ComfyUI Node
Mpi Krea 2 Ostris Edit Model Patch
Enable reference latents on a Krea 2 model (index_timestep_zero method, as trained by ai-toolkit). Chain conditioning from MpiKrea2OstrisEditTextEncode or Set Reference Latent nodes. kv_cache enables the cached one-pass reference mode; the LoRA must be trained with ai-toolkit's kv_cache option for it to work properly.
Mpi Krea 2 Ostris Edit Model Patch
- model
- MODEL
◄kv_cachefalse►
Categoryostris/krea2
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| kv_cache | BOOLEAN | false | Cache the reference tokens' attention K/V: they are precomputed in one t=0 pass and reused every denoising step, so the refs never ride along in the per-step sequence. The LoRA must be trained with ai-toolkit's kv_cache model kwarg for this to work properly. Leave off for normal edit LoRAs. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| MODEL | MODEL | — |