@@ -79,28 +79,6 @@ class ForwardBatch:
7979 image_embeds : list [torch .Tensor ] = field (default_factory = list )
8080 pil_image : torch .Tensor | PIL .Image .Image | None = None
8181 preprocessed_image : torch .Tensor | None = None
82- # LTX-2 image conditioning. Each entry is (path, frame_idx, strength)
83- # — strength 1.0 anchors the latent at frame_idx exactly to the
84- # encoded image; <1.0 mixes denoised content with the conditioning
85- # image. ``ltx2_image_crf`` controls the H.264 CRF re-encode applied
86- # to the image before VAE encoding (matches official LTX-2 quality
87- # adaptation; 0.0 disables the re-encode).
88- ltx2_images : list [tuple [str , int , float ]] | None = None
89- ltx2_image_crf : float = 33.0
90- # Stage-specific continuation latents — populated by the streaming
91- # session controller between segments. ``stage1`` is the last latent
92- # from the previous segment's stage-1 denoise (half-res); ``stage2``
93- # is the last latent from the previous segment's stage-2 refine
94- # (full-res). The image-conditioning builder reads whichever the
95- # current pass needs (``base_clean_latent is None`` -> stage1).
96- ltx2_conditioning_latent_stage1 : torch .Tensor | None = None
97- ltx2_conditioning_latent_stage2 : torch .Tensor | None = None
98- # Video-clip conditioning: list of (frame_paths, frame_idx, strength)
99- # — multi-frame VAE-encoded conditioning so the resulting latent
100- # captures temporal/motion info (used by the streaming server's
101- # mid-roll continuation flow).
102- ltx2_video_conditions : list [tuple [list [str ], int , float ]] | None = None
103-
10482 # Text inputs
10583 prompt : str | list [str ] | None = None
10684 negative_prompt : str | list [str ] | None = None
0 commit comments