NoobAI-XL-Merges

Various merges built on Laxhar Lab's Illustrious-xl-based text to image model, uploaded for testing purposes.

These are provided as-is, and YMMV. The user is responsible for any outputs produced using these checkpoints.

Other models involved in these merges include:

• comin/IterComp
• CyberRealistic XL

Methods

Perpendicular merges are done via sd-mecha using the Python API, for example:

1) Merge noobaiXLNAIXL_vPred10Version-cyberrealistic4-perpendicular

import sd_mecha
sd_mecha.set_log_level()


text_encoder_recipe = sd_mecha.model("noobaiXLNAIXL_vPred10Version.safetensors", "sdxl")

unet_recipe = sd_mecha.add_perpendicular(
    sd_mecha.model("noobaiXLNAIXL_vPred10Version.safetensors", "sdxl"),
    sd_mecha.model("cyberrealisticXL_v4.safetensors", "sdxl"),
    sd_mecha.model("sd_xl_base_1.0_0.9vae.safetensors", "sdxl"),
)

recipe = sd_mecha.weighted_sum(
    text_encoder_recipe,
    unet_recipe,
    alpha=(
        sd_mecha.blocks("sdxl", "txt") |
        sd_mecha.blocks("sdxl", "txt2") |
        sd_mecha.default("sdxl", "unet", 1)
    ),
)

merger = sd_mecha.RecipeMerger(
    models_dir=r"C:\path\to\models\directory",
)

merger.merge_and_save(recipe, output="output.safetensors")

2) Add v_pred and ztsnr keys to the resulting model for autodetection in Comfy/Forge

from safetensors.torch import load_file, save_file
import torch

state_dict = load_file("output.safetensors")
state_dict["v_pred"] = torch.tensor([])
state_dict["ztsnr"] = torch.tensor([])
save_file(state_dict, "noobaiXLNAIXL_vPred10Version-cyberrealistic4-perpendicular.safetensors")