Free image generation AI ``Stable Diffusion'' can be easily installed and used by beginners on Windows & ``Artroom Stable Diffusion'' usage summary that also supports img2img
Google Colaboratory (Google Colab), etc. You need to use the online execution environment . However, command line installation requires knowledge and has a high threshold, and Google Colab has a computational limitation problem. ' Artroom Stable Diffusion ' can be installed in the Windows environment just by clicking, and it also supports img2img.
To generate an image using the image generation AI ``Stable Diffusion'', which was released to the public in August 2022, you can either install it in a local environment equipped with an NVIDIA GPU using the command line, or use
GitHub - artmamedov/artroom-stable-diffusion
Stable Diffusion One-Click Install Local GUI : StableDiffusion
Access the Artroom Stable Diffusion release page and download the EXE format installer (file size 192MB). At the time of writing the article, Artroom Stable Diffusion was an open beta version and the version was 0.3.6.
As a preparation for installation, the PC must be equipped with an NVIDIA GPU and the driver must be updated to the latest version. The PC specifications used in the review are Windows 10 Pro, memory 32 GB, GeForce GTX 1060 and VRAM 6 GB. The recommended spec for Stable Diffusion is VRAM 12B or higher, so I'm a little worried.
Once you're ready, double-click the downloaded installer to launch it.
NMKD Stable Diffusion GUI was built in Anaconda based distribution, but Artroom Stable Diffusion is not. Instead, Anaconda is automatically installed at this stage.
Then, a command prompt will be launched with administrator privileges and the installation will start.
After that, the CUDA Tool Kit and PyTorch are also installed.
When the installation is complete, the following screen will appear. Press the key.
Double-click 'Artroom.exe' in the installation folder to start it.
It looks like this when you start it. First, enter 'cat and dog' at the prompt and click 'Run'.
The first time, the Stable Diffusion model is downloaded, and then the image is generated.
When the generation is finished, the image will be displayed in the center. I tried using it on a weak PC, but the time it took to generate an image was about 1 minute per image. The menu on the right is various settings, from the top, batch name setting, number of images to be generated, batch size, number of steps to be generated, aspect ratio setting, image resolution, CFGScale, sampler, seed value.
You can view the storage folder of generated images by clicking 'View Images' in the left column.
To specify the folder for image output, click 'Setteings', specify the folder with 'Image Output Folder', and click 'Update Settings' OK
Artroom Stable Diffusion can also use img2img, which can generate images from model images. Click 'Upload', select the model image, and click 'Run'. This time I used a photo of a Shiba Inu.
Lexica or clip-retrieval and specify more detailed image output prompts, you may get closer to the image you want to output.
This is the output image. Although it is 'cat and dog', there is only a cat, and it should have been a photograph, but for some reason it became a stained glass image. If you use sites such as