LTX-2.3 CompanionVAE + Vocoder + Connectors9 GB instead of 43***Not a model on its own Nothing quantized***The ProblemIf you run an LTX-2.3 GGUF DiT on a 64 GB machine and ComfyUI dies while loading with <code>Windows fatal exception: access violation</code> — this is the fix.A GGUF carries only the transformer blocks. The loader still needs a <code>checkpoint_path</code> for everything else, and it materializes that WHOLE file in system RAM as owned copies — roughly 60 GB peak against a 43 GB bf16 checkpoint. On a 64 GB box that usually falls over, and it looks like a VRAM problem even though the GPU is idle.What's Inside — 9.02 GB, cut byte-for-byteEmbedding connectors + adaLN singlesThe DiT machinery GGUFs don't carry4.89 GBText embedding projectionKeeps voice and identity conditioning wired up2.31 GBVideo VAE1.45 GBVocoder0.26 GBAudio VAE0.11 GBThe Result~11 GBPeak system RAM60 GB → 11 GBBefore → AfterOutput is identical — nothing is quantized, and the config metadata rides along so the loader configures the transformer exactly as the full checkpoint would.UsagePut it wherever your checkpoints live. In <code>JoyEcho_ModelLoader</code>: your GGUF in <code>model_file</code>, this file in <code>checkpoint_path</code>. Nothing else changes.Pairing NoteCut from the e50 merge — the exact match for the e50 GGUF family. It loads with any LTX-2.3 GGUF, but the connectors and gate tensors differ between merge variants, so for a different checkpoint family, cut your own with the included one-command script (on the HuggingFace repo).Not a model on its own.There is no DiT in here — it is the other half of a GGUF.Also on HuggingFacejoeygambino/ltx23-gguf-vae-companion

Files with SHA256: 3065330be56e96132a87e639be447436897489199fbb9dea7f5c788decc7bcc8