My approach: I generate everything in "Fast" mode first to quickly iterate and find the best results, then I selectively upscale only the videos worth keeping by reusing the same seed.<h2 id="generation-modalities">Generation Modalities</h2><ul><li>📝➡️🎥 Text-to-Video Create completely new videos from scratch using text prompts.</li><li>🖼️➡️🎥 Image-to-Video Animate static reference images using text prompts.</li><li>🖼️🖼️➡️🎥 First-Last-Frame-to-Video Generates a coherent video sequence that bridges a defined starting and ending image.</li><li>🎥➡️🎥 Video-to-Video Generates synchronized audio and speech driven by video visuals. Offers options to pass through or entirely regenerate the source video.</li></ul><h2 id="audio-input-settings">Audio Input Settings</h2><ul><li>🔇 No Audio Input No external audio file used. The AI generates new audio based on your text prompt.</li><li>🔊 Audio Input Upload an existing voice or music file to drive the animation and lip-sync.</li></ul><h2 id="vram-optimization-and-long-videos">🚀 VRAM Optimization &amp; Long Videos</h2><h3 id="chunking-settings-for-longer-videos">⚙️ Chunking Settings for Longer Videos</h3>Adjust ffn_chunks based on your video length (at 24fps):<ul><li>10 seconds (240 frames): ffn_chunks=1-2</li><li>15 seconds (360 frames): ffn_chunks=2-4</li><li>20 seconds (480 frames): ffn_chunks=4-6</li><li>25 seconds (600 frames): ffn_chunks=8-10</li><li>33 seconds (800 frames): ffn_chunks=12-16</li></ul>If you encounter OOM (Out of Memory) errors:<ul><li>Increase ffn_chunks value</li><li>Reduce resolution slightly</li></ul>

LTX2 All-in-One ComfyUI Workflow

v2.46 — Negative Promt + PROMPT ENHANCERv2.45 — Small fixes, plus new model links for distilled-1.1 with improved aesthetics and better audio over v1.0, and gemma3-12B-hereticx-sikaworld-ltx-2, a fully uncensored LTX-2 text encoder, higher embedding variance, stronger motion output, and BF16-protected layers 0–1, 44–47, all LayerNorms, and biases for better spatial stability, facial symmetry, and anatomical coherence. v2.43 — FLF2V: added Ignore first frame, use end frame onlyv2.41 — Multiple LoRAs loader added * Added an optional 2x Frame Interpolation during the upscaling process.* implemented the Clownshark Sampler for better results.* Integrated RTX Video Super Resolution Node for fast video upscaling in seconds.

Files with SHA256: cf2e511d4d8ec43ff2757a44ee96dc010887e08c7879f2adb5528c73a41ebf6a