DPM++ 2M Alt Sampler for ComfyUI

DPM++ 2M Alt Sampler for ComfyUI

Made the DPM++ 2M Alt Karras Sampler to work with ComfyUI.

Comparison: Left DPMPP 2M Karras | Right DPMPP 2M Alt Karras

Note: The face isn’t an accurate comparison due to different seed in facedetailer, other parts of the picture is using the same seed.

Credits to hallastore, original thread: https://github.com/AUTOMATIC1111/stable-diffusion-webui/discussions/8457

Instructions:

Unzip method:

IMPORTANT: Make backup of these two files of just in case new update affects the samplers.

For ComfyUI Portable:

ComfyUI_windows_portable\ComfyUI\comfy\samplers.py
ComfyUI_windows_portable\ComfyUI\comfy\k_diffusion\sampling.py

For Automatic 1111:
Stable-diffusion-webui\extensions\sd-webui-comfyui\ComfyUI\comfy\samplers.py
Stable-diffusion-webui\extensions\sd-webui-comfyui\ComfyUI\comfy\k_diffusion\sampling.py

Unpack zip to directory

For Portable ComfyUI: ComfyUI_windows_portable\ComfyUI\comfy
For Automatic 1111 ComfyUI extension: Stable-diffusion-webui\extensions\sd-webui-comfyui\ComfyUI\comfy

Sampler is named dpmpp_2m_alt in ComfyUI

Manual Method:

IMPORTANT: Make backup of these two files of just in case new update affects the samplers.

For ComfyUI Portable:

ComfyUI_windows_portable\ComfyUI\comfy\samplers.py
ComfyUI_windows_portable\ComfyUI\comfy\k_diffusion\sampling.py

For Automatic 1111:
Stable-diffusion-webui\extensions\sd-webui-comfyui\ComfyUI\comfy\samplers.py
Stable-diffusion-webui\extensions\sd-webui-comfyui\ComfyUI\comfy\k_diffusion\sampling.py

Tips: The following instructions is for ComfyUI Portable but it is the same for Automatic 1111, just navigate to the path given above for the *.py files with the same name.

Open ComfyUI_windows_portable\ComfyUI\comfy\samplers.py

search for KSAMPLER_NAMES
Add dpmpp_2m_alt as a new value and save file, make sure you include the comma and quotes.
Open ComfyUI_windows_portable\ComfyUI\comfy\k_diffusion\sampling.py\

Copy the following code.

@torch.no_grad()
def sample_dpmpp_2m_alt(model, x, sigmas, extra_args=None, callback=None, disable=None):
    """DPM-Solver++(2M)."""
    extra_args = {} if extra_args is None else extra_args
    s_in = x.new_ones([x.shape[0]])
    sigma_fn = lambda t: t.neg().exp()
    t_fn = lambda sigma: sigma.log().neg()
    old_denoised = None

    for i in trange(len(sigmas) - 1, disable=disable):
        denoised = model(x, sigmas[i] * s_in, **extra_args)
        if callback is not None:
            callback({'x': x, 'i': i, 'sigma': sigmas[i], 'sigma_hat': sigmas[i], 'denoised': denoised})
        t, t_next = t_fn(sigmas[i]), t_fn(sigmas[i + 1])
        h = t_next - t
        
        t_min = min(sigma_fn(t_next), sigma_fn(t))
        t_max = max(sigma_fn(t_next), sigma_fn(t))

        if old_denoised is None or sigmas[i + 1] == 0:
            x = (t_min / t_max) * x - (-h).expm1() * denoised
        else:
            h_last = t - t_fn(sigmas[i - 1])

            h_min = min(h_last, h)
            h_max = max(h_last, h)
            r = h_max / h_min

            h_d = (h_max + h_min) / 2
            denoised_d = (1 + 1 / (2 * r)) * denoised - (1 / (2 * r)) * old_denoised
            x = (t_min / t_max) * x - (-h_d).expm1() * denoised_d

        old_denoised = denoised
    return x