Skip to content

Load FreeVideo for offline environment packages; wider attention head groups on Linux GPUs with memory to spare (0.3.20) - #137

Merged
KBRASK merged 1 commit into
mainfrom
release/0.3.20-public
Oct 8, 2026
Merged

KBRASK merged 1 commit into
mainfrom
release/0.3.20-public

Conversation

@KBRASK

@KBRASK KBRASK commented Oct 8, 2026

Copy link
Copy Markdown
Collaborator

Offline installations imported from the environment package now open FreeVideo in ComfyUI. The package carries ComfyUI without custom_nodes/FreeVideo, and connecting an offline installation never installed the node, so ComfyUI started without FreeVideo and the launch stopped at "could not load FreeVideo". Connecting now installs the node from the launcher source the installation was imported with, as every other installation does; a full bundle, whose source is the node itself, is unchanged. A retry restarts the ComfyUI that an earlier attempt started instead of asking for a restart, and both messages get their own explanation.

Offline installations in a longer folder path now also get their desktop and Start menu shortcuts. Windows cannot create a shortcut to a program whose path is 260 characters or longer, and the folders of an offline installation add about 210 characters to the chosen folder before the launcher copy that the shortcut starts. When that path would be too long, the copy is kept in the launcher folder (%LOCALAPPDATA%\FreeVideo-launcher\application), and version cleanup removes copies there that no shortcut, running launcher or installation record uses.

On Linux, when the GPU has memory to spare, attention now runs 28 or all 56 heads at a time instead of 16. Across a planner comparison of systems, architectures, VRAM, RAM, video sizes and precisions, only the head group changes: resident layers, pinned memory, streaming and output placement stay as they are, and feed-forward tiles stay at 2048 rows. Video decoding on Linux also uses its FP16 cache on every GPU, with identical output. Windows and smaller GPUs keep their current settings.

Tested on a Linux RTX PRO 6000 with complete two-pass 1344×768 10-second requests, against 16 heads: sampling took 71.4 s instead of 80.0 s with the whole GPU, 73.1 s instead of 80.0 s at a 20 GB budget and 78.1 s instead of 83.1 s at 16 GB; FP8 69.7 s instead of 76.5 s. Decoding took 15.4 s instead of 19.2 s with 4.2 GB less memory. The same seed can give a slightly different video on these GPUs, as it already does between GPUs with different memory. The offline node installation was checked through the launcher on a Windows test machine, where ComfyUI loaded the FreeVideo nodes and the launcher reached "Open FreeVideo", and the shortcut path limit was reproduced and checked with the Windows shortcut API.

… groups on Linux GPUs with memory to spare (0.3.20)
@KBRASK
KBRASK merged commit ebc94ec into main Oct 8, 2026
4 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant