Release Notes

Updates to ltx workflow and LTX 2.3 model releases.

2026-09-19

Creative Lab LoRAs — 24 Official Lightricks Adapters for LTX 2.3 & 2.5

+New: 9 LTX 2.5 Creative Lab LoRAs released 2026-09-08..15 — Ingredients (reference-sheet control), Clean Plate, Deblur, Decompression, Colorization, Water Simulation, Day-to-Night, Cinemagraph, and Slow Motion Control (I2V speed knob, ships an official ComfyUI workflow). The LTX 2.3 files don't work on 2.5; these are retrained ports
+New: 15 LTX 2.3 Creative Lab LoRAs we'd missed (published June–July 2026, each in its own gated repo) — Ingredients, In/Outpainting, Dub-It, Deblur, Decompression, Colorization, Water Simulation, Day-to-Night, Relight (sun direction), Clean Plate, Instant Shave, Cross-Eyed, Pixel Spatial Upscaler ×2 / ×4, Cinemagraph, Foley V2A (video-to-audio)
+Each entry lists the trigger word, recommended LoRA strength, trained resolution bucket, and out-of-scope cases taken from the Lightricks model card
+Models page: new 'Creative Lab LoRAs' groups for both families; LTX 2.5 setup guide gets a matching section
+Update check now scans every Lightricks repo, not just the main LTX-2.3 / LTX-2.5 ones — that's how this batch slipped through

2026-09-09

+New: LTX 2.3 Pruna VAE (Kijai, 1.33 GB) — drop-in ComfyUI conversion of Pruna AI's PrunaVAED decoder; ~1.7× faster decode and ~50% lower peak decode VRAM, encoder and latents unchanged. Natively supported in current ComfyUI builds
+New: Gemma 4 12B GGUF Q4_K_M (8.41 GB) and Q2_K (5.96 GB) text encoders for LTX 2.5 (elix3r) — 1.1 GB and 3.5 GB smaller than the Q5_K_M that was previously the only GGUF encoder option
+LTX 2.5 guides: encoder tables and 16GB recommendations updated for the new quants
+Online generator: page copy now reflects LTX 2.5 Fast/Pro alongside LTX 2.3

2026-08-26

+New: LTX 2.5 official transformer, Gemma 4 text encoder, VAEs, LoRA, upscalers, and duration-head patch — all from the gated Lightricks/LTX-2.5 repo
+New: Community GGUF quants (Abiray, elix3r) for the only sub-24GB LTX 2.5 path
+New: /guide/ltx-2-5-comfyui setup guide and /guide/ltx-2-5-vram-requirements VRAM guide
+VRAM Adapter: added an LTX 2.5 / LTX 2.3 family toggle
+Note: LTX 2.5's official files need 24GB+ VRAM combined (Gemma 4 encoder alone is 15.37GB+) — 16GB cards need the GGUF path

2026-06-15

+New: LTX 2.3 Distilled 1.1 INT8 convrot (Kijai, ~21.5 GB) — rotation-based INT8 that runs on Ampere/Turing INT8 tensor cores, where FP8 matmul is unavailable and MXFP8 only gets BF16-fallback speed
+New: LTX 2.3 Dev INT8 convrot (Kijai, ~21.5 GB) — same low-VRAM speedup for the dev/quality + LoRA path

2026-06-03

+New: IC-LoRA Union Control (Canny + Depth combined) — official Lightricks release
+New: IC-LoRA Motion Track Control — guide motion with sparse point trajectories
+New: IC-LoRA HDR + scene embeddings — 16-bit HDR generation, SDR→HDR via LogC3
+New: IC-LoRA LipDub — joint audio-visual lip-dubbing (JustDubIt research)
+New: LTX 2.3 Dev NVFP4 (21.7 GB) — official quantization for Blackwell / RTX 50xx
+New: Gemma 3 12B IT text encoder (FP8 13.2 GB + full ~24 GB) from Comfy-Org — required for all ComfyUI workflows
+New: Kijai dash-naming FP8 input-scaled variant (alias)

2026-06-01

+New: LTX 2.3 OmniNFT RL LoRA (Kijai, 617 MB) — reinforcement-learning LoRA that improves audio-video sync, lip-sync, and temporal stability (JavisBench DeSync 0.569 → 0.269)

2026-05-27

+New: LTX23 Audio VAE (Kijai) — enables audio-conditioned video generation workflows
+New: LTX23 Video VAE BF16 and Text Projection BF16 components (Kijai)
+New: Spatial Upscaler x2 v1.1 — updated upscaler from Lightricks
+New: Distilled LoRA rank-384 v1.0 (Official) and Dynamic LoRA rank-105 (Kijai)
+New: Distilled 1.1 BF16 and MXFP8 transformer-only variants (Kijai)
+New: Dev BF16, FP8 Scaled, and MXFP8 transformer-only variants (Kijai)
+New: Distilled v1.0 BF16, FP8 v1/v2/Scaled, and MXFP8 transformer-only variants (Kijai)

2026-04-27

+New Models: Added official Lightricks FP8 variants for dev and distilled models (29.1GB / 29.5GB)
+New Models: Added Spatial Upscaler x1.5 and Temporal Upscaler x2 for two-stage pipelines
+New Models: Added official LoRA rank-384 v1.1 from Lightricks
+Pricing: Added one-time payment option ($4.99 for 30 days Pro access, no auto-renewal)
+Pricing: Updated currency display to USD ($)
+Video Tutorials: Added 3 Bilibili video tutorials to the Guide page

2026-04-21

+Pro Plan: Subscribe for ¥2.99/month to unlock unlimited prompt enhancements and workflow generations (free users get 3 uses per day)
+Content Library: Added 13+ new resources including tutorials, community discussions, showcase examples, research papers, and tool guides
+Navigation: Renamed 'Models' to 'Downloads' for clarity
+VRAM Adapter: Added quick download button to jump directly to model downloads
+Payment Policy: Clear no-refund policy displayed on pricing and success pages
+Workflows Page: Removed subscription card for cleaner layout - visit Pricing page to subscribe

2026-04-16

+Blog: Published 4 new articles (taeltx2_3 guide, LTX 2.3 release, model comparison, LoRA training deep dive)
+Nav: Added Resources dropdown with Blog and Release Notes sub-items
+Nav: Mobile responsive hamburger menu
+Nav: Logo now links to homepage with gradient text
+Footer: Added to all pages, Feedback link moved to footer
+Models: Added Quick Start 3-step guide to Models page
+Favicon: Replaced black icon with purple gradient to match site branding

2026-04-13

+Homepage: Replaced model grid with two-button entry cards (Models / Workflows)
+Models page: Added 'How to Choose' official guide (Distilled vs Dev, FP8 vs bf16)
+Models page: Separated 16GB/24GB/32GB models with VRAM-specific recommendations
+Workflows page: Added 'How to Choose' decision guide with 4 scenario cards
+VRAM Adapter: Separated checkpoints from required files (VAE always shown)
+All pages: Unified two-button pattern (Download / How to Choose)

2026-04-08

2026-04-01