imagine
Generate images via AI.
Can be invoked via either aimg imagine
or just imagine
.
Usage:
Options:
--negative-prompt TEXT Negative prompt. Things to try and exclude
from images. Same negative prompt will be
used for all images.
--prompt-strength FLOAT How closely to follow the prompt. Image
looks unnatural at higher values [default:
7.5]
--init-image PATH|URL Starting image.
--init-image-strength FLOAT Starting image strength. Between 0 and 1.
--image-prompt PATH|URL Starting image.
--image-prompt-strength FLOAT Starting image strength. Between 0 and 1.
--outdir PATH Where to write results to. [default:
./outputs]
--output-file-extension [jpg|png]
Where to write results to. [default: jpg]
-r, --repeats INTEGER How many times to repeat the renders. If you
provide two prompts and --repeat=3 then six
images will be generated. [default: 1]
--size TEXT Image size as a string. Can be a named size,
WIDTHxHEIGHT, or single integer. Should be
multiple of 8. Examples: 512x512, 4k, UHD,
8k, 512, 1080p
--steps INTEGER How many diffusion steps to run. More steps,
more detail, but with diminishing returns.
--seed INTEGER What seed to use for randomness. Allows
reproducible image renders.
--upscale
--fix-faces
--fix-faces-fidelity FLOAT How faithful to the original should face
enhancement be. 1 = best fidelity, 0 = best
looking face.
--solver, --sampler [ddim|dpmpp]
Solver algorithm to generate the image with.
(AKA 'Sampler' or 'Scheduler' in other
libraries. [default: ddim]
--log-level [DEBUG|INFO|WARNING|ERROR]
What level of logs to show. [default: INFO]
-q, --quiet Suppress logs. Alias of `--log-level ERROR`.
--show-work Output a debug images to `steps` folder.
--tile Any images rendered will be tileable in both
X and Y directions.
--tile-x Any images rendered will be tileable in the
X direction.
--tile-y Any images rendered will be tileable in the
Y direction.
--allow-compose-phase / --no-compose-phase
Allow the image to be composed at a lower
resolution.
--mask-image PATH|URL A mask to use for inpainting. White gets
painted, Black is left alone.
--mask-prompt TEXT Describe what you want masked and the AI
will mask it for you. You can describe
complex masks with AND, OR, NOT keywords and
parentheses. The strength of each mask can
be modified with {*1.5} notation.
Examples: car AND (wheels{*1.1} OR trunk
OR engine OR windows OR headlights) AND NOT
(truck OR headlights){*10} fruit|fruit stem
--mask-mode [keep|replace] Should we replace the masked area or keep
it? [default: replace]
--mask-modify-original After the inpainting is done, apply the
changes to a copy of the original image.
--outpaint TEXT Specify in what directions to expand the
image. Values will be snapped such that
output image size is multiples of 8.
Examples `--outpaint
up10,down300,left50,right50` `--outpaint
u10,d300,l50,r50` `--outpaint all200`
`--outpaint a200`
--caption Generate a text description of the generated
image.
--precision [full|autocast] Evaluate at this precision. [default:
autocast]
--model-weights-path, --model TEXT
Model to use. Should be one of flux,
miniaturuspotentia, miniaturuspotentia12,
modern-disney, modern-disney-15, modi,
modi15, mp, mp12, od, odv11, oj, oj1, oj2,
oj4, ojv1, ojv2, ojv4, opendalle,
opendalle11, openjourney, openjourney-v1,
openjourney-v2, openjourney-v4,
openjourney1, openjourney2, openjourney4,
potentia, potentia12, sd-1.5,
sd-1.5-inpaint, sd-15, sd-15-inpaint, sd-xl,
sd-xlinpaint, sd1.5, sd1.5-inpaint,
sd1.5inpaint, sd15, sd15-inpaint,
sd15inpaint, sdxl, sdxl-inpaint,
sdxlinpaint, or a path to custom weights.
[default: sd15]
--model-architecture TEXT Model architecture. When specifying custom
weights the model architecture must be
specified. (sd15, sdxl, etc).
--prompt-library-path PATH Path to folder containing phrase lists in
txt files. Use txt filename in prompt:
{_filename_}.
--version Print the version and exit.
--gif Create a gif of the generation.
--compare-gif Create a gif comparing the original image to
the modified one.
--arg-schedule TEXT Schedule how an argument should change over
several generations. Format: `--arg-schedule
arg_name[start:end:increment]` or `--arg-
schedule arg_name[val,val2,val3]`
--compilation-anim [gif|mp4] Generate an animation composed of all the
images generated in this run. Defaults to
gif but `--compilation-anim mp4` will
generate an mp4 instead.
--caption-text TEXT Specify the text to write onto the image.
--composition-strength FLOAT Strength of the composition phase.
--control-image PATH|URL Image used for control signal in image
generation. For example if control-mode is
depth, then the generated image will match
the depth map extracted from the control
image. Defaults to the `--init-image`
--control-image-raw PATH|URL Preprocessed image used for control signal
in image generation. Like `--control-image`
but expects the already extracted signal.
For example the raw control image would be a
depth map orpose information.
--control-strength TEXT Strength of the control signal.
--control-mode [|canny|depth|details|normal|hed|openpose|shuffle|edit|inpaint|colorize|qrcode|densepose]
how the control image is used as signal
--videogen Turns the generated photo into video
--help Show this message and exit.