* Simplified the UI, using JS, removing the need for image select input
Signed-off-by: Alex Volkov <alex.volkov@fundbox.com>
* Merging incoming changes before PR
Signed-off-by: Alex Volkov <alex.volkov@fundbox.com>
* Removing pychache
Signed-off-by: Alex Volkov <alex.volkov@fundbox.com>
* adding gitignore
Signed-off-by: Alex Volkov <alex.volkov@fundbox.com>
Signed-off-by: Alex Volkov <alex.volkov@fundbox.com>
Fixed
At most should produce 2 samples
>original & gfpgan
>original & esrgan
>original & gfpgan -> esrgan
Turn off save individual samples to save only gfpgan, esrgan or gfpgan -> esrgan
The previous color magenta (255, 0, 255) distracts from the actual
images. Use white instead.
The text already has black stroke, so it remains readable even on white
backgrounds.
* Fix minor things from last big UI restyle
These two changes were applied to txt2img tab but not img2img tab:
Sample select changed from Radio to Dropdown,
Seed Input restricted to single line, which also removes the resize handlebar
Added a small CSS Workaround for a gradio bug in firefox: The Seed output box is a disabled number field, in chromium gradio will hide the up/down arrows but on firefox it leaves a weird grey square, the CSS rule included in this commit fixes that issue.
* remove full_width from where it doesn't belong
Co-authored-by: hlky <106811348+hlky@users.noreply.github.com>
The negative filename reported in #112 was caused by the old (now
replaced) listdir-based sequence number calculation, that incorrectly
subtracted 1 from the number of files already in the directory:
base_count = len([x for x in os.listdir(sample_path_i) if x.endswith(('.png', '.jpg'))]) - 1 # start at 0
The `- 1` was probably copied from the code for the base directory,
which had to subtract 1 to account for the `samples/` subdirectory (that
was before the listdir code started checking file extensions).
The get_next_sequence_number could never return negative numbers, since
it returns `result + 1`, meaning the sequence starts at 0 as intended.
Commit 82770bacae broke regular prompt
matrix, causing the following error:
Traceback (most recent call last):
File "/.../sd/miniconda3/envs/ldx/lib/python3.8/site-packages/gradio/routes.py", line 247, in run_predict
output = await app.blocks.process_api(
File "/.../sd/miniconda3/envs/ldx/lib/python3.8/site-packages/gradio/blocks.py", line 641, in process_api
predictions, duration = await self.call_function(fn_index, processed_input)
File "/.../sd/miniconda3/envs/ldx/lib/python3.8/site-packages/gradio/blocks.py", line 556, in call_function
prediction = await anyio.to_thread.run_sync(
File "/.../sd/miniconda3/envs/ldx/lib/python3.8/site-packages/anyio/to_thread.py", line 31, in run_sync
return await get_asynclib().run_sync_in_worker_thread(
File "/.../sd/miniconda3/envs/ldx/lib/python3.8/site-packages/anyio/_backends/_asyncio.py", line 937, in run_sync_in_worker_thread
return await future
File "/.../sd/miniconda3/envs/ldx/lib/python3.8/site-packages/anyio/_backends/_asyncio.py", line 867, in run
result = context.run(func, *args)
File "scripts/webui.py", line 933, in txt2img
output_images, seed, info, stats = process_images(
File "scripts/webui.py", line 841, in process_images
grid = image_grid(output_images, batch_size, force_n_rows=1 << ((len(prompt_matrix_parts)-1)//2), captions=prompt_matrix_parts)
File "scripts/webui.py", line 384, in image_grid
size = d.textbbox( (0,0), captions[i], font=fnt, stroke_width=2, align="center" )
IndexError: list index out of range
* Expose some useful Gradio server launch arguments to command line
* add --grid-format, --gradio-share args
* fixes, add --gradio-disable-queue
* space
This should behave much better, for example if (re)moving older files
from the output directories.
Also remove the increments (which never had any effect), and move the
grid_count calculation right before where it's used.
Max value for a 32bit unsigned int is 2**32 - 1, not 2**32 (and the
range of possible values returned by randint includes both of its
arguments).
Python's hash() function is not deterministic across different
invocations of the interpreter. This meant a given string seed would
produce different results after restarting the script. Use the passed
strings to seed a Random instance instead.
This may open the option to read data from images dragged into the tool later. Activated with --save_metadata
Properties (example output from imagemagic 'identify -verbose' command:
SD:cfg_scale: 7.5
SD:GFPGAN: False
SD:height: 512
SD:normalize_prompt_weights: True
SD:prompt: a beautiful matte painting of a cottage on a magical fantasy island, unreal engine, barometric projection, rectilinear
SD:seed: 247624793
SD:steps: 50
SD:width: 512
If loading Arial fails, try loading a font commonly installed on Linux
distros.
This means it continues to work on Windows, and will also just work on
most Linux machines (DejaVu Sans is widely available and often installed
by default).
