ComfyUI-Extra-Samplers

ComfyUI Extra Samplers

Currently included extra samplers:

• RES (+ Somewhat naively momentumized and modified, thanks to Kat and Birch-San for the source implementation!)
• DPMPP Dual SDE (+ Somewhat naively momentumized, tis simply DPMPP SDE with an added SDE akin to how 3M samples)
• Clyb 4M SDE (A modified DPMPP 3M SDE, with an added SDE, egotisticalized by yours truly)
• TTM (Thanks to Kat and Birch-San for the source implementation!)
• LCM Custom Noise (Supports different types of noise other than generic gaussian)
• DPMPP 3M SDE with Dynamic ETA (Anneals down towards a minimum eta via a cosine curve)
• Supreme (Many extra functionalities and step methods available)

Currently included extra K-Sampling nodes:

• SamplerCustomNoise (Supports custom noises other than gaussian noise for init noise)
• SamplerCustomNoiseDuo (Same as above, but with an added High-res fix for simplicity.)
• SamplerCustomModelMixtureDuo (Samples with custom noises, and switches between model1 and model2 every step. If you encounter vram errors, try adding/removing --disable-smart-memory when launching ComfyUI)

Currently included extra Guider nodes:

• GeometricCFGGuider: Samples the two conditionings, then blends between them using a user-chosen alpha.
• ScaledCFGGuider: Samples the two conditionings, then adds it using a method similar to "Add Trained Difference" from merging models.
• ImageAssistedCFGGuider: Samples the conditioning, then adds in the latent image using vector projection onto the CFG. Image latent ought to be of the same size as the diffusion latent.

Supreme Sampler features:

• step_method: You have the ability to choose your own step method with this sampler! Optionally, there's a dynamic step method, which chooses the appropriate order based on the calculated error between steps, allowing you to obtain higher quality when it matters in the sampling process. Defaults to (Euler).
• centralization: Subtracts mean from the denoised latent. This can lead to perceptually sharper results, though may change the perceivable brightness of the image. Conservatively defaults to (0.05).
• normalization: Divides the denoised latent by the standard deviation. Can increase contrast in the image, though may hurt fidelity and coherency at high strengths. Conservatively defaults to (0.05).
• edge_enhancement: Sharpens the latent, and then applies a bilateral blur, leaving the edges sharpened. Defaults to (0.25).
• perphist: Adds previous denoised variable to the current denoised using perpendicular vector projection. Default of (0.5), where higher values add the old denoised variable, and negative values subtract the old denoised variable.
• substeps: Amount of times to iterate over each step and average the results. Can be useful for obtaining higher quality at a given step count. Inspired by ReNoise. Default of (2).
• noise_modulation: Modulates the noise based on the denoising step or other functionality. Current options with a default of ("intensity"): ("none", "intensity", "frequency")
• modulation_strength: Strength of the modulation, utilizing a weighted sum between the modulated noise and normal noise. Default of (2.0). Only has an effect when noise_modulation != "none".
• modulation_dims: Chooses where to apply noise modulation. (1) is in the channels only, perceptually the strongest effect. (2) is in the height and width of the noisy tensor, and has the weakest effect (but that doesn't mean worse). (3) is both channels and height n' width.
• reversible_dampen: Multiplier of the reversible correction in the reversible_ samplers. Increase for less reversibility, and decrease for more. There be dragons lower than .5 with low steps.

Noise Modulation Functionality:

• modulation_dims: Choose between C, HW, or CHW. (In regards to the latent's channels, height + width, or channels + height + width)

• none: No change from normal ancestraln