Two ComfyUI workflows for LTX-2.5 that add first-frame/last-frame keyframe conditioning and audio-driven lipsync to the stock image-to-video template. The official 2.5 I2V workflow only conditions on a starting image and lets the model improvise where the clip ends up; these versions lock in an ending image too, so the motion travels from your chosen start frame to your chosen end frame, while your own audio clip drives the speech/singing.

Built on top of Comfy's own LTX-2.5 template nodes. No custom node packs required.

WHICH ONE SHOULD SHOULD YOU USE?

SinglePass_HQ (LTX2.5_FLF2V_Lipsync_SinglePass_HQ.json)

• One full-resolution pass

• Best quality, noticeably cleaner

• Slower: about 20 min for a 28 s clip on my machine

• Based on Comfy’s official LTX-2.5 FLF2V template, plus lipsync

• Extra model: prompt enhancer (only if you turn that optional toggle on)

TwoStage_Fast (LTX2.5_FLF2V_Lipsync_TwoStage_Fast.json)

• Low-res pass, then a 2x latent upscale and refine pass

• Good quality, slightly softer

• Faster: under 10 min for the same clip

• Based on my original lipsync workflow, repaired

• Extra model: latent spatial upscaler

Rule of thumb: use TwoStage_Fast for drafts, long clips, or lower VRAM, and SinglePass_HQ for the final render.

Usage

1. Load the workflow into ComfyUI.

2. In Load First Frame, upload your starting image.

3. In Load Last Frame, upload your ending image.

4. In Load Driving Audio, pick the audio clip (speech or song) you want the character to perform.

5. Write your prompt describing the scene and motion. For lipsync, say who is speaking or singing, and include the spoken or sung words.

6. Set duration (seconds). It must be no longer than your audio, counted from audio_start.

7. Pick the aspect ratio and size in the Resolution Selector node.

8. Queue the prompt.

REQUIREMENTS

• ComfyUI with the LTX-2.5 nodes (comfy-core, no extra custom nodes)

• LTX-2.5 model files as used by the stock LTX-2.5 template:

  • distilled transformer: ltx-2.5-22b-distilled-transformer-comfy-int8-convrot

  • video VAE and audio VAE: ltx-2.5-video-vae-bf16, ltx-2.5-audio-vae-bf16

  • Gemma 4 12B text encoder: gemma4-12b-with-proj-ltx-2.5-comfy-int8-convrot

  • TwoStage_Fast only: latent spatial upscaler, ltx-2.5-latent-spatial-upscaler-x2-bf16-1.0

  • SinglePass_HQ only: prompt enhancer model, gemma4_e2b_it_int8_convrot (only used if you turn the optional prompt enhance toggle on)

Credit

Adapted from Comfy's stock LTX-2.5 image-to-video and first/last-frame templates, using LTXVAddGuide, the same keyframe-conditioning node used in Lightricks' official first-frame/last-frame example workflows. Lipsync uses the LTX audio VAE encode path with a zero noise mask to hold the driving audio fixed.