mirror of
https://github.com/AcademiaSD/sd-forge-fluxtools-v2.git
synced 2026-11-04 23:27:24 +00:00
[GH-ISSUE #5] FluxTools-V2 causes mat1 and mat2 shapes cannot be multiplied error in FLUX1 models even after being disabled #6
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @thejohnd0e on GitHub (Apr 21, 2025).
Original GitHub issue: https://github.com/AcademiaSD/sd-forge-fluxtools-v2/issues/5
Using FluxTools-V2 with any flux1 model (tested with flux1-dev-bnb-nf4.safetensors) causes a runtime crash during sampling, with the error:
RuntimeError: mat1 and mat2 shapes cannot be multiplied (2304x128 and 64x3072)
Once the error happens, disabling FluxTools does not fix it — Forge needs to be fully restarted to recover.
The issue occurs even when Canny preprocessor works visually, and the UI loads normally.
FluxTools likely injects an incorrectly-shaped conditioning tensor (dim 128) into FLUX's img_in() or control pipeline, which expects a different shape (64). Even if FluxTools is visually off, the malformed tensor may remain in memory or model state.
webui_forge_cu124_torch24
Python 3.10.6
Torch 2.6.0+cu124
GPU: RTX 3060 (12 GB VRAM)
Model: flux1-dev-bnb-nf4.safetensors
@AcademiaSD commented on GitHub (Apr 22, 2025):
You must select the appropriate checkpoint for each tool, flux-depth-dev for the depth tool, flux-canny-dev for canny, flux-fill-dev for fill, and any flux-dev checkpoints for redux.
@AcademiaSD commented on GitHub (Apr 22, 2025):
https://www.youtube.com/watch?v=MHYSFBkF36s
@thejohnd0e commented on GitHub (Apr 24, 2025):
Yes, you're right, I was inattentive, but who reads these manuals ) Everything works now.
Thanks a lot