Add support for Chroma Radiance (#9682)

* Initial Chroma Radiance support

* Minor Chroma Radiance cleanups

* Update Radiance nodes to ensure latents/images are on the intermediate device

* Fix Chroma Radiance memory estimation.

* Increase Chroma Radiance memory usage factor

* Increase Chroma Radiance memory usage factor once again

* Ensure images are multiples of 16 for Chroma Radiance
Add batch dimension and fix channels when necessary in ChromaRadianceImageToLatent node

* Tile Chroma Radiance NeRF to reduce memory consumption, update memory usage factor

* Update Radiance to support conv nerf final head type.

* Allow setting NeRF embedder dtype for Radiance
Bump Radiance nerf tile size to 32
Support EasyCache/LazyCache on Radiance (maybe)

* Add ChromaRadianceStubVAE node

* Crop Radiance image inputs to multiples of 16 instead of erroring to be in line with existing VAE behavior

* Convert Chroma Radiance nodes to V3 schema.

* Add ChromaRadianceOptions node and backend support.
Cleanups/refactoring to reduce code duplication with Chroma.

* Fix overriding the NeRF embedder dtype for Chroma Radiance

* Minor Chroma Radiance cleanups

* Move Chroma Radiance to its own directory in ldm
Minor code cleanups and tooltip improvements

* Fix Chroma Radiance embedder dtype overriding

* Remove Radiance dynamic nerf_embedder dtype override feature

* Unbork Radiance NeRF embedder init

* Remove Chroma Radiance image conversion and stub VAE nodes
Add a chroma_radiance option to the VAELoader builtin node which uses comfy.sd.PixelspaceConversionVAE
Add a PixelspaceConversionVAE to comfy.sd for converting BHWC 0..1 <-> BCHW -1..1

comfy_extras/nodes_chroma_radiance.py

@@ -0,0 +1,114 @@
+from typing_extensions import override
+from typing import Callable
+
+import torch
+
+import comfy.model_management
+from comfy_api.latest import ComfyExtension, io
+
+import nodes
+
+class EmptyChromaRadianceLatentImage(io.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="EmptyChromaRadianceLatentImage",
+            category="latent/chroma_radiance",
+            inputs=[
+                io.Int.Input(id="width", default=1024, min=16, max=nodes.MAX_RESOLUTION, step=16),
+                io.Int.Input(id="height", default=1024, min=16, max=nodes.MAX_RESOLUTION, step=16),
+                io.Int.Input(id="batch_size", default=1, min=1, max=4096),
+            ],
+            outputs=[io.Latent().Output()],
+        )
+
+    @classmethod
+    def execute(cls, *, width: int, height: int, batch_size: int=1) -> io.NodeOutput:
+        latent = torch.zeros((batch_size, 3, height, width), device=comfy.model_management.intermediate_device())
+        return io.NodeOutput({"samples":latent})
+
+
+class ChromaRadianceOptions(io.ComfyNode):
+    @classmethod
+    def define_schema(cls) -> io.Schema:
+        return io.Schema(
+            node_id="ChromaRadianceOptions",
+            category="model_patches/chroma_radiance",
+            description="Allows setting advanced options for the Chroma Radiance model.",
+            inputs=[
+                io.Model.Input(id="model"),
+                io.Boolean.Input(
+                    id="preserve_wrapper",
+                    default=True,
+                    tooltip="When enabled, will delegate to an existing model function wrapper if it exists. Generally should be left enabled.",
+                ),
+                io.Float.Input(
+                    id="start_sigma",
+                    default=1.0,
+                    min=0.0,
+                    max=1.0,
+                    tooltip="First sigma that these options will be in effect.",
+                ),
+                io.Float.Input(
+                    id="end_sigma",
+                    default=0.0,
+                    min=0.0,
+                    max=1.0,
+                    tooltip="Last sigma that these options will be in effect.",
+                ),
+                io.Int.Input(
+                    id="nerf_tile_size",
+                    default=-1,
+                    min=-1,
+                    tooltip="Allows overriding the default NeRF tile size. -1 means use the default (32). 0 means use non-tiling mode (may require a lot of VRAM).",
+                ),
+            ],
+            outputs=[io.Model.Output()],
+        )
+
+    @classmethod
+    def execute(
+        cls,
+        *,
+        model: io.Model.Type,
+        preserve_wrapper: bool,
+        start_sigma: float,
+        end_sigma: float,
+        nerf_tile_size: int,
+    ) -> io.NodeOutput:
+        radiance_options = {}
+        if nerf_tile_size >= 0:
+            radiance_options["nerf_tile_size"] = nerf_tile_size
+
+        if not radiance_options:
+            return io.NodeOutput(model)
+
+        old_wrapper = model.model_options.get("model_function_wrapper")
+
+        def model_function_wrapper(apply_model: Callable, args: dict) -> torch.Tensor:
+            c = args["c"].copy()
+            sigma = args["timestep"].max().detach().cpu().item()
+            if end_sigma <= sigma <= start_sigma:
+                transformer_options = c.get("transformer_options", {}).copy()
+                transformer_options["chroma_radiance_options"] = radiance_options.copy()
+                c["transformer_options"] = transformer_options
+            if not (preserve_wrapper and old_wrapper):
+                return apply_model(args["input"], args["timestep"], **c)
+            return old_wrapper(apply_model, args | {"c": c})
+
+        model = model.clone()
+        model.set_model_unet_function_wrapper(model_function_wrapper)
+        return io.NodeOutput(model)
+
+
+class ChromaRadianceExtension(ComfyExtension):
+    @override
+    async def get_node_list(self) -> list[type[io.ComfyNode]]:
+        return [
+            EmptyChromaRadianceLatentImage,
+            ChromaRadianceOptions,
+        ]
+
+
+async def comfy_entrypoint() -> ChromaRadianceExtension:
+    return ChromaRadianceExtension()