LTX-2 handler cannot find video output — 'images' key not checked

[arthurtravers]

Bug Description

LTX-2 I2V (and likely T2V) jobs always return {"error": "비디오를를 찾을 수 없습니다."} even though the ComfyUI workflow executes successfully.

The workflow completes fine (logs show Prompt executed in 17.25 seconds), but the handler never finds the video output.

Root Cause

The LTX-2 workflows (ltx2_i2v.json, ltx2_t2v.json) use ComfyUI's built-in SaveVideo node (node 75), which internally extends PreviewVideo.

PreviewVideo.as_dict() returns:

{"images": self.values, "animated": (True,)}

The output key is "images".

However, the get_videos() function in handler.py only checks for "gifs" and "videos":

if 'gifs' in node_output:
    video_list = node_output['gifs']
elif 'videos' in node_output:
    video_list = node_output['videos']

Since "images" is never checked, the handler always falls through to the error return.

Why WAN branches work

The WAN workflows use VHS_VideoCombine (from ComfyUI-VideoHelperSuite), which outputs under the "gifs" key — matching what the handler expects.

Branch	Output Node	History Key	Handler Checks	Works?
main (WAN)	VHS_VideoCombine	gifs	gifs	✅
ksampler (WAN)	VHS_VideoCombine	gifs	gifs	✅
LTX v1.3 blackwell	SaveVideo	images	gifs, videos	❌

Suggested Fix

Add an elif for the "images" key in the LTX handler's get_videos():

if 'gifs' in node_output:
    video_list = node_output['gifs']
elif 'videos' in node_output:
    video_list = node_output['videos']
elif 'images' in node_output:
    video_list = node_output['images']

Alternatively, install ComfyUI-VideoHelperSuite in the LTX Docker image and replace SaveVideo with VHS_VideoCombine in the workflow JSONs (consistent with the WAN branches).

Environment

• Branch: generate_video_ltx_v1.3_blackwell
• Release: https://github.com/wlsdml1114/generate_video/releases/tag/generate_video_ltx_v1.3_blackwell
• ComfyUI: latest main (unpinned in Dockerfile)
• Checkpoint: ltx-2-19b-dev-fp8.safetensors (26GB, verified not corrupt)