34 lines
1.2 KiB
Markdown
34 lines
1.2 KiB
Markdown
# float-increment-per-latent
|
|
|
|
## Node for single-pass LoRA sweep
|
|
`Per Sample LoRA Loader (Single Pass)`
|
|
|
|
Inputs:
|
|
- `model`
|
|
- `lora_name`
|
|
- `mode`: `range`, `range_step_by_batch`, `manual_values`
|
|
- `start`, `stop`
|
|
- `step` (used only in `range`)
|
|
- `direction` (used only in `range`)
|
|
- `manual_values` (used only in `manual_values`)
|
|
|
|
`range_step_by_batch` formula:
|
|
- uses the runtime latent batch from `EmptySD3LatentImage`
|
|
- if runtime batch is 1: value = `start`
|
|
- else: `auto_step = (stop - start) / (runtime_batch - 1)`
|
|
|
|
Examples:
|
|
- start=-4, stop=5, batch_size=10 -> step=1
|
|
- start=-10, stop=10, batch_size=10 -> step=2.222222...
|
|
|
|
Outputs:
|
|
- `model` (patched)
|
|
- `weights_used` (string preview)
|
|
|
|
Multiple `Per Sample LoRA Loader (Single Pass)` nodes can be stacked. Stacked LoRAs are aggregated into shared per-layer bypass hooks, and standard LoRA adapters are fused per layer into one larger rank projection where possible.
|
|
|
|
## Wiring
|
|
1. `CheckpointLoaderSimple.model` -> `Per Sample LoRA Loader (Single Pass).model`
|
|
2. Output `model` from this node -> `KSampler.model`
|
|
3. Set `EmptySD3LatentImage.batch_size` normally; `range_step_by_batch` reads the runtime batch during sampling.
|