comfyui-mcp/packs/ltx-2.3-img2vid/manifest.yaml

Local-first, agent-native control plane for ComfyUI — MCP server + autonomous sidebar agent that drives your live graph in natural language on ANY LLM: Claude/ChatGPT/Gemini on your subscription (no API key), free local models via Ollama (fully offline),

# LTX-2.3 img2vid — official Comfy-Org two-stage image-to-video template.
#
# Workflow: Comfy-Org's LTX-2.3 I2V template (video_ltx2_3_i2v). Render-verified
# sharp at 1280x704 with synchronized 48kHz stereo audio. All nodes are CORE
# ComfyUI (comfy_extras.*) — no custom node packs required beyond ComfyUI-Manager.
#
# Pipeline: LTXAVTextEncoderLoader (gemma + checkpoint, drives video + audio/voice)
# + gemma abliterated LoRA on the CLIP (prompt/eyes accuracy) + dynamic distilled
# LoRA + two-stage spatial upscale. Load a first-frame image into LoadImage; set
# the prompt to describe the motion/scene. Flip the "Switch to Text to Video?"
# boolean for pure T2V (see ltx-2.3-txt2vid).

pip:
  - imageio-ffmpeg   # CreateVideo/SaveVideo need an ffmpeg for the mux

# All LTX 2.3 nodes are native to current ComfyUI (comfy_extras); this pack
# needs no custom nodes. Update ComfyUI itself to a recent build.
custom_nodes: []

models:
  # --- Main checkpoint: 22B dev transformer + audio VAE (~46 GB) ---
  # fp8 (~23 GB) at Lightricks/LTX-2.3-fp8 is the VRAM-friendly alternative.
  - url: https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-22b-dev.safetensors
    local_path: checkpoints/ltx-2.3-22b-dev.safetensors

  # --- Gemma text encoder (fp8_scaled; powers video + audio/voice) ---
  - url: https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/text_encoders/gemma_3_12B_it_fp8_scaled.safetensors
    local_path: text_encoders/gemma_3_12B_it_fp8_scaled.safetensors

  # --- Distilled speed LoRA (dynamic rank-111) @ 0.5 ---
  - url: https://huggingface.co/Comfy-Org/ltx-2.3/resolve/main/split_files/loras/ltx_2.3_22b_distilled_1.1_lora_dynamic_fro09_avg_rank_111_bf16.safetensors
    local_path: loras/ltx_2.3_22b_distilled_1.1_lora_dynamic_fro09_avg_rank_111_bf16.safetensors

  # --- Gemma abliterated LoRA @ 1.0 (text-encoder CLIP; prompt/eyes accuracy) ---
  - url: https://huggingface.co/Comfy-Org/ltx-2/resolve/main/split_files/loras/gemma-3-12b-it-abliterated_lora_rank64_bf16.safetensors
    local_path: loras/gemma-3-12b-it-abliterated_lora_rank64_bf16.safetensors

  # --- Spatial upscaler (stage-2 LTXVLatentUpsampler) ---
  - url: https://huggingface.co/Lightricks/LTX-2.3/resolve/main/ltx-2.3-spatial-upscaler-x2-1.1.safetensors
    local_path: latent_upscale_models/ltx-2.3-spatial-upscaler-x2-1.1.safetensors