AI imagined images. Pythonic generation of images.

Primary LanguagePythonMIT LicenseMIT

ImaginAIry ๐Ÿค–๐Ÿง 

Downloads Downloads image image Code style: black Python Checks Discord

AI imagined images. Pythonic generation of stable diffusion images and videos *!.

"just works" on Linux and macOS(M1) (and sometimes windows).

# on macOS, make sure rust is installed first
# be sure to use Python 3.10, Python 3.11 is not supported at the moment
>> pip install imaginairy
>> imagine "a scenic landscape" "a photo of a dog" "photo of a fruit bowl" "portrait photo of a freckled woman" "a bluejay"
# Make an animation showing the generation process
>> imagine --gif "a flower"
# Make an AI video
>> aimg videogen --start-image rocket.png

Stable Video Diffusion

Rushed release of Stable Diffusion Video!

Works with Nvidia GPUs. Does not work on Mac or CPU.

On windows you'll need to install torch 2.0 first via https://pytorch.org/get-started/locally/

Usage: aimg videogen [OPTIONS]

  AI generate a video from an image


      aimg videogen --start-image assets/rocket-wide.png

  --start-image TEXT       Input path for image file.
  --num-frames INTEGER     Number of frames.
  --num-steps INTEGER      Number of steps.
  --model TEXT             Model to use. One of: svd, svd_xt, svd_image_decoder, svd_xt_image_decoder
  --fps INTEGER            FPS for the AI to target when generating video
  --output-fps INTEGER     FPS for the output video
  --motion-amount INTEGER  How much motion to generate. value between 0 and 255.
  -r, --repeats INTEGER    How many times to repeat the renders.   [default: 1]
  --cond-aug FLOAT         Conditional augmentation.
  --seed INTEGER           Seed for random number generator.
  --decoding_t INTEGER     Number of frames decoded at a time.
  --output_folder TEXT     Output folder.
  --help                   Show this message and exit.


Whats New


  • ๐ŸŽ‰ video generation using Stable Video Diffusion
  • ๐ŸŽ‰ imaginairy is now backed by the refiners library
    • This was a huge re-write which is why some features are not yet supported. On the plus side, refiners supports cutting edge features (SDXL, image prompts, etc) which will be added to imaginairy soon.
    • self-attention guidance which makes details of images more accurate
  • feature: better logging output: color, error handling
  • feature: support for pytorch 2.0
  • deprecated: support for python 3.8, 3.9
  • deprecated: support for torch 1.13
  • deprecated: support for Stable Diffusion versions 1.4, 2.0, and 2.1
  • broken: most samplers, tile/details controlnet, and memory management

Run API server and StableStudio web interface (alpha)

Generate images via API or web interface. Much smaller featureset compared to the command line tool.

>> aimg server

Visit http://localhost:8000/ and http://localhost:8000/docs

Image Structure Control by ControlNet

(Broken as of 14.0.0)

Generate images guided by body poses, depth maps, canny edges, hed boundaries, or normal maps.

Openpose Control

imagine --control-image assets/indiana.jpg  --control-mode openpose --caption-text openpose "photo of a polar bear"

Canny Edge Control

imagine --control-image assets/lena.png  --control-mode canny "photo of a woman with a hat looking at the camera"

HED Boundary Control

imagine --control-image dog.jpg  --control-mode hed  "photo of a dalmation"

Depth Map Control

imagine --control-image fancy-living.jpg  --control-mode depth  "a modern living room"

Normal Map Control

imagine --control-image bird.jpg  --control-mode normal  "a bird"

Image Shuffle Control

Generates the image based on elements of the control image. Kind of similar to style transfer.

imagine --control-image pearl-girl.jpg  --control-mode shuffle  "a clown"

The middle image is the "shuffled" input image

Editing Instructions Control

Similar to instructPix2Pix (below) but works with any SD 1.5 based model.

imagine --control-image pearl-girl.jpg  --control-mode edit --init-image-strength 0.01 --steps 30  --negative-prompt "" --model openjourney-v2 "make it anime" "make it at the beach" 

Add Details Control (upscaling/super-resolution)

Replaces existing details in an image. Good to use with --init-image-strength 0.2

imagine --control-image "assets/wishbone.jpg" --control-mode details "sharp focus, high-resolution" --init-image-strength 0.2 --steps 30 -w 2048 -h 2048 

Image (re)Colorization (using brightness control)

Colorize black and white images or re-color existing images.

The generated colors will be applied back to the original image. You can either provide a caption or allow the tool to generate one for you.

aimg colorize pearl-girl.jpg --caption "photo of a woman"

Instruction based image edits by InstructPix2Pix

(Broken as of 14.0.0)

Just tell imaginairy how to edit the image and it will do it for you!

Click to see shell commands Use prompt strength to control how strong the edit is. For extra control you can combine with prompt-based masking.
# enter imaginairy shell
>> aimg
๐Ÿค–๐Ÿง > edit scenic_landscape.jpg -p "make it winter" --prompt-strength 20
๐Ÿค–๐Ÿง > edit dog.jpg -p "make the dog red" --prompt-strength 5
๐Ÿค–๐Ÿง > edit bowl_of_fruit.jpg -p "replace the fruit with strawberries"
๐Ÿค–๐Ÿง > edit freckled_woman.jpg -p "make her a cyborg" --prompt-strength 13
๐Ÿค–๐Ÿง > edit bluebird.jpg -p "make the bird wear a cowboy hat" --prompt-strength 10
๐Ÿค–๐Ÿง > edit flower.jpg -p "make the flower out of paper origami" --arg-schedule prompt-strength[1:11:0.3]  --steps 25 --compilation-anim gif

# create a comparison gif
๐Ÿค–๐Ÿง > edit pearl_girl.jpg -p "make her wear clown makeup" --compare-gif
# create an animation showing the edit with increasing prompt strengths
๐Ÿค–๐Ÿง > edit mona-lisa.jpg -p "make it a color professional photo headshot" --negative-prompt "old, ugly, blurry" --arg-schedule "prompt-strength[2:8:0.5]" --compilation-anim gif
๐Ÿค–๐Ÿง > edit gg-bridge.jpg -p "make it night time" --prompt-strength 15  --steps 30 --arg-schedule prompt-strength[1:15:1] --compilation-anim gif

Quick Image Edit Demo

Want just quickly have some fun? Try edit-demo to apply some pre-defined edits.

>> aimg edit-demo pearl_girl.jpg

Prompt Based Masking by clipseg

Specify advanced text based masks using boolean logic and strength modifiers. Mask syntax:

  • mask descriptions must be lowercase
  • keywords (AND, OR, NOT) must be uppercase
  • parentheses are supported
  • mask modifiers may be appended to any mask or group of masks. Example: (dog OR cat){+5} means that we'll select any dog or cat and then expand the size of the mask area by 5 pixels. Valid mask modifiers:
    • {+n} - expand mask by n pixels
    • {-n} - shrink mask by n pixels
    • {*n} - multiply mask strength. will expand mask to areas that weakly matched the mask description
    • {/n} - divide mask strength. will reduce mask to areas that most strongly matched the mask description. probably not useful

When writing strength modifiers keep in mind that pixel values are between 0 and 1.

>> imagine \
    --init-image pearl_earring.jpg \
    --mask-prompt "face AND NOT (bandana OR hair OR blue fabric){*6}" \
    --mask-mode keep \
    --init-image-strength .2 \
    --fix-faces \
    "a modern female president" "a female robot" "a female doctor" "a female firefighter"


>> imagine \
    --init-image fruit-bowl.jpg \
    --mask-prompt "fruit OR fruit stem{*6}" \
    --mask-mode replace \
    --mask-modify-original \
    --init-image-strength .1 \
    "a bowl of kittens" "a bowl of gold coins" "a bowl of popcorn" "a bowl of spaghetti"


Face Enhancement by CodeFormer

>> imagine "a couple smiling" --steps 40 --seed 1 --fix-faces


Upscaling by RealESRGAN

>> imagine "colorful smoke" --steps 40 --upscale
# upscale an existing image
>> aimg upscale my-image.jpg
Python Example
from imaginairy.enhancers.upscale_realesrgan import upscale_image
from PIL import Image
img = Image.open("my-image.jpg")
big_img = upscale_image(i)


Tiled Images

>> imagine  "gold coins" "a lush forest" "piles of old books" leaves --tile

360 degree images

imagine --tile-x -w 1024 -h 512 "360 degree equirectangular panorama photograph of the desert"  --upscale


Use depth maps for amazing "translations" of existing images.

>> imagine --model SD-2.0-depth --init-image girl_with_a_pearl_earring_large.jpg --init-image-strength 0.05  "professional headshot photo of a woman with a pearl earring" -r 4 -w 1024 -h 1024 --steps 50



Given a starting image, one can generate it's "surroundings".

Example: imagine --init-image pearl-earring.jpg --init-image-strength 0 --outpaint all250,up0,down600 "woman standing"


Work with different generation models

Click to see shell command
imagine "valley, fairytale treehouse village covered, , matte painting, highly detailed, dynamic lighting, cinematic, realism, realistic, photo real, sunset, detailed, high contrast, denoised, centered, michael whelan" --steps 60 --seed 1 --arg-schedule model[sd14,sd15,sd20,sd21,openjourney-v1,openjourney-v2] --arg-schedule "caption-text[sd14,sd15,sd20,sd21,openjourney-v1,openjourney-v2]"

Prompt Expansion

You can use {} to randomly pull values from lists. A list of values separated by | and enclosed in { } will be randomly drawn from in a non-repeating fashion. Values that are surrounded by _ _ will pull from a phrase list of the same name. Folders containing .txt phraselist files may be specified via --prompt_library_path. The option may be specified multiple times. Built-in categories:

  3d-term, adj-architecture, adj-beauty, adj-detailed, adj-emotion, adj-general, adj-horror, animal, art-scene, art-movement, 
  art-site, artist, artist-botanical, artist-surreal, aspect-ratio, bird, body-of-water, body-pose, camera-brand,
  camera-model, color, cosmic-galaxy, cosmic-nebula, cosmic-star, cosmic-term, desktop-background, dinosaur, eyecolor, f-stop, 
  fantasy-creature, fantasy-setting, fish, flower, focal-length, food, fruit, games, gen-modifier, hair, hd,
  iso-stop, landscape-type, national-park, nationality, neg-weight, noun-beauty, noun-fantasy, noun-general, 
  noun-horror, occupation, painting-style, photo-term, pop-culture, pop-location, punk-style, quantity, rpg-item, scenario-desc, 
  skin-color, spaceship, style, tree-species, trippy, world-heritage-site


imagine "a {lime|blue|silver|aqua} colored dog" -r 4 --seed 0 (note that it generates a dog of each color without repetition)

imagine "a {_color_} dog" -r 4 --seed 0 will generate four, different colored dogs. The colors will be pulled from an included phraselist of colors.

imagine "a {_spaceship_|_fruit_|hot air balloon}. low-poly" -r 4 --seed 0 will generate images of spaceships or fruits or a hot air balloon

Python example
from imaginairy.enhancers.prompt_expansion import expand_prompts

my_prompt = "a giant {_animal_}"

expanded_prompts = expand_prompts(n=10, prompt_text=my_prompt, prompt_library_paths=["./prompts"])

Credit to noodle-soup-prompts where most, but not all, of the wordlists originate.

Generate image captions (via BLIP)

>> aimg describe assets/mask_examples/bowl001.jpg
a bowl full of gold bars sitting on a table

Example Use Cases

>> aimg
# Generate endless 8k art
๐Ÿค–๐Ÿง > imagine -w 1920 -h 1080 --upscale "{_art-scene_}. {_painting-style_} by {_artist_}" -r 1000 --steps 30 --model sd21v

# generate endless desktop backgrounds 
๐Ÿค–๐Ÿง > imagine --tile "{_desktop-background_}" -r 100

# convert a folder of images to pencil sketches
๐Ÿค–๐Ÿง > edit other/images/*.jpg -p "make it a pencil sketch"

# upscale a folder of images
๐Ÿค–๐Ÿง > upscale my-images/*.jpg

# generate kitchen remodel ideas
๐Ÿค–๐Ÿง > imagine --control-image kitchen.jpg -w 1024 -h 1024 "{_interior-style_} kitchen" --control-mode depth -r 100 --init-image 0.01 --upscale --steps 35 --caption-text "{prompt}"

Additional Features

  • Generate images either in code or from command line.
  • It just works. Proper requirements are installed. Model weights are automatically downloaded. No huggingface account needed. (if you have the right hardware... and aren't on windows)
  • Noisy logs are gone (which was surprisingly hard to accomplish)
  • WeightedPrompts let you smash together separate prompts (cat-dog)
  • Prompt metadata saved into image file metadata
  • Have AI generate captions for images aimg describe <filename-or-url>
  • Interactive prompt: just run aimg
  • finetune your own image model. kind of like dreambooth. Read instructions on "Concept Training" page

How To

For full command line instructions run aimg --help

from imaginairy import imagine, imagine_image_files, ImaginePrompt, WeightedPrompt, LazyLoadingImage

url = "https://upload.wikimedia.org/wikipedia/commons/thumb/6/6c/Thomas_Cole_-_Architect%E2%80%99s_Dream_-_Google_Art_Project.jpg/540px-Thomas_Cole_-_Architect%E2%80%99s_Dream_-_Google_Art_Project.jpg"
prompts = [
    ImaginePrompt("a scenic landscape", seed=1, upscale=True),
    ImaginePrompt("a bowl of fruit"),
        WeightedPrompt("cat", weight=1),
        WeightedPrompt("dog", weight=1),
        "a spacious building", 
        "a bowl of strawberries", 
        mask_prompt="fruit OR stem{*2}",  # amplify the stem mask x2
    ImaginePrompt("strawberries", tile_mode=True),
for result in imagine(prompts):
    # do something

# or

imagine_image_files(prompts, outdir="./my-art")


  • ~10 gb space for models to download
  • A CUDA supported graphics card with >= 11gb VRAM (and CUDA installed) or an M1 processor.
  • Python installed. Preferably Python 3.10. (not conda)
  • For macOS rust and setuptools-rust must be installed to compile the tokenizer library. They can be installed via: curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh and pip install setuptools-rust

Running in Docker

See example Dockerfile (works on machine where you can pass the gpu into the container)

docker build . -t imaginairy
# you really want to map the cache or you end up wasting a lot of time and space redownloading the model weights
docker run -it --gpus all -v $HOME/.cache/huggingface:/root/.cache/huggingface -v $HOME/.cache/torch:/root/.cache/torch -v `pwd`/outputs:/outputs imaginairy /bin/bash

Running on Google Colab

Example Colab


Q: How do I change the cache directory for where models are stored?

A: Set the HUGGINGFACE_HUB_CACHE environment variable.

Q: How do I free up disk space?

A: The AI models are cached in ~/.cache/ (or HUGGINGFACE_HUB_CACHE). To delete the cache remove the following folders:

  • ~/.cache/imaginairy
  • ~/.cache/clip
  • ~/.cache/torch
  • ~/.cache/huggingface



  • ๐ŸŽ‰ feature: video generation using Stable Video Diffusion
  • ๐ŸŽ‰ feature: imaginairy is now backed by the refiners library
    • This was a huge re-write which is why some features are not yet supported. On the plus side, refiners supports cutting edge features (SDXL, image prompts, etc) which will be added to imaginairy soon.
    • feature: added support for self-attention guidance which makes details of images more accurate
  • feature: adds support for pytorch 2.0
  • deprecated: support for python 3.8
  • deprecated: support for torch 1.13
  • deprecated: support for sd 1.4, 2.0, 2.1
  • broken: pix2pix editing, most samplers, some controlnets, and memory management


  • fix: pydantic models for http server working now. Fixes #380
  • fix: install triton so annoying message is gone


  • fix: allow tile_mode to be set to True or False for backward compatibility
  • fix: various pydantic issues have been resolved
  • feature: switch to pydantic 2.3 (faster but was a pain to migrate)


  • feature: api server now has feature parity with the python API. View the docs at after running aimg server
    • ImaginePrompt is now a pydantic model and can thus be sent over the rest API
    • images are expected in base64 string format
  • fix: pin pydantic to 2.0 for now
  • build: better python 3.11 incompatibility messaging (fixes #342)
  • build: add minimum versions to requirements to improve dependency resolution
  • docs: add a discord link


  • feature: show full stack trace when there is an api error
  • fix: make lack of support for python 3.11 explicit
  • fix: add some routes to match StableStudio routes


  • ๐ŸŽ‰ feature: multi-controlnet support. pass in multiple --control-mode, --control-image, and --control-image-raw arguments.
  • ๐ŸŽ‰ feature: add colorization controlnet. improve aimg colorize command
  • ๐ŸŽ‰๐Ÿงช feature: Graphical Web Interface StableStudio. run aimg server and visit
  • ๐ŸŽ‰๐Ÿงช feature: API server aimg server command. Runs a http webserver (not finished). After running, visit for api.
  • ๐ŸŽ‰๐Ÿงช feature: API support for Stablity AI's new open-source Generative AI interface, StableStudio.
  • ๐ŸŽ‰๐Ÿงช feature: "better" memory management. If GPU is full, least-recently-used model is moved to RAM. I'm not confident this works well.
  • feature: [disabled] inpainting controlnet can be used instead of finetuned inpainting model
    • The inpainting controlnet doesn't work as well as the finetuned model
  • feature: python interface allows configuration of controlnet strength
  • feature: show full stack trace on error in cli
  • fix: hide the "triton" error messages
  • fix: package will not try to install xformers on aarch64 machines. While this will allow the dockerfile to build on MacOS M1, torch will not be able to use the M1 when generating images.
  • build: specify proper Pillow minimum version (fixes #325)
  • build: check for torch version at runtime (fixes #329)


  • fix: exclude broken versions of timm as dependencies


  • fix: move normal map preprocessor for conda compatibility


  • fix: use correct device for depth images on mps. Fixes #300


  • ๐ŸŽ‰ feature: add "detail" control mode. Add details to an image. Great for upscaling an image.
  • ๐ŸŽ‰ feature: add "edit" control mode. Edit images using text instructions with any SD 1.5 based model. Similar to instructPix2Pix.
  • ๐ŸŽ‰ feature: add "shuffle" control mode. Image is generated from elements of control image. Similar to style transfer.
  • ๐ŸŽ‰ feature: upgrade to controlnet 1.1
  • ๐ŸŽ‰ fix: controlnet now works with all SD 1.5 based models
  • feature: add openjourney-v4
  • fix: raw control images are now properly loaded. fixes #296
  • fix: filenames start numbers after latest image, even if some previous images were deleted


  • fix: fix globbing bug with input image path handling
  • fix: changed sample to True to generate caption using blip model


  • docs: add some example use cases
  • feature: add art-scene, desktop-background, interior-style, painting-style phraselists
  • fix: compilation animations create normal slideshows instead of "bounces"
  • fix: file globbing works in the interactive shell
  • fix: fix model downloads that were broken by library change in transformers 4.27.0


  • all these changes together mean same seed/sampler will not be guaranteed to produce same image (thus the version bump)
  • fix: image composition didn't work very well. Works well now but probably very slow on non-cuda platforms
  • fix: remove upscaler tiling message
  • fix: improve k-diffusion sampler schedule. significantly improves image quality of default sampler
  • fix: img2img was broken for all samplers except plms and ddim when init image strength was >~0.25


  • feature: input raw control images (a pose, canny map, depth map, etc) directly using --control-image-raw This is opposed to current behavior of extracting the control signal from an input image via --control-image
  • feature: aimg model-list command lists included models
  • feature: system memory added to aimg system-info command
  • feature: add --fix-faces options to aimg upscale command
  • fix: add missing metadata attributes to generated images
  • fix: image composition step was producing unnecessarily blurry images
  • refactor: split aimg cli code into multiple files
  • docs: pypi docs now link properly to github automatically


  • feature: ๐ŸŽ‰ ControlNet integration! Control the structure of generated images.
  • feature: aimg colorize attempts to use controlnet to colorize images
  • feature: --caption-text command adds text at the bottom left of an image


  • fix: edit was broken


  • feature: ๐ŸŽ‰๐ŸŽ‰ Make large images while retaining composition. Try imagine "a flower" -w 1920 -h 1080
  • fix: create compilations directory automatically
  • perf: sliced encoding of images to latents (removes memory bottleneck)
  • perf: use Silu for performance improvement over nonlinearity
  • perf: xformers added as a dependency for linux and windows. Gives a nice speed boost.
  • perf: sliced attention now runs on MacOS. A typo prevented that from happening previously.
  • perf: sliced latent decoding - now possible to make much bigger images. 3310x3310 on 11 GB GPU.


  • fix: edit interface was broken


  • fix: use entry_points for windows since setup.py scripts doesn't work on windows #239


  • perf: cli now has minimal overhead such that aimg --help runs in ~650ms instead of ~3400ms
  • feature: edit and imagine commands now accept multiple images (which they will process separately). This allows batch editing of images as requested in #229
  • refactor: move --surprise-me to its own subcommand edit-demo
  • feature: allow selection of output image format with --output-file-extension
  • docs: make training fail on MPS platform with useful error message
  • docs: add directions on how to change model cache path


  • fix: init-image-strength type


  • feature: create gifs or mp4s from any images made in a single run with --compilation-anim gif
  • feature: create a series of images or edits by iterating over a parameter with the --arg-schedule argument
  • feature: openjourney-v1 and openjourney-v2 models added. available via --model openjourney-v2
  • feature: add upscale command line function: aimg upscale
  • feature: --gif option will create a gif showing the generation process for a single image
  • feature: --compare-gif option will create a comparison gif for any image edits
  • fix: tile mode was broken since latest perf improvements


  • feature: added aimg system-info command to help debug issues


  • feature: some memory optimizations and documentation
  • feature: surprise-me improvements
  • feature: image sizes can now be multiples of 8 instead of 64. Inputs will be silently rounded down.
  • feature: cleaned up aimg shell logs
  • feature: auto-regen for unsafe images
  • fix: make blip filename windows compatible
  • fix: make captioning work with alpha pngs


  • fix: bypass huggingface cache retrieval bug


  • fix: limit attention slice size on MacOS machines with 64gb (#175)


  • fix: use python 3.7 compatible lru_cache
  • fix: use windows compatible filenames


  • fix: hf_hub_download() got an unexpected keyword argument 'token'


  • fix: spelling mistake of "surprise"


  • feature: ๐ŸŽ‰ edit images with instructions alone!
  • feature: when editing an image add --gif to create a comparision gif
  • feature: aimg edit --surprise-me --gif my-image.jpg for some fun pre-programmed edits
  • feature: prune-ckpt command also removes the non-ema weights


  • fix: default model config was broken
  • feature: print version with --version
  • feature: ability to load safetensors
  • feature: ๐ŸŽ‰ outpainting. Examples: --outpaint up10,down300,left50,right50 or --outpaint all100 or --outpaint u100,d200,l300,r400


  • fix: handle old pytorch lightning imports with a graceful failure (fixes #161)
  • fix: handle failed image generations better (fixes #83)


  • fix: run face enhancement on GPU for 10x speedup


  • fix: incorrect config files being used for non-1.0 models


  • feature: ๐ŸŽ‰ finetune your own image model. kind of like dreambooth. Read instructions on "Concept Training" page
  • feature: image prep command. crops to face or other interesting parts of photo
  • fix: back-compat for hf_hub_download
  • feature: add prune-ckpt command
  • feature: allow specification of model config file


  • feature: ๐ŸŽ‰ depth-based image-to-image generations (and inpainting)
  • fix: k_euler_a produces more consistent images per seed (randomization respects the seed again)


  • feature: ๐ŸŽ‰ tile in a single dimension ("x" or "y"). This enables, with a bit of luck, generation of 360 VR images. Try this for example: imagine --tile-x -w 1024 -h 512 "360 degree equirectangular panorama photograph of the mountains" --upscale


  • fix: memory/speed regression introduced in 6.1.0
  • fix: model switching now clears memory better, thus avoiding out of memory errors


  • feature: ๐ŸŽ‰ Stable Diffusion 2.1. Generated people are no longer (completely) distorted. Use with --model SD-2.1 or --model SD-2.0-v


  • feature: negative prompting. --negative-prompt or ImaginePrompt(..., negative_prompt="ugly, deformed, extra arms, etc")
  • feature: a default negative prompt is added to all generations. Images in SD-2.0 don't look bad anymore. Images in 1.5 look improved as well.


  • fix: add back in memory-efficient algorithms


  • feature: xformers will be used if available (for faster generation)
  • fix: version metadata was broken


  • feature: use different default steps and image sizes depending on sampler and model selected
  • fix: #110 use proper version in image metadata
  • refactor: samplers all have their own class that inherits from ImageSampler
  • feature: ๐ŸŽ‰๐ŸŽ‰๐ŸŽ‰ Stable Diffusion 2.0
    • --model SD-2.0 to use (it makes worse images than 1.5 though...)
    • Tested on macOS and Linux
    • All samplers working for new 512x512 model
    • New inpainting model working
    • 768x768 model working for all samplers except PLMS (--model SD-2.0-v )


  • feature: add progress image callback


  • fix: support larger images on M1. Fixes #8
  • fix: support CPU generation by disabling autocast on CPU. Fixes #81


  • feature: ๐ŸŽ‰ inpainting support using new inpainting model from RunwayML. It works really well! By default, the inpainting model will automatically be used for any image-masking task
  • feature: ๐ŸŽ‰ new default sampler makes image generation more than twice as fast
  • feature: added DPM++ 2S a and DPM++ 2M samplers.
  • feature: improve progress image logging
  • fix: fix bug with --show-work. fixes #84
  • fix: add workaround for pytorch bug affecting macOS users using the new DPM++ 2S a and DPM++ 2M samplers.
  • fix: add workaround for pytorch mps bug affecting k_dpm_fast sampler. fixes #75
  • fix: larger image sizes now work on macOS. fixes #8


  • feature: allow dynamic switching between models/weights --model SD-1.5 or --model SD-1.4 or --model path/my-custom-weights.ckpt)
  • feature: log total progress when generating images (image X out of Y)


  • feature: stable diffusion 1.5 (slightly improved image quality)
  • feature: dilation and erosion of masks Previously the + and - characters in a mask (example: face{+0.1}) added to the grayscale value of any masked areas. This wasn't very useful. The new behavior is that the mask will expand or contract by the number of pixel specified. The technical terms for this are dilation and erosion. This allows much greater control over the masked area.
  • feature: update k-diffusion samplers. add k_dpm_adaptive and k_dpm_fast
  • feature: img2img/inpainting supported on all samplers
  • refactor: consolidates img2img/txt2img code. consolidates schedules. consolidates masking
  • ci: minor logging improvements


  • fix: k-samplers were broken


  • feature: improved safety filter


  • ๐ŸŽ‰ feature: prompt expansion
  • feature: make (blip) photo captions more descriptive


  • fix: face fidelity default was broken


  • feature: model weights file can be specified via --model-weights-path argument at the command line
  • fix: set face fidelity default back to old value
  • fix: handle small images without throwing exception. credit to @NiclasEriksen
  • docs: add setuptools-rust as dependency for macos


  • fix: init image is fully ignored if init-image-strength = 0


  • feature: face enhancement fidelity is now configurable



  • fix memory leak in face enhancer
  • fix blurry inpainting
  • fix for pillow compatibility


  • ๐ŸŽ‰ fix: inpainted areas correlate with surrounding image, even at 100% generation strength. Previously if the generation strength was high enough the generated image would be uncorrelated to the rest of the surrounding image. It created terrible looking images.
  • ๐ŸŽ‰ feature: interactive prompt added. access by running aimg
  • ๐ŸŽ‰ feature: Specify advanced text based masks using boolean logic and strength modifiers. Mask descriptions must be lowercase. Keywords uppercase. Valid symbols: AND, OR, NOT, (), and mask strength modifier {+0.1} where + can be any of + - * /. Single character boolean operators also work (|, &, !)
  • ๐ŸŽ‰ feature: apply mask edits to original files with mask_modify_original (on by default)
  • feature: auto-rotate images if exif data specifies to do so
  • fix: mask boundaries are more accurate
  • fix: accept mask images in command line
  • fix: img2img algorithm was wrong and wouldn't at values close to 0 or 1


  • fix: another bfloat16 fix


  • fix: make sure image tensors come to the CPU as float32 so there aren't compatibility issues with non-bfloat16 cpus


  • fix: maybe address #13 with expected scalar type BFloat16 but found Float
    • at minimum one can specify --precision full now and that will probably fix the issue
  • feature: tile mode can now be specified per-prompt


  • fix: missing config file for describe feature


  • img2img now supported with PLMS (instead of just DDIM)
  • added image captioning feature aimg describe dog.jpg => a brown dog sitting on grass
  • added new commandline tool aimg for additional image manipulation functionality


  • support multiple additive targets for masking with | symbol. Example: "fruit|stem|fruit stem"


  • added prompt based image editing. Example: "fruit => gold coins"
  • test coverage improved


  • allow urls as init-images


  • img2img actually does # of steps you specify
  • performance optimizations
  • numerous other changes

Not Supported

  • exploratory features that don't work well