The base model was then fused with a selection of LoRAs, some of which are NSFW in nature. Please use responsibily. Please be aware of the licenses of both the models before using the params commercially.

This model was created as part of the ongoing OpenSight project, mostly for dataset generation and evaluation purposes.

The following context provided by sayakpaul, of the original merged model.

Prompt: `An Instagram profile picture of an Asian model taken at a rooftop penthouse pool party.`

Dev (50 steps)	Dev (4 steps)	Dev + Schnell Merge (4 steps)	This Model (6-8 steps recommended)

Sub-memory-efficient merging code

python

from diffusers import FluxTransformer2DModel
from huggingface_hub import snapshot_download
from accelerate import init_empty_weights
from diffusers.models.model_loading_utils import load_model_dict_into_meta
import safetensors.torch
import glob
import torch


with init_empty_weights():
    config = FluxTransformer2DModel.load_config("black-forest-labs/FLUX.1-dev", subfolder="transformer")
    model = FluxTransformer2DModel.from_config(config)

dev_ckpt = snapshot_download(repo_id="black-forest-labs/FLUX.1-dev", allow_patterns="transformer/*")
schnell_ckpt = snapshot_download(repo_id="black-forest-labs/FLUX.1-schnell", allow_patterns="transformer/*")

dev_shards = sorted(glob.glob(f"{dev_ckpt}/transformer/*.safetensors"))
schnell_shards = sorted(glob.glob(f"{schnell_ckpt}/transformer/*.safetensors"))

merged_state_dict = {}
guidance_state_dict = {}

for i in range(len((dev_shards))):
    state_dict_dev_temp = safetensors.torch.load_file(dev_shards[i])
    state_dict_schnell_temp = safetensors.torch.load_file(schnell_shards[i])

    keys = list(state_dict_dev_temp.keys())
    for k in keys:
        if "guidance" not in k:
            merged_state_dict[k] = (state_dict_dev_temp.pop(k) + state_dict_schnell_temp.pop(k)) / 2
        else:
            guidance_state_dict[k] = state_dict_dev_temp.pop(k)

    if len(state_dict_dev_temp) > 0:
        raise ValueError(f"There should not be any residue but got: {list(state_dict_dev_temp.keys())}.")
    if len(state_dict_schnell_temp) > 0:
        raise ValueError(f"There should not be any residue but got: {list(state_dict_dev_temp.keys())}.")

merged_state_dict.update(guidance_state_dict)
load_model_dict_into_meta(model, merged_state_dict)

model.to(torch.bfloat16).save_pretrained("merged-flux")

Changelog:

7 April 2025 - Just noticed a potential mistake when loading some components (particularly the text_encoder2), feel free to load from the base dev folder, works the same as it is a merge.

buildborderless/FLUX.MF-8step_lightning_v2

FLUX-merged_lightning-v2

The following context provided by sayakpaul, of the original merged model.

Sub-memory-efficient merging code

buildborderless/
FLUX.MF-8step_lightning_v2