ComfyUI Node
RS Anima Reference Frame (Experimental)
Experimental Anima reference-frame adapter for RS MultiCharacter Sampler. Connect image for a global reference, or leave it unconnected and use numbered Character Reference nodes. Uses a separately downloaded learned adapter. Not Wan/AniSora or a universal IP-Adapter. RS progressive sampling is supported. The detailer ignores image references.
RS Anima Reference Frame (Experimental)
- model
- image
- MODEL
◄adapter_name▾►
◄strength1.00►
◄reference_guidance4.5►
◄correctionapg►
◄cache_text_kvfalse►
◄reference_scale1.00►
CategoryRS MultiCharacter/Reference
Inputs (8)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| adapter_name | COMBO | Anima reference weights. Auto v2 downloads verified 2B weights on use; other variants can be installed in loras. See weight license. | |
| strength | FLOAT | 1.000–2 | Learned adapter strength. Zero bypasses reference processing. |
| reference_guidance | FLOAT | 4.50–20 | Reference influence. With one global image, 1 skips an extra prediction. Zero bypasses the reference. |
| correction | COMBO | apg | 3 options: apg, rescale, none |
| cache_text_kv | BOOLEAN | false | Reuse unchanged text K/V during this sample, up to 256 MiB. Automatically disabled with per-character LoRAs. |
| reference_scale | FLOAT | 1.000.5–1 | Reference width/height relative to target. 0.8 uses about 64% reference tokens; quality can change. |
| imageopt | IMAGE | Global reference. Leave unconnected to use numbered Character Reference nodes instead. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |