Try 4.38.0

suggested_hardware: zero-a10g

Allocation is important again

No more quota

Update and block dependencies

Image format for result

Readd heic format support

open-clip-torch==2.24.0

Rollback

HEIF

Link, prompt, time & stat

torch.cuda.empty_cache()

No interactive outputs

with torch.no_grad():

No more used

More than double

Do not use LLAVA_CLIP_PATH

Delete gradio_demo_tiled.py

Adjust stats

Delete gradio_demo_face.py

4.37.2

Due to Gradio issues, the generated image is slightly less satured than the original

Square root steps

Set max seed

Cube root

print("Estimated minutes: " + f'{((result_width * result_height * input_width * input_height * edm_steps * (num_sampl

Fix downscale

Allow gif

Square root

Log statistics

Reset button

Rollback

Reset button

log_information

Fix moving code

Select model without GPU

Image.fromarray(results[0]).save(unique_name)

Stable Diffusion 3 Medium

Fix by rollback

denoise_image = upscale_image(denoise_image, upscale, unit_resolution=32, min_size=min_size)

Code out of GPU compute

Remove Llava

if type(e).__name__ == 'gradio.exceptions.Error':

"The image" + (" has" if len(results) == 1 else "s have") + " been generated in " + \

print('Exception of type ' + str(type(e)))

Adapt text

GitHub stars

fp32

Useless requirement

As input by default