stable-diffusion
stable-diffusion copied to clipboard
img2img blurred images with low strength
Hello. Could you maybe point me to the direction, why am i getting too blurred images with the --strength=0.5 setting even with 100-150 steps. Is there any way to improve detalisation or it is the limitations of this method?
for example, with --strength=0.5:
--strength=0.75:

it's kind of hard to produce the same results without knowing the exact command you ran, include it so we can test and see if it does the same thing in our machine as well
not knowing what you ran all i can say is that it doesn't look blurry on my end, here is an example
source

python optimizedSD/optimized_img2img.py --prompt "a man smiling and showing his teeth" --init-img ./chad.png --H 512 --W 512 --seed 1 --n_iter 1 --n_samples 1 --ddim_steps 50 --skip_grid --turbo --precision full
--strength=0.75

--strength=0.5

--strength=0.25

Well, what a weird thing. Updated all files, if i run it on my machine - i got a result that you have. But if i run it through my tg bot - it generates a different image even with same seed. Weird thing but at least i know where the problem is, thank you
its like some args wont pass correctly..
i'm happy you figured it out, sorry if i'm late, i somehow didn't get notified about your reply
i'm happy you figured it out, sorry if i'm late, i somehow didn't get notified about your reply
weell.. kinda
i've manage to figure that most of the problems were because of the input image was 90x90 instead of 512x512
but still with the same seed there is slight differences
couldn't understand fully what can telegram bot do different in this case
fine enough tho =)

i see, at least it's good you found the issue behind the blurry images, keep at it, you'll eventually get there