🎬 Joy-LTX 2.5 — one prompt, one take, picture + sound

JoyAI-Echo's performance on LTX-2.5's engine. The Echo video attention / feed-forward delta on the official LTX-2.5 dev transformer with the official distilled LoRA baked in at 0.5 — few-step, joint audio + video, lip-sync and expression from Echo, any length on your own card. This demo renders one take with the same Multishot sampler the ComfyUI canvases use (two passes, x2 latent upscale).

Models: GGUF (RTX 30/40) · comfy-native (RTX 50) · DEV merges · x Z-Image texture graft (GGUF) - Workflows: ComfyUI-JoyLTX25. LTX-2.x Community License; generated content is machine-generated.

4 12
render size (pass 1)
model
0.6 1.2

Tips: say the sounds you want (LTX invents drones for anything left unsaid), say the accent, keep the camera still or slow, quote the line. Multishot (seamless joins, identity across cuts, refs by name) lives in the ComfyUI canvases.