lllyasviel / lllyasviel/sd-forge-layerdiffuse
[fix] Fix for Forge Neo 2026 compatibility (type "lora" is not recognized)
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 4.1k
- Forks
- 352
- PR merge metrics
- No merged PRs in 30d
Description
Tested on Forge Neo: 2.29 (september 2026)
`unet.load_frozen_patcher()` builds LoRA patches in a raw format `("lora", [tensors])` that Forge Neo's `merge_lora_to_weight` no longer recognizes, causing a crash when using any of the SDXL layer methods
(FG_ONLY_ATTN, FG_ONLY_CONV, BG_TO_BLEND, FG_TO_BLEND, BG_BLEND_TO_FG, FG_BLEND_TO_BG):
`ValueError: "diffusion_model.output_blocks.2.1.transformer_blocks.9.attn2.to_v.weight" of type "lora" is not recognized...`
Fix:
~~sd-forge-layerdiffuse (for Forge Neo 2026).zip~~
update: fixed the errors for "SDXL From Foreground to Blending" and some others.
[sd-forge-layerdiffuse (for Forge Neo 2026) fix 2.zip](https://github.com/user-attachments/files/32353176/sd-forge-layerdiffuse.for.Forge.Neo.2026.fix.2.zip)
Too many edits to the files to list them here...
SD1.5 modes still don't work though.
update: fixed the errors in SD1.5 modes
[sd-forge-layerdiffuse (for Forge Neo 2026) fix 3.zip](https://github.com/user-attachments/files/32383696/sd-forge-layerdiffuse.for.Forge.Neo.2026.fix.3.zip)
Posting it here in case it helps someone, or someone wants to pick it up properly and start a fork. I won't fork it myself because I am not a python dev, most of the job for this fix was done by Claude AI.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start at unet.load_frozen_patcher() and inspect how raw ("lora", [tensors]) patches reach Forge Neo's merge_lora_to_weight. Reproduce the failure with the listed SDXL and SD1.5 modes, then compare behavior with the attached fix archives. Done means all listed modes no longer raise the unrecognized-type ValueError.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 38/100