Merge master into cep (resolve TrainingTab.py split conflict, add CEP to Anima/Ideogram/Krea2) - #4

Open

dxqb wants to merge 43 commits into

Koratahiu:cepfrom

dxqb:merge-master-into-cep-pr1235

Commits on Jun 21, 2026

Disable LoRA rank check (false positive on Flux2 LoHA loading) (#1549)
dxqb
authored
07254ad

Commits on Jul 1, 2026

ctk UI file split (#1564)
dxqb
authored
10890e5
Decouple CTK from UI logic; lay groundwork for Qt6 (#1445)
dxqb
authored
75c4082
Qt UI file split (#1565)
dxqb
authored
ebb6f38
implement Qt6 views (#1446)
dxqb
authored
4a660d4
[pre-commit.ci] pre-commit autoupdate (#1553)
pre-commit-ci[bot]
authored
431abe0
Add --preset-path and --config-value CLI overrides to scripts/train.py (#1531)
dxqb
authored
e042774
Model saver code deduplication (#1518)
dxqb
authored
560ce35
fix: missing commit for #1518 (#1567)
dxqb
authored
ef7bf6c
Attention backend selection (#1227)
dxqb
authored
58955db
Fix frozen-text-encoder training crash for Würstchen / Stable Cascade (#1558)
dxqb
authored
736cc75

Commits on Jul 2, 2026

Bump diffusers pin to latest upstream commit (#1569)
dxqb
authored
622d453
Bump safetensors pin to 0.8.0 final release (#1570)
dxqb
authored
1e0263b

Commits on Jul 3, 2026

LoRA output formats Diffusers, Kohya, Comfy, Legacy (#1563)
dxqb
authored
de1437c
New Model: Krea 2 (#1568)
dxqb
authored
e991240
Fix PySide6 param dialogs leaking field validators after close (#1573)
dxqb
authored
95b3392

Commits on Jul 4, 2026

Fix full-finetune save crash when checkpointing + layer offloading are enabled + diffusers output format (#1574)
dxqb
authored
41ee4da
Qt thread safety (#1575)
dxqb
authored
f610daf
Auto-detect legacy CUDA 12.6 install for old drivers (#1572)
dxqb
authored
5f05ea9
Cloud fixes (#1576)
dxqb
authored
597dcd7
New model: Anima (#1487)
dxqb
authored
9547963
Better preset selection (#1578)
dxqb
authored
d4dc082
Remove dead text_encoder None-checks for models that always load their TE (#1579)
dxqb
authored
06f3e0d
Centralize model composition and training methods in ModelType (#1577)
dxqb
authored
6e5ad7a

Commits on Jul 5, 2026

Make gradient checkpointing and offloading per-component (#1476)
dxqb
authored
6fd9b7f
New Model: Ideogram 4 (#1522)
dxqb
authored
2d1f97b
always enable mixed-precision / autocasting (#1481)
dxqb
authored
a2047df
Hide additional embeddings tab for models without embedding support (#1581)
dxqb
authored
65e954e
Register HiDream fine-tune model saver (#1582)
dxqb
authored
e3b8099
Bump huggingface-hub to 1.22.0 and scenedetect to 0.7 (#1583)
dxqb
authored
ec5c333
Disable transformer override UI for Krea2 (#1584)

dxqb
and
claude
authored
81bf744
Fix stale train_ui.py directory guard in update.bat and export_debug.bat (#1586)
dxqb
authored
32f228a
Force Fusion style for PySide6 UI (#1585)
dxqb
authored
978d38b
Presets for Krea, Ideogram, and change dataloader threads (#1588)
dxqb
authored
82173b1
Bump mgds pin to latest master (3a6994a) (#1587)
dxqb
authored
0373945
Fix Ideogram4 full-model save to apply diffusers-to-original qkv fusion (#1589)
dxqb
authored
f8a5bec
Merge pull request #1566 from Nerogar/merge
dxqb
authored
ccf1b63
Restore scripts/train_ui.py as a compat shim for external launchers (#1591)
dxqb
authored
192c6ac

Commits on Jul 11, 2026