Added ui sopport for multi control samples and datasets. Added qwen image edit 5209 to the ui

This commit is contained in:
Jaret Burkett
2025-09-25 11:10:02 -06:00
parent 454be0958a
commit 1069dee0e4
10 changed files with 574 additions and 88 deletions

View File

@@ -0,0 +1,105 @@
---
job: extension
config:
# this name will be the folder and filename name
name: "my_first_qwen_image_edit_2509_lora_v1"
process:
- type: 'diffusion_trainer'
# root folder to save training sessions/samples/weights
training_folder: "output"
# uncomment to see performance stats in the terminal every N steps
# performance_log_every: 1000
device: cuda:0
network:
type: "lora"
linear: 16
linear_alpha: 16
save:
dtype: float16 # precision to save
save_every: 250 # save every this many steps
max_step_saves_to_keep: 4 # how many intermittent saves to keep
datasets:
# datasets are a folder of images. captions need to be txt files with the same name as the image
# for instance image2.jpg and image2.txt. Only jpg, jpeg, and png are supported currently
# images will automatically be resized and bucketed into the resolution specified
# on windows, escape back slashes with another backslash so
# "C:\\path\\to\\images\\folder"
- folder_path: "/path/to/images/folder"
# can do up to 3 control image folders, file names must match target file names, but aspect/size can be different
control_path:
- "/path/to/control/images/folder1"
- "/path/to/control/images/folder2"
- "/path/to/control/images/folder3"
caption_ext: "txt"
# default_caption: "a person" # if caching text embeddings, if you don't have captions, this will get cached
caption_dropout_rate: 0.05 # will drop out the caption 5% of time
resolution: [ 512, 768, 1024 ] # qwen image enjoys multiple resolutions
# a trigger word that can be cached with the text embeddings
# trigger_word: "optional trigger word"
train:
batch_size: 1
# caching text embeddings is required for 32GB
cache_text_embeddings: true
# unload_text_encoder: true
steps: 3000 # total number of steps to train 500 - 4000 is a good range
gradient_accumulation: 1
timestep_type: "weighted"
train_unet: true
train_text_encoder: false # probably won't work with qwen image
gradient_checkpointing: true # need the on unless you have a ton of vram
noise_scheduler: "flowmatch" # for training only
optimizer: "adamw8bit"
lr: 1e-4
# uncomment this to skip the pre training sample
# skip_first_sample: true
# uncomment to completely disable sampling
# disable_sampling: true
dtype: bf16
model:
# huggingface model name or path
name_or_path: "Qwen/Qwen-Image-Edit-2509"
arch: "qwen_image_edit_plus"
quantize: true
# to use the ARA use the | pipe to point to hf path, or a local path if you have one.
# 3bit is required for 32GB
qtype: "uint3|ostris/accuracy_recovery_adapters/qwen_image_edit_2509_torchao_uint3.safetensors"
quantize_te: true
qtype_te: "qfloat8"
low_vram: true
sample:
sampler: "flowmatch" # must match train.noise_scheduler
sample_every: 250 # sample every this many steps
width: 1024
height: 1024
# you can provide up to 3 control images here
samples:
- prompt: "Do whatever with Image1 and Image2"
ctrl_img_1: "/path/to/image1.png"
ctrl_img_2: "/path/to/image2.png"
# ctrl_img_3: "/path/to/image3.png"
- prompt: "Do whatever with Image1 and Image2"
ctrl_img_1: "/path/to/image1.png"
ctrl_img_2: "/path/to/image2.png"
# ctrl_img_3: "/path/to/image3.png"
- prompt: "Do whatever with Image1 and Image2"
ctrl_img_1: "/path/to/image1.png"
ctrl_img_2: "/path/to/image2.png"
# ctrl_img_3: "/path/to/image3.png"
- prompt: "Do whatever with Image1 and Image2"
ctrl_img_1: "/path/to/image1.png"
ctrl_img_2: "/path/to/image2.png"
# ctrl_img_3: "/path/to/image3.png"
- prompt: "Do whatever with Image1 and Image2"
ctrl_img_1: "/path/to/image1.png"
ctrl_img_2: "/path/to/image2.png"
# ctrl_img_3: "/path/to/image3.png"
neg: ""
seed: 42
walk_seed: true
guidance_scale: 3
sample_steps: 25
# you can add any additional meta info here. [name] is replaced with config name at top
meta:
name: "[name]"
version: '1.0'