stabilityai

/

sdxl-turbo

like 2.08k

You can add ,width=1024, height=1024 like that :
image = pipe(prompt=prompt, negative_prompt=negative_prompt, num_inference_steps=4, guidance_scale=0.0, width=1024, height=1024).images[0]

It work pretty good for me.

where do you add that, sorry, pls give step-by-step, thx.

I just used the python code in this model card here : https://huggingface.co/stabilityai/sdxl-turbo#diffusers, you can replace the last line of code in the "Text-to-image" code.

Whit more generations, i saw that it depend on what you generate, but the model may be better in 512 x 512 for some cases.

this lloks not good enough

I have trid some examples, and i find that in 1024x1024, it always draw two person when i just input "an amine boy in the room".