Skip to content

v4.3.2 (PAG. Flux Control Nets, img2img & inpainting) Patch

Compare
Choose a tag to compare
@Teriks Teriks released this 03 Nov 11:11
· 102 commits to master since this release

v4.3.2 release with Windows installer.

Due to the size of the packaged python environment, the installer is within a multi-part zip file.

The multipart zip can be extracted using 7-Zip: https://www.7-zip.org/

Download both dgenerate_installer.zip.001 and dgenerate_installer.zip.002 to a folder.

Unzip dgenerate_installer.zip.001 to a directory (Right click, 7-Zip -> Extract to "dgenerate_installer") and then run dgenerate_installer\dgenerate.msi to install.

dgenerate will be installed under C:\Program Files\dgenerate by default with an isolated python environment provided.

The install directory will be added to PATH, and dgenerate will be available from the command line.

Portable Install

A portable install is provided via dgenerate_portable.zip.001 and dgenerate_portable.zip.002, these contain
nothing but the dgenerate executable and a frozen python environment which can be placed anywhere.

4.3.2 Patch

Fix for --model-cpu-offload with 8 bit bitsandbytes quantized weights.

Fix for diffusers / accelerate randn_tensor device mismatch error, occasionally occurring in the following scripted construct upon subsequent invocations:

#! /usr/bin/env dgenerate --file
#! dgenerate 4.3.2

# Iterate over any scheduler options

\setp schedulers [ \
    'EulerAncestralDiscreteScheduler', \
    'KDPM2AncestralDiscreteScheduler', \
    'EulerDiscreteScheduler', \
    'KDPM2DiscreteScheduler', \
    'HeunDiscreteScheduler']

{% for scheduler in schedulers %}
    stabilityai/stable-diffusion-xl-base-1.0
    --model-type torch-sdxl
    --dtype float16
    --variant fp16
    --scheduler {{ scheduler }}
    --model-cpu-offload
    --prompts "add your prompt here"
{% endfor %} !END

Issue only occurred with --model-cpu-offload/--model-sequential-offload enabled due to quirks with accelerate offloading and high memory usage.

4.3.0 Features

  • Flux img2img, inpainting, and Control Net support. InstantX/FLUX.1-dev-Controlnet-Union is also supported with --model-type torch-flux via a mode URI argument to --control-nets, see: Flux Union Control Net Mode in the README.

  • PAG perturbed attention guidance support for Stable Diffusion, SDXL, and SD3, see: Feature Table for support caveats. New arguments have been added for working with PAG: --pag, --pag-scales, --pag-adaptive-scales, --sdxl-refiner-pag, --sdxl-refiner-pag-scales, and --sdxl-refiner-pag-adaptive-scales.

  • Console UI recipe form updates for Flux control nets, added basic UI for specifying one control net model, control net scale, and control net mode for Flux ControlNet Union.