tavi d6d8ebb601 feat: per-guide attention strength control in self-attention ... Implements per-guide attention attenuation via log-space additive bias in self-attention. Each guide reference tracks its own strength and optional spatial mask in conditioning metadata (guide_attention_entries). - model.py: build composite attention mask from per-guide entries - av_model.py: thread self_attention_mask through AV transformer blocks - model_base.py: pass guide_attention_entries via extra_conds - nodes_lt.py: guide entry tracking in LTXVAddGuide, cleanup in CropGuides - Submodule updated to feature/per-guide-attention-strength (cherry picked from commit 7a302cc1649fed036fe17aa89daecdb66b76705e) # Conflicts: # comfy/ldm/lightricks/av_model.py # comfy/ldm/lightricks/model.py # custom_nodes/ComfyUI-LTXVideo-Internal		2026-02-24 08:59:23 +02:00
..
vae	Reduce LTX2 VAE VRAM consumption (#12028)	2026-01-22 16:54:18 -05:00
vocoders	Support the LTXV 2 model. (#11632)	2026-01-05 01:58:59 -05:00
av_model.py	feat: per-guide attention strength control in self-attention	2026-02-24 08:59:23 +02:00
embeddings_connector.py	Fix issue loading fp8 ltxav checkpoints. (#12582)	2026-02-22 16:00:02 -05:00
latent_upsampler.py	Support the LTXV 2 model. (#11632)	2026-01-05 01:58:59 -05:00
model.py	feat: per-guide attention strength control in self-attention	2026-02-24 08:59:23 +02:00
symmetric_patchifier.py	Support the LTXV 2 model. (#11632)	2026-01-05 01:58:59 -05:00