xiaolibai-sys
a177ad5dbc
- FreeNoise: applied exactly once on the single-sampler path; now also applied on the dual-expert path (was zero) - TeaCache: restores eager transformer_forward when the model is torch.compiled, so block-skipping actually engages (previously silent no-op) - Compile cache: isolated to %TEMP%/bernini_r_inductor_cache (env-overridable), version-gated purge + manual purge via BerniniR_CompileModel.purge_cache / BERNINI_PURGE_COMPILE_CACHE - Removed web/guidance_ui.js frontend (guidance moved to the BerniniR_GuidanceConfig node); dropped WEB_DIRECTORY export - Stale-comment cleanup across nodes/ and utils/; refreshed README and added README_DEVELOPER - Bump version to 1.0.3
43 lines
2.1 KiB
Python
43 lines
2.1 KiB
Python
"""ComfyUI-BerniniR: Custom node package for the Bernini-R 1.3B/14B video model.
|
|
|
|
Features:
|
|
- Model loading with separate LoRA-aware torch.compile node
|
|
- Temporal context window tiling for long video generation
|
|
- Enhanced KSampler with native context window support
|
|
- Seven-mode guidance family: CFG, APG, RAAG, S2, Z2, STG_A, STG_R
|
|
- Prompt embedding with task presets, system prompts, and CLIP caching
|
|
- In-context conditioning for video editing
|
|
- Five attention backends: SageAttention 3, SageAttention, FlashAttention, xFormers, SDPA
|
|
- TeaCache acceleration and GPU<->CPU block swap
|
|
|
|
Registered nodes (19):
|
|
BerniniR_ModelLoader — Load diffusion model
|
|
BerniniR_CompileModel — Apply torch.compile after LoRA loading
|
|
BerniniR_CLIPLoader — Load Wan text encoder
|
|
BerniniR_VAELoader — Load Bernini-R VAE
|
|
BerniniR_Conditioning — Create latents + attach in-context references
|
|
BerniniR_KSampler — Enhanced sampler with context window support
|
|
BerniniR_TeaCacheArgs — TeaCache block-skipping args
|
|
BerniniR_ContextWindow — Configure temporal context windows
|
|
BerniniR_PromptEmbedding — Task-aware prompt embedding & CLIP encoding
|
|
BerniniR_VAEDecode — VAE decode with color matching
|
|
BerniniR_VAEEncode — VAE encode with tiling
|
|
BerniniR_ApplyNAG — Normalized Attention Guidance for detail enhancement
|
|
BerniniR_GuidanceStrengthSchedule — Per-step guidance scale curve
|
|
BerniniR_DualExpertSampler — High/low-noise model switching per step
|
|
BerniniR_BlockSwapArgs — Configure GPU<->CPU block swapping
|
|
BerniniR_GuidanceConfig — Select guidance mode & parameters
|
|
BerniniR_SegmentSchedule — Per-segment schedule
|
|
BerniniR_LoadLoRA — Append LoRA spec (inline merge)
|
|
BerniniR_AttentionConfig — Select attention backend
|
|
"""
|
|
|
|
import logging
|
|
|
|
from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
|
|
|
|
logger = logging.getLogger(__name__)
|
|
logger.info("[BerniniR] ComfyUI-BerniniR node package loaded successfully.")
|
|
|
|
__all__ = ["NODE_CLASS_MAPPINGS", "NODE_DISPLAY_NAME_MAPPINGS"]
|