LTX Workflow is a free resource hub for generating AI videos locally using LTX 2.3 models and ComfyUI, providing downloadable model files, ready-to-use workflow JSONs, and an interactive configurator to match your GPU hardware.
What is LTX Workflow?
LTX Workflow aggregates official and community LTX 2.3 model releases, ComfyUI workflow templates, and a VRAM-based model selector for local AI video generation. It takes text prompts or source images as input and outputs video clips (up to 161 frames, 5–8 seconds), running on ComfyUI on a local PC with an NVIDIA GPU. The platform is maintained by Lightricks (model weights) and third-party contributors like Kijai (FP8 quantizations).
Key Features
- Model Downloads — Direct links to official LTX 2.3 checkpoints (e.g.,
ltx-2.3-22b-distilled-1.1_transformer_only_fp8_scaled.safetensors), the required VAE (taeltx2_3.safetensors), audio VAE (LTX23_audio_vae_bf16.safetensors), text encoder (gemma_3_12B_it_fp4_mixed.safetensors), spatial upscalers (x1.5, x2 v1.1), temporal upscaler, and LoRA files for motion control. - VRAM Adapter — Select your GPU VRAM (16, 24, or 32 GB) to see recommended models; for 16 GB, the FP8 distilled model is suggested (requires RTX 40xx+).
- Workflow Configurator — Answer questions about VRAM, generation type (text-to-video or image-to-video), speed vs. quality, video length, orientation, resolution, frame rate, and optional upscaling; outputs a tailored model list and a download link to the matching ComfyUI workflow JSON.
- Official Workflow JSONs — Drag-and-drop workflows for T2V, I2V, two-stage upscaling, and ICLoRA motion tracking, sourced directly from Lightricks’ GitHub repository.
- Newsletter (LTX Insider) — Occasional updates on new model releases (e.g., Kijai FP8 quants, Lightricks updates, IC-LoRAs), with no fixed cadence.
- Community Showcase — Real examples and use cases from the LTX 2.3 community (implied from the earlier overview).
Who is it for?
- Content creators — generate unique video clips for social media or marketing using local AI without cloud API costs.
- Developers — integrate LTX 2.3 into applications with pre-configured workflows and technical guides.
- Hobbyists — learn AI video generation step by step with free model downloads and ready-to-use workflows.
What can you do with it?
- Text-to-video: Input a text prompt, choose a workflow, and generate a video up to 121 frames (approx. 5 seconds at 24 fps) using the FP8 distilled model on a 16GB GPU.
- Image-to-video: Provide a source image and produce a video with synchronized audio via the audio VAE and conditioning nodes; the recommended I2V workflow includes one-pass audio generation.
- Motion control: Use ICLoRA LoRAs or motion track workflows to guide object movement from a reference video or drawn motion paths.
- Two-stage upscaling: Generate at a lower resolution (e.g., 768×512) then apply the spatial upscaler x2 v1.1 for sharper output; the upscaler is under 1 GB and VRAM-light.
How does it work?
- Visit ltxworkflow.com and choose your use case (text-to-video, image-to-video, or motion control).
- Use the VRAM adapter or Workflow Configurator to select the correct model files and workflow JSON for your GPU.
- Download the recommended model files (transformer, VAE, text encoder, LoRA, upscalers as needed) and the workflow JSON.
- Open ComfyUI, drag in the workflow JSON, update any node settings as instructed (e.g., set LoRA to v1.1), and run generation.
Pricing
LTX Workflow is a free resource: all model downloads and workflow files are provided at no cost. The online image-to-video demo is also free to try (no signup required). There are no paid tiers mentioned on the page.
