: 10

……but sometimes produce totally different results

Steps: 30, Sampler: DPM fast, CFG scale: 10

Steps: 30, Sampler: DPM2 a Karras, CFG scale: 10

Steps: 30, Sampler: Euler a, CFG scale: 10

And changing models, of course, also changing outputs, but not as drastically as I expect
Once again, here is model sd-v1-4.ckpt 7460a6fa
SD1.4, Steps: 30, Sampler: DDIM, CFG scale: 10

v1-5-pruned-emaonly.ckpt 81761151
SD1.5, Steps: 30, Sampler: DDIM, CFG scale: 10

sd-wikiart-v2.ckpt 8f32b8df
Wiki-art, Steps: 30, Sampler: DDIM, CFG scale: 10

wd-v1-3-float16.ckpt 84692140
WD1.3, Steps: 30, Sampler: DDIM, CFG scale: 10

That's all for now

PS: 30 more outputs for better representation of what you could find in empty prompt outputs
seeds 6-36

StableDiffusion experiments with empty prompts

Warning