A Python script that lets you paint on a canvas and sends that image every stroke to the automatic1111 API and updates the canvas when the image is generated.

Note : "Cycle" shortcuts type will wait for the shift key to be released before launching the rendering.

Windows: Link to step-by-step Windows installation instructions with screenshots

macOS: Link to step-by-step MacOS installation instructions with screenshots

Linux: To Do

TLDR; Web UI with the ControlNet extension, API mode enabled in settings, and the AI Models

On first launch of the Start file, the script will create the configs\config.json, configs\controlnet.json and configs\img2img.json configuration files as needed. The ControlNet available models for scribble and lineart will be automatically fetched from your API and set in configuration.

The config.json file handles global interface and script configuration. For these settings, the value used on application start is the first value of each of those list:

• controlnet_models
• detectors
• samplers
• hr_scales
• hr_upscalers
• denoising_strengths

The configs\controlnet.json or configs\img2img.json files can be used to configure the prompt, negative prompt, seed, controlnet model, etc. When you save the json file the program will use it after the next brush stroke or when you press enter.

A controlnet-high.json-dist example configuration file is available for better image quality, at the cost of longer rendering time. Use only with a powerful graphics card.

If you want to add additional models into your controlnet extension you can do so by adding the model folder into the models folder of the controlnet extension.

    ".\stable-diffusion-webui\extensions\sd-webui-controlnet\models"

The quick mode is able to use a very low number of steps with Latent Consistency Models LoRAs . To be able to use it fully, please download the LCM LoRAs for SD 1.5 and/or SDXL on huggingface :

• LCM LoRA for SD 1.5
• LCM LoRA for SDXL

Save the safetensors files as LCM_LoRA_Weights.safetensors in the ./models/LoRA/ directory of your Stable Diffusion WebUI installation. The name of the LoRA is configurable in the JSON configuration files controlnet.json and img2img.json with the quick.lora entry. You can also configure the steps, cfg scale, sampler, and LoRA weight used. The default quick setting gives a good quality/performance ratio with DPM++ SDE Karras and 6 steps. HRFix is disabled when first entering quick mode, but you can activate it if wanted by pressing the h key.

The images can be auto-saved after each rendering into outputs and outputs/autosave directories. The maximum number of images saved is set by autosave_images_max in config.json.

This feature can be disabled on start by setting autosave_images to "false".

You can update the config.json "controlnet_models" list to have multiple ControlNet models available. You can then cycle between the models using the l and m keys.

The models set by default correspond to the ones available on https://huggingface.co/comfyanonymous/ControlNet-v1-1_fp16_safetensors :

• Scribble
• Lineart
• Lineart anime

Note: "pixel perfect" preprocessor mode is disabled by default, because it can be detrimental with scribble and lineart models when used with a resolution > 512. Try out and compare the results.

You can also update the config.json "detectors" list to configure the line detection on the generated image. Cycle between detections with the d key.

The default detectors are:

• Lineart
• Lineart coarse
• Sketch (pidinet)
• Scribble (pidinet)

You can find the full list of supported modules with this URL http://127.0.0.1:7860/controlnet/module_list .

You can save the current rendering settings by using ctrl + keypad 1-9, and the current ControlNet settings by using ctrl + alt + keypad 1-9. Those presets persist in a local configs\presets.json file and are available even after the application is restarted.

You can load rendering settings with keypad 1-9, and ControlNet settings with alt + keypad 1-9. Not all the saved fields are applied by default : the settings that are applied are determined respectively by the preset_fields and cn_preset_fields entries of the configs\config.json file. For example, if you want to also apply the sampler value of the preset on recall, add 'sampler' to your preset_fields list.

Alternatively, you can use web interface. Just run SdPaint\StartWeb.bat. After creating virtual environment and installing packages it will open interface in your default browser. More in this instruction

Launch the program with --img2img --source <image_file_path> to watch an image file for changes, and use it as img2img source. If the script is launched without --source argument, a loading file dialog will be displayed. The configs\img2img.json file is used in this mode.

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

Please make sure to update tests as appropriate.

MIT