Smart batching

[Aatricks]

This pull request introduces several improvements and bug fixes across the codebase, focusing on enhanced batch conditioning support, improved robustness when optional dependencies are missing, and better file management and user documentation. The most significant changes are summarized below.

Batch Conditioning and Compatibility Improvements:

• Updated CLIPTextEncode.encode in src/clip/Clip.py to support encoding a list of prompts, returning results in a backward-compatible format for both single and batch prompt scenarios.
• Enhanced convert_cond and calc_cond_batch in src/cond/cond.py to robustly handle both legacy and new condition formats, support batch processing with per-chunk timestep alignment, and provide optional debug logging for batch issues. [1] [2] [3]

Robustness and Fallbacks for Optional Dependencies:

• Added fallback behavior in gen_stable_fast_config and ApplyStableFastUnet.apply_stable_fast in src/StableFast/StableFast.py to gracefully handle missing optional sfast dependencies, ensuring the application continues to function with conservative defaults and clear logging. [1] [2]

File Management and Metadata:

• Modified save_images in src/FileManaging/ImageSaver.py to return the actual subfolder path relative to ./output and include the originally requested subfolder, improving traceability and file location for callers. [1] [2]

Documentation and User Guidance:

• Updated README.md to add a note instructing users to manually download the FLUX VAE model due to its gated status on Huggingface, clarifying setup steps for new users.

Miscellaneous:

• Minor code hygiene: imported os and logging in src/cond/cond.py for new functionality.
• Updated generate_noise in src/AutoDetailer/ADetailer.py to explicitly pass seeds_per_sample=None for improved clarity and future compatibility.