⭐ Star LTXV 2.5 All-in-One (BETA)

The name says "All-in-One" and it means it. Star LTXV 2.5 All-in-One (BETA) is the StarNodes port of the official LTX 2.5 workflow templates, compressed into one node that handles text-to-video, image-to-video, image-plus-audio, first-and-last-frame keyframing, and even an audio-only mode - plus the two-pass render pipeline, model/LoRA/CLIP/VAE caching, and internal sound processing. If you want to run LTX 2.5, this is the fastest way from zero to a finished clip that the pack offers.

It's in ⭐StarNodes/Video, and it's the pack's newest and most ambitious node - which is also why the BETA label is honest, not decorative.

How it works - the modes

The mode dropdown is where you start, and it decides the whole graph:

• text_to_video - prompt only.
• image_to_video - connect first_frame.
• image_audio_to_video - connect first_frame and an audio file; the audio is trimmed to the video length and preserved.
• first_last_frame_to_video - connect first_frame and last_frame; runs a single full-resolution pass with both frames added as keyframe guides (LTXVAddGuide, same approach as the official FLF2V template). This is the one mode that skips the two-pass structure.
• audio_only - no real video at all; a single 30-step pass at 64×64 whose only output that matters is the audio. Yes, that's as odd as it sounds, but it's the official workflow's approach and it works.

The main modes (T2V / I2V / I2V+audio) use the two-pass pattern: pass 1 at half resolution, a 2× latent upscale in between, pass 2 at full resolution for the refine. LTX 2.5 uses a single text encoder (a ltxv-type Gemma encoder), which is a real simplification versus the older two-encoder setups.

The inputs that matter

• base_model, clip_1, vae, audio_vae, upscale_model - the internal loaders. Defaults point at LTX 2.5 files (ltx-2.5-22b-distilled-transformer-comfy-int8-convrot, the gemma4-12b encoder, ltx-2.5-video-vae-conv-bf16, etc.) from the standard model folders. Each is cached and only re-loaded when you change the selection.
• sigma_preset - the baked 8/12/16-step schedules from the original workflow, plus plain 20/30/40/50-step sampler options, or custom with custom_sigmas_pass1.
• cfg (default 1.0) and the two sampler dropdowns - both passes.
• sound_settings - optional; plug in the pack's Star Video Sound Enricher Option bundle and the audio output gets cleaned and enriched (at least 44.1 kHz, never downsampled) before it leaves the node.
• override_audio - the subtle one. By default (in T2V/I2V/FLF modes) a connected audio input is ignored and the model-generated audio goes to the output; flip it on to pass your own audio through instead. In image_audio_to_video mode the connected audio always passes through.

Outputs

images, audio, and frame_rate - the same trio as the LTX 2.3 node, and the audio output is the part most people underestimate. LTX 2.5 generates synchronized audio, and this node decodes it from the first (high-step) pass, which is the version with the best audio quality.

Installing it

ComfyUI Manager → search Starnodes → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/Starnodes2024/ComfyUI_StarNodes
cd ComfyUI_StarNodes
pip install -r requirements.txt

The honest BETA warning

Take the label seriously. LTX 2.5 is brand-new territory (the pack's own README calls this node the newest thing in the pack), and this node wraps a huge amount of machinery - five internal model loaders, two-pass sampling, keyframe guides, an audio path, vendored fallback code for older ComfyUI. When it works it's remarkable; when it doesn't, the single-node black box makes debugging harder than a normal graph would be. Your first three problems will be: (1) missing model files, because the dropdowns only populate from the exact models/ subfolders; (2) an outdated ComfyUI core - this node leans on recent comfy_extras.nodes_lt internals, so update ComfyUI before blaming the pack; (3) VRAM, because a 22B transformer plus a 12B text encoder is a serious ask regardless of how convenient the node is. Update everything, check the console, and budget your card's memory - then enjoy what is genuinely the easiest way to run 2.5.