Random tests on low VRAM

[mkondel]

I used the same parameters and was able to get identical images from both forks. InvokeAI seems faster. The UI on AUTOMATIC1111 is killer tho. I tried frankensteining them together, but that doesn't just work.

GPU: 1050Ti
VRAM: 4GB
Model: sd-v1-4.ckpt
Prompt: fibonacci goddess
Sampling: LMS or KLMS
CFG scale: 7.5
Size: 512x512
Seed: 270890088

InvokeAI

time	s/it
1:10	1.42
1:10	1.41
1:10	1.42

AUTOMATIC1111

arguments	time	s/it
--opt-split-attention	1:24	1.67
--opt-split-attention-v1	1:38	1.97
--lowvram	2:28	2.97
--lowvram --opt-split-attention	2:31	3.03
--lowvram --opt-split-attention-v1	2:44	3.29

fibonacci goddess