sayakpaul

When loading the LoRA params (that were obtained on a quantized base model) and merging them into the base model, it is recommended to first dequantize the base model, merge the LoRA params into it, and then quantize the model again. This is because merging into 4bit quantized models can lead to some rounding errors. Below, we provide an end-to-end example:

1. First, load the original model and merge the LoRA params into it:

from diffusers import FluxPipeline 
import torch 

ckpt_id = "black-forest-labs/FLUX.1-dev"
pipeline = FluxPipeline.from_pretrained(

ariG23498

# Taken from: https://gist.github.com/sayakpaul/23862a2e7f5ab73dfdcc513751289bea

from diffusers import FluxPipeline, FluxTransformer2DModel
from transformers import T5EncoderModel
import torch
import gc


def flush():
    gc.collect()

sayakpaul

from diffusers import FluxPipeline, AutoencoderKL
from diffusers.image_processor import VaeImageProcessor
from transformers import T5EncoderModel, T5TokenizerFast, CLIPTokenizer, CLIPTextModel
import torch
import gc


def flush():
    gc.collect()
    torch.cuda.empty_cache()

kohya-ss

    def forward(self, x, timesteps=None, context=None, y=None, **kwargs):
        # broadcast timesteps to batch dimension
        timesteps = timesteps.expand(x.shape[0])

        hs = []
        t_emb = get_timestep_embedding(timesteps, self.model_channels)  # , repeat_only=False)
        t_emb = t_emb.to(x.dtype)
        emb = self.time_embed(t_emb)

        assert x.shape[0] == y.shape[0], f"batch size mismatch: {x.shape[0]} != {y.shape[0]}"

Stella2211

# huggingface login
from huggingface_hub import login

login()


# load textencorder in 8bit quantized
from transformers import T5EncoderModel
from diffusers import DiffusionPipeline
import datetime

Norod

# Ran it with the following packages installed:
# accelerate                   0.18.0
# diffusers                    0.16.0.dev0
# torch                        2.0.0+cu118
# torchvision                  0.15.0+cu118
# transformers                 4.28.1
# xformers                     0.0.18

import torch
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler

AmericanPresidentJimmyCarter

from PIL import Image
import stable_inference
import numpy as np
from einops import repeat

# Interpolate video between two conditionings
FOLDER = 'test/painting'
MAX_STRENGTH = 0.5 # Strength at maximum in the middle of the interpolation
SEED = 9001
SECONDS = 10

ds54e

bl_info = {
    "name": "Kuchi Paku Light",
    "author": "ds54e",
    "version": (1, 1, 1),
    "blender": (2, 80, 0),
    "location": "View3D > Sidebar > KPL",
    "description": "Generate Kuchi-Paku animations from the sound sequences in the VSE",
    "warning": "",
    "doc_url": "",
    "category": "Animation",

ZoomTen

General Tips and Tricks

• Think of clip nodes as containers!
  • You can plug in an arbitrary composite node and make them "instances" of said node

Common Node Configurations

Drop shadow (TinglesMagicWords)

A simple drop shadow effect, assuming you have a solid color and transparent clip such as text.

Celeborn2BeAlive

#!/bin/bash

# This script generate a wheel for OpenTimelineIO in dist folder
# We need it has a patch to install otio 0.12.1 in Windows Blender because it requires compilation
# This wheel should be distributed with code that depends on otio 0.12.1 and installed using Blender's python -m pip

# You need python (3.7.*) and Visual Studio 2017+

CURRENT_DIR=`dirname $0`
ROOT_DIR=$CURRENT_DIR/