ComfyUI-Anima29-LegacyLoRA

A ComfyUI compatibility loader for using legacy 28-layer Anima LoRAs with 40-layer Anima 2.9B / expanded Anima checkpoints.

Older Anima LoRAs were trained for the original 28-block DiT architecture.
Anima 2.9B expands the main DiT to 40 blocks, so the original LoRA block indices no longer line up directly.

This custom node remaps the old 28-layer LoRA blocks to the corresponding positions in the 40-layer model and can optionally apply a weak LoRA contribution to the 12 newly inserted 2.9B blocks.

Main features

• Legacy 28-layer Anima LoRA support for 40-layer Anima 2.9B

• Automatic 28 → 40 block remapping

• Optional LoRA propagation to the inserted 12 blocks

• Sharp / anti-blur oriented modes

• Compatibility with older workflow JSON files

• No permanent conversion of the LoRA file

• No checkpoint or LoRA weights included

## 28 → 40 block mapping

0→0, 1→1, 2→3, 3→4, 4→6, 5→7, 6→9, 7→10, 8→12, 9→13, 10→15, 11→16, 12→18, 13→19, 14→20, 15→22, 16→23, 17→25, 18→26, 19→28, 20→29, 21→31, 22→32, 23→34, 24→35, 25→37, 26→38, 27→39

Inserted Anima 2.9B blocks:

2, 5, 8, 11, 14, 17, 21, 24, 27, 30, 33, 36

Sharp / anti-blur mode

A simple 28 → 40 remap can sometimes make legacy LoRAs look softer or blurrier on a 40-layer checkpoint.

This version includes several methods for applying a weak LoRA contribution to the inserted blocks.

Recommended starting settings

strength_model:     0.65 – 0.85

inserted_mode: copy_prev

inserted_strength: 0.12 – 0.22

inserted_scope: attn_only

A good first test is:

strength_model = 0.75

inserted_mode = copy_prev

inserted_strength = 0.18

inserted_scope = attn_only

If the image is still too soft, try lowering strength_model and inserted_strength.

Inserted modes

copy_prev
Copies the LoRA from the previous legacy block into the inserted block at reduced strength.
This is the recommended starting mode for sharper results.

sharp_blend
Uses approximately 70% of the previous block and 30% of the next block.

blend
Uses a 50/50 blend of the neighboring blocks.
This can give smoother transitions, but may also produce softer results.

copy_next
Copies the LoRA from the next legacy block.

disabled
Disables synthetic LoRA application to the inserted 12 blocks.

Inserted scope

attn_only
Only propagates attention-related LoRA components.
Recommended for the first test.

attn_and_mlp
Applies the compatibility mapping to attention and MLP components.

all
Applies all supported LoRA components to the inserted blocks.

Workflow compatibility

This version supports both custom node IDs:

Anima29LegacyLoRALoader

Anima29LegacyLoRALoaderSharp

So workflows created with the earlier version should continue to load.

The node is displayed as:

Anima 2.9B Legacy 28L LoRA Loader (Sharp)

Installation

Extract the package so that the node is located here:

ComfyUI/
└── custom_nodes/
    └── ComfyUI-Anima29-LegacyLoRA/
        └── __init__.py

Then restart ComfyUI.

Important notes

This node does not retrain the LoRA.

It performs compatibility remapping when the LoRA is loaded into ComfyUI.

Because the 40-layer Anima 2.9B architecture contains 12 additional blocks, the result may not exactly reproduce the appearance of the same LoRA on the original 28-layer Anima model.

Compatibility also depends on how the LoRA was trained. Some unusual LoRA key formats may not be recognized.

This is an experimental compatibility tool intended to make legacy Anima LoRAs more usable with 40-layer Anima 2.9B checkpoints.

License

MIT License for the custom node code.

No checkpoint weights, LoRA weights, or third-party model files are included.