Unverified50% confidenceFactExact time
The starting point of the H3 video generation speed regression has been pinpointed to ComfyUI version v0.32.0, with the related issue recorded in GitHub Issue #15665.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/30/2026
First Seen
Valid until: 11/28/2026
Sources
Related Entities
Related Claims
UnverifiedStarting from an update approximately three weeks ago, the MiniMax H3 video generation model has been running approximately 4 times slower at full resolution.67% similarUnverifiedA Reddit user successfully ran the MiniMax H3 reference-image-to-video (Ref2VID) workflow on a system with 32 GB of system memory and an NVIDIA RTX 3080 (12 GB VRAM)63% similarUnverifiedIn ComfyUI's default MiniMax H3 i2v workflow, fast audio generation can be achieved by disconnecting the load image node, setting the width and height to 32, and specifying the duration.62% similarUnverifiedAt a 30fps setting, frame 90 corresponds to the 3-second mark in the video.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/825620API
curl https://kongchang.com/api/v1/knowledge/claims/825620MCP
get_claim(id=825620)