Merge pull request #5047 from vladmandic/dev

refresh master

CHANGELOG.md

@@ -1,6 +1,6 @@
 # Change Log for SD.Next
 
-## Highlights for 2026-08-21
+## Highlights for 2026-08-22
 
 Time for a new release, this is a larger one!
 Main focus is improving video workflows which also brings full support for new [MiniMax H3](https://vladmandic.github.io/sdnext-docs/MiniMax) and [LTXVideo-2.5](https://vladmandic.github.io/sdnext-docs/LTX)  
@@ -11,12 +11,13 @@ and improves general video processing with flexible video upscaling, updated int
 - New [group offload](https://vladmandic.github.io/sdnext-docs/Offload/#group) option which is more aggressive than the default balanced offload
 - Extended model support for [Nunchaku-Lite](https://github.com/rootonchair/nunchaku-lite) engine
 - A lot of [SDNQ](https://github.com/Disty0/sdnq) *quantization and attention* optimizations and features
+- Updated [SD.Next Launcher](https://github.com/vladmandic/sdnext-launcher)
 
 Plus quite a lot more, see full [changelog](https://github.com/vladmandic/automatic/blob/dev/CHANGELOG.md) for details!
 
 [Home](https://vladmandic.github.io/sdnext/) | [ChangeLog](https://github.com/vladmandic/automatic/blob/master/CHANGELOG.md) | [Docs](https://vladmandic.github.io/sdnext-docs/) | [Discord](https://discord.com/invite/sd-next-federal-batch-inspectors-1101998836328697867) | [Sponsor](https://github.com/sponsors/vladmandic)  
 
-## Details for 2026-08-21
+## Details for 2026-08-22
 
 - **Models**
   - [MiniMax H3](https://huggingface.co/MiniMaxAI/MiniMax-H3) available in *base* and *ref* variants  
@@ -90,6 +91,7 @@ Plus quite a lot more, see full [changelog](https://github.com/vladmandic/automa
   - hf: init hf env variables before gradio load
   - lora: skip init and rebuild offload state
   - lora: keep network multiplier on change
+  - lora: handle networks when delta does not fit
   - auth: improve handling of hf auth
   - load: improve pipeline detection for non-cached models
   - offload: cleanup alt offload codepaths
@@ -112,6 +114,7 @@ Plus quite a lot more, see full [changelog](https://github.com/vladmandic/automa
   - api: process
   - api: auth via remote-ip
   - krea2: fallback to base pipeline/transformer for nunchaku-lite
+  - torchsde: handle obsolete dependency
 
 ## Update for 2026-08-07
 

TODO.md

@@ -12,35 +12,30 @@
 
 ## Features
 
+### Assigned
+
+- Object clear remover for Kanvas: [Object clear](https://huggingface.co/jixin0101/ObjectClear), @vladmandic
+- OpenAI API interface for image generation, @vladmandic
+- Lightweight scheduler/queue manager, @vladmandic
+- Integrate natural language image search: [ImageDB](https://github.com/vladmandic/imagedb), @vladmandic
+- Support cloud providers, @CalamitousFelicitousness
+
 ### Roadmap
 
-- Video upscaling: nvidia-vfx, ltx-upscaler, etc.
+- Video upscaling: LTX-Upscaler
 - Video capabilities to processing tab, add RIFE, upscaling (once available)
-- Object clear remover for Kanvas: [Object clear](https://huggingface.co/jixin0101/ObjectClear)
-- OpenAI API interface for image generation
-- Lightweight scheduler/queue manager
 - Distraction-free UI mode with prompt-only, chat-based interface
 - Revisit transformer caching for modular pipelines
 - Revisit guidance for modular pipelines
 - Implement modular for some image models
 - Video models: support finetunes
-
-### Assigned
-
-- Support cloud providers, @CalamitousFelicitousness
-
-### Unassigned
-
 - Incorporate [prompting guides](https://github.com/CalamitousFelicitousness/ai-prompting-guides)
-- Video models: add to Reference
+- Video models: use Networks/Reference instead of custom
 - UI Lite vs Expert mode
-- Auto handle scheduler `prediction_type`
-- JSON image metadata
 - Expand custom VAE support
 - Refactor: remove obsolete code:
   - Remove `olive-ai`
-- Integrate natural language image search
-  - [ImageDB](https://github.com/vladmandic/imagedb)
+
 
 ### OnHold
 
@@ -48,11 +43,13 @@
 - [QuantFunc](https://huggingface.co/QuantFunc/Klein-9B-Series): once its released as sdk
 - LoRA add OMI format support for SD35/FLUX.1
 - Remote Text-Encoder support, sidelined for the moment
+- Auto handle scheduler `prediction_type`
 - Multi-user support
 - Settings profile manager
 - TensorRT acceleration
 - Cache models in memory
 - Unify *huggingface* and *diffusers* model folders
+- JSON image metadata
 
 ### Modular
 

modules/sd_samplers_diffusers.py

@@ -3,6 +3,7 @@ import re
 import copy
 import inspect
 import diffusers
+from installer import install
 from modules import shared, errors
 from modules.logger import log
 from modules.sd_hijack_schedulers import init_hijack, hijack_unipc, attach_scale_noise_if_missing # pylint: disable=unused-import
@@ -20,13 +21,11 @@ hijack_unipc()
 try:
     from diffusers import (
         CMStochasticIterativeScheduler,
-        CosineDPMSolverMultistepScheduler,
         DDIMScheduler,
         DDPMScheduler,
         DEISMultistepScheduler,
         DPMSolverMultistepInverseScheduler,
         DPMSolverMultistepScheduler,
-        DPMSolverSDEScheduler,
         DPMSolverSinglestepScheduler,
         EDMDPMSolverMultistepScheduler,
         EDMEulerScheduler,
@@ -54,6 +53,20 @@ except Exception as e:
     if os.environ.get('SD_SAMPLER_DEBUG', None) is not None:
         errors.display(e, 'Samplers')
 
+try:
+    install('trampoline==0.1.2', 'trampoline', quiet=True, no_deps=True)
+    install('torchsde==0.2.6', 'torchsde', quiet=True, no_deps=True)
+    from diffusers import (
+        DPMSolverSDEScheduler,
+        CosineDPMSolverMultistepScheduler,
+    )
+except Exception as e:
+    DPMSolverSDEScheduler = None
+    CosineDPMSolverMultistepScheduler = None
+    log.error(f'Sampler import: version={diffusers.__version__} error: {e}')
+    if os.environ.get('SD_SAMPLER_DEBUG', None) is not None:
+        errors.display(e, 'Samplers')
+
 # SD.Next Schedulers
 try:
     # from modules.schedulers.scheduler_tcd import TCDScheduler # pylint: disable=ungrouped-imports