olivv-cs-ComfyUI-FunPack

Files

T

Claude a13f037f72 H3 training: modulate blocks out of place so the backward pass stops keeping every block's recompute alive

Measured on a tiny H3: backward growth was ~75 MB per block, now flat (61-88 MB at 2, 8 and 16 blocks).
The in-place mul_/addcmul_ on slices of the stream in the block helpers were the cause.
UNTESTED on the real weights.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MpnF7GwNGMbngU11wjym7P

2026-10-17 19:47:34 +00:00

conditioning

fix: Shot Camera only reports 'changed' when it will rewrite, so the H3 prompt/reference encode (~7s) is cached between runs

2026-10-17 01:53:46 +03:00

latent

fix: text-only runs do not hand the grey stand-in to the enhancer; the encoder pick is text so a vanished file cannot stop a run with Enhance off; truer retain_first, montage and upscaler hints (attacker round)

2026-10-15 16:59:48 +03:00

loaders

LoRA training add-on: sets, tagger, trainer node on ComfyUI's Train LoRA, H3 support

2026-10-17 15:05:57 +00:00

media

feat: MiniMax H3 Reference to Video works text-only; with no picture a plain stand-in of the Width/Height shape feeds the size chain (found on a real GPU)