New Release 01-15-2024

[vladmandic]

Update for 2025-01-15

Highlights for 2025-01-15

Two weeks since last release, time for update!
This time a bit shorter highligh reel as this is primarily a service release, but still there is more than few updates
(actually, there are ~60 commits, so its not that tiny)

*What's New?"

• Large Wiki/Docs updates
• New models: Allegro Video, new pipelines: PixelSmith, updates: Hunyuan-Video, LTX-Video, Sana 4k
• New version for ZLUDA
• New features in Detailer, XYZ grid, Sysinfo, Logging, Schedulers, Video save/create
• And a tons of hotfixes...

Details for 2025-01-15

• Wiki/Docs:
  • updated: Detailer, Install, Update, Debug, Control-HowTo, ZLUDA
• Allegro Video
  • optimizations: full offload and quantization support
  • reference values: width 1280 height 720 frames 88 steps 100 guidance 7.5
  • note: allegro model is really sensitive to input width/height/frames/steps
    and may result in completely corrupt output if those are not within expected range
• PixelSmith
  • available for SD-XL in txt2img and img2img workflows
  • select from scripts -> pixelsmith
• Hunyuan Video LoRA support
  • example: https://huggingface.co/Cseti/HunyuanVideo-LoRA-Arcane_Jinx-v1
• LTX Video framewise decoding
  • enabled by default, allows generating longer videos with reduced memory requirements
• Sana 4k
  • new Sana variation with support of directly generating 4k images
  • simply select from networks -> models -> reference
  • tip: enable vae tiling when generating very large images
• Logging:
  • reverted enable debug by default
  • updated debug wiki
  • sort logged timers by duration
  • allow min duration env variable for timers: SD_MIN_TIMER=0.1 (default)
  • update installer messages
• Refactor:
  • refactored progress monitoring, job updates and live preview
  • improved metadata save and restore
  • startup tracing and optimizations
  • threading load locks on model loads
  • refactor native vs legacy model loader
  • video save/create
• Schedulers:
  • TDD new super-fast scheduler that can generate images in 4-8 steps
    recommended to use with TDD LoRA
• Detailer:
  • add explicit detailer prompt and negative prompt
  • add explicit detailer steps setting
  • move steps, strength, prompt, negative from settings into ui params
  • set/restore detailer metadata
  • new detailer wiki
• Preview
  • since different TAESD versions produce different results and latest is not necessarily greatest
    you can choose TAESD version in settings -> live preview
    also added is support for another finetuned version of TAESD Hybrid TinyVAE
• Video
  • all video create/save code is now unified
  • add support for video formats: GIF, PNG, MP4/MP4V, MP4/AVC1, MP4/JVT3, MKV/H264, AVI/DIVX, AVI/RGBA, MJPEG/MJPG, MPG/MPG1, AVR/AVR1
  • note: video format support is platform dependent and not all formats may be available on all platforms
  • note: avc1 and h264 need custom opencv due to oss licensing issues
• ZLUDA v3.8.7
  • new runtime compiler implementation: complex types, JIT are now available
  • fast fourier transformation is implemented
  • experimental BLASLt support via nightly build
    • set ZLUDA_NIGHTLY=1 to install nightly ZLUDA: newer torch such as 2.4.x (default) and 2.5.x are now available
    • requirements: unofficial hipBLASLt
• Other
  • XYZ Grid: add prompt search&replace options: primary, refine, detailer, all
  • SysInfo: update to collected data and benchmarks
• Fixes:
  • explict clear caches on model load
  • lock adetailer commit: #a89c01d
  • xyzgrid progress calculation
  • xyzgrid detailer
  • vae tiling use default value if not set
  • sd35 img2img
  • samplers test for scale noise before using
  • scheduler api
  • sampler create error handling
  • controlnet with hires
  • controlnet with batch count
  • apply settings skip hidden settings
  • lora diffusers method apply only once
  • lora diffusers method set prompt tags and metadata
  • flux support on-the-fly quantization for bnb of unet only
  • control restore pipeline before running hires
  • restore args after batch run
  • flux controlnet
  • zluda installer
  • control inherit parent pipe settings
  • control logging
  • hf cache folder settings
  • fluxfill should not require base model

[vladmandic]

@chain2k please either create an issue or discussion thread specific to your problem, not comment on release announcement!

[chain2k]

Sorry, I did it