/Blender-ControlNet

Using ControlNet right in Blender.

Primary LanguagePythonMIT LicenseMIT

Blender-ControlNet

Using Multiple ControlNet in Blender.

Required

Usage

1. Start A1111 in API mode.

First, of course, is to run web ui with --api commandline argument

  • example in your "webui-user.bat": set COMMANDLINE_ARGS=--api

2. Install Mikubill/sd-webui-controlnet extension

You have to install the Mikubill/sd-webui-controlnet extension in A1111 and download the ControlNet models.
Please refer to the installation instructions from Mikubill/sd-webui-controlnet.

Notes

  • In new version of Mikubill/sd-webui-controlnet, you need to enable Allow other script to control this extension in settings for API access.
  • To enable Multi ControlNet, change Multi ControlNet: Max models amount (requires restart) in the settings. Note that you will need to restart the WebUI for changes to take effect.

3. Copy and paste the multicn.py code into your blender Scripting pane.

4. How to use the script?

Basically, the script utilizes Blender Compositor to generate the required maps and then sends them to AUTOMATIC1111.

To generate the desired output, you need to make adjustments to either the code or Blender Compositor nodes before pressing F12. To simplify this process, I have provided a basic Blender template that sends depth and segmentation maps to ControlNet.

Here is a brief tutorial on how to modify to suit @toyxyz3's rig if you wish to send openpose/depth/canny maps.

Notes

  • Make sure you have the right name for controlnet_model, hash does matter. You can get your local installed ControlNet models list from here.

5. Hit "Run Script"

Before you hit "Run Script", here are the parameters that you may want to modify in the scripts:

# specify your images output folder
IMAGE_FOLDER = "//sd_results"

# if you don't want to send your maps to AI, set this option to False
is_using_ai = True

# which maps are you going to send to AI
is_send_canny = False
is_send_depth = False
is_send_bone = True
is_send_seg = False

# prepare data for API
params = {
    "prompt": "a room",
    "negative_prompt": "(worst quality:2), (low quality:2), (normal quality:2), lowres, normal quality",
    "width": get_output_width(scene),
    "height": get_output_height(scene),
    "sampler_index": "DPM++ SDE Karras",
    "sampler_name": "",
    "batch_size": 1,
    "n_iter": 1,
    "steps": 20,
    "cfg_scale": 7,
    "seed": -1,
    "subseed": -1,
    "subseed_strength": 0,
    "restore_faces": False,
    "enable_hr": False,
    "hr_scale": 1.5,
    "hr_upscaler": "R-ESRGAN General WDN 4xV3",
    "denoising_strength": 0.5,
    "hr_second_pass_steps": 10,
    "hr_resize_x": 0,
    "hr_resize_y": 0,
    "firstphase_width": 0,
    "firstphase_height": 0,
    "override_settings": {"CLIP_stop_at_last_layers": 2},
    "override_settings_restore_afterwards": True,
    "alwayson_scripts": {"controlnet": {"args": []}},
}

canny_cn_units = {
    "mask": "",
    "module": "none",
    "model": "diff_control_sd15_canny_fp16 [ea6e3b9c]",
    "weight": 1.2,
    "resize_mode": "Scale to Fit (Inner Fit)",
    "lowvram": False,
    "processor_res": 64,
    "threshold_a": 64,
    "threshold_b": 64,
    "guidance": 1,
    "guidance_start": 0.19,
    "guidance_end": 1,
    "guessmode": False,
}

depth_cn_units = {
    "mask": "",
    "module": "none",
    "model": "diff_control_sd15_depth_fp16 [978ef0a1]",
    "weight": 1.2,
    "resize_mode": "Scale to Fit (Inner Fit)",
    "lowvram": False,
    "processor_res": 64,
    "threshold_a": 64,
    "threshold_b": 64,
    "guidance": 1,
    "guidance_start": 0.19,
    "guidance_end": 1,
    "guessmode": False,
}

bone_cn_units = {
    "mask": "",
    "module": "none",
    "model": "diff_control_sd15_openpose_fp16 [1723948e]",
    "weight": 1.1,
    "resize_mode": "Scale to Fit (Inner Fit)",
    "lowvram": False,
    "processor_res": 64,
    "threshold_a": 64,
    "threshold_b": 64,
    "guidance": 1,
    "guidance_start": 0,
    "guidance_end": 1,
    "guessmode": False,
}

seg_cn_units = {
    "mask": "",
    "module": "none",
    "model": "diff_control_sd15_seg_fp16 [a1e85e27]",
    "weight": 1,
    "resize_mode": "Scale to Fit (Inner Fit)",
    "lowvram": False,
    "processor_res": 64,
    "threshold_a": 64,
    "threshold_b": 64,
    "guidance": 1,
    "guidance_start": 0,
    "guidance_end": 1,
    "guessmode": False,
}

6. Hit F12, wait for it...

Bonus

To create 150 ControlNet segmentation colors materials, run seg.py. Check out this tweet for instructions.