JoyAI-Echo x LTX-2.5 (echoVid) - DEV merges
Bring your own distill LoRA. These are the LTX-2.5 dev transformer with JoyAI-Echo's video
attention / feed-forward delta transplanted in (070T30 = 0.7 x delta, 0.3 x on the modulation
tables; 100T50 = 1.0 x / 0.5 x) and NOTHING baked on top - no distillation. Use them like LTX-2.5 dev:
apply the official ltx-2.5-22b-distilled-lora-450 (Lightricks/LTX-2.5, loras/) with a
Load LoRA (model only) node at the strength you like (0.4-0.6 is where the picture holds its natural
grade; 1.0 is fully distilled), or any other distill / style LoRA, then run the distilled 8-step
schedule. Or run them slow, as a dev model (25-30 steps, cfg ~3).
The ready-made few-step files (this delta with the official distill LoRA baked at 0.5) are in joyai-echo-ltx25-echoVid-gguf (RTX 30/40) and joyai-echo-ltx25-echoVid-comfy-native (RTX 50). Workflows + nodes: https://github.com/jlucasmcrell/ComfyUI-JoyLTX25
Try it live: https://huggingface.co/spaces/joeygambino/joy-ltx-25 (one take, ZeroGPU)
What it looks like
Rendered with the files on this page (070T30, distilled LoRA baked at 0.5), the ComfyUI-JoyLTX25 canvases, 8 steps at cfg 1. Sound is generated with the picture, in the same pass - turn it on.
Three shots joined into one take
Multishot, 3 x 8 s at 1280x736, AV-extend joins - the speech and the room carry across both joins with no reference photo attached.
Beach, hard sun
10 s, single generation, picture and sound together.
Wet neon street
10 s, single generation. Reflections and rain with a voice over them.
Snow, flat overcast
10 s, single generation. The grade holds in high key - the failure mode of the first build.
Files
| file | GB | notes |
|---|---|---|
LTX25dev-echoVid-<dose>-DiT-bf16.safetensors |
42.0 | the master; needs a card that streams 42 GB (or offload) |
LTX25dev-echoVid-<dose>-DiT-comfy-fp8.safetensors |
~21 | comfy fp8_e4m3fn scaled; stock Load Diffusion Model |
LTX25dev-echoVid-<dose>-DiT-comfy-int8.safetensors |
21.5 | comfy int8 (convrot); fastest on RTX 50, works on RTX 30/40 (needs ComfyUI 0.32+) |
Both doses: 070T30 and 100T50 in the file name. LoRA-on-int8 works in stock ComfyUI (it is how
Lightricks ship LTX-2.5 dev + distilled LoRA).
Credits
JoyAI-Echo by JD (jdopensource/JoyAI-Echo). LTX-2.5 by Lightricks. Merge and files by me. Licensed under the LTX-2.x Community License (inherited from both parents).