Fix: reload base model before HiresFix + stability & UX improvements (Flux2, VAE, ADetailer, settings history) #19
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Reverted to original: - Attention.py: removed SDXL-specific code and mask param addition - SDXLModel.py: restored unconditional HiDiffusion application - upscale.py: restored bislerp implementation - Clip.py: restored original fallback behavior - cond.py: removed extra pooled_output from model_conds - ModelBase.py: restored original sdxl_pooled function ADetailer fixes remain intact: - crop_cond with SDXL size conditioning - noise_mask_feather=0 (prevents DifferentialDiffusion mask destruction) - SDXL-aware guide_size/max_size/denoise - Original positive conditioning passed through pipeline
… for VAE compatibility
…ting - Fix VAE encode GPU->CPU race condition (non_blocking=True -> blocking transfer) matching the pattern already used in decode(). Without this fix, VAE-encoded latents become all zeros after refiner model swap. - Ensure ADetailer always uses base model, not refiner, in all Pipeline methods (run, run_batched, run_img2img, run_controlnet). Refiner model is not suited for text-guided crop enhancement. - Remove diagnostic print statements from Adetailer.py
…ecks; add tests for tokenizer-only case
…ithout diffusion submodule
…ing tiled outputs; add tests
- Added LD_MAX_IMAGES_PER_GROUP to control maximum images processed in a single pipeline run. - Enhanced GenerationBuffer to chunk requests exceeding LD_MAX_IMAGES_PER_GROUP and ImageSaver.MAX_IMAGES_PER_SAVE. - Updated AutoHDR to handle RGBA images and preserve alpha channels. - Improved ImageSaver to respect runtime configuration for MAX_IMAGES_PER_SAVE via environment variable. - Introduced comprehensive tests for chunking behavior, ensuring proper handling of large requests and image saving limits. - Added telemetry to report max_images_per_group in server status.
… mismatches input; add RoPE seq-length assert and tests
…RoPE feature dimension check and padding adjustments in Flux2 model
…endpoints for last seed and settings history; update UI components to utilize new features
…duplicates, add fixtures, add SettingsStore
… at bottom of GenerationSettings
…runtime crash on image display
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Ensure HiresFix runs against the base model after a refiner swap and add a collection of stability fixes, frontend settings/history features, and comprehensive tests to prevent regressions.
Key changes by theme
Pipeline / HiresFix
VAE & ADetailer (stability)
VariationalAE.encode()(use blocking transfer). (VariationalAE.py)Flux2 / Latent handling
src/NeuralNetwork/flux2/*, Latent.py)Downloader & File validation
Frontend: settings history & metadata import
Docs & installation
lcms2installation for AutoHDR color transforms. (docs/*.md)Tests / Test-restructure
tests/**)