Diffusion (Experimental) node
Diffusion Model
Experimental — results depend entirely on the model you supply, and a good one is not easy to come by. Loads a converted Core ML diffusion model from a folder you choose on your own disk (Settings tab) and hands it to Diffusion nodes. Grafit includes no model and downloads none: the model and its license are yours. Loading runs in the background — the first load of a model compiles it for the chosen hardware and can take a minute; Status says where it is.
Inputs
| Port | Type | Default |
|---|---|---|
| EnableOff unloads the model and frees its memory (2–8 GB). On loads the folder chosen in the Settings tab again — fast after the first time. | bool | true |
| Compute UnitsWhich hardware runs the model. GPU is fastest on Pro/Max/Ultra chips and for "original" conversions. Neural Engine suits "split-einsum" conversions and leaves the GPU to Grafit's rendering. Changing it reloads the model. | int | 0 |
| Reduce MemoryLoads each part of the model only while it is needed and unloads it after. Much less memory, clearly slower per image. For 8–16 GB Macs or XL models. Changing it reloads the model. | bool | false |
Outputs
| Port | Type |
|---|---|
| Model | structure |
| Loaded | bool |
| Status | string |
| Native Size | int |
| Info | string |
Open Diffusion Model in the interactive catalog →
Diffusion Model is one of 381 nodes in Grafit Pro.
Free download. Every feature for 7 days, no card. Then €99.99 once — or keep playing your projects as a Player.