WIP on updating FluxDenoise to support FLUX Fill.

invokeai/app/invocations/flux_denoise.py

@@ -267,8 +267,22 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         if is_schnell and self.control_lora:
             raise ValueError("Control LoRAs cannot be used with FLUX Schnell")
 
-        # Prepare the extra image conditioning tensor if a FLUX structural control image is provided.
-        img_cond = self._prep_structural_control_img_cond(context)
+        # TODO(ryand): It's a bit confusing that we support inpainting via both FLUX Fill and masked image-to-image.
+        # Think about ways to tidy this interface, or at least add clear error messages when incompatible inputs are
+        # provided.
+
+        # Prepare the extra image conditioning tensor if either of the following are provided:
+        # - FLUX structural control image
+        # - FLUX Fill conditioning
+        img_cond: torch.Tensor | None = None
+        if self.control_lora is not None and self.fill_conditioning is not None:
+            raise ValueError("Control LoRA and Fill conditioning cannot be used together.")
+        elif self.control_lora is not None:
+            img_cond = self._prep_structural_control_img_cond(context)
+        elif self.fill_conditioning is not None:
+            img_cond = self._prep_flux_fill_img_cond(
+                context, device=TorchDevice.choose_torch_device(), dtype=inference_dtype
+            )
 
         inpaint_mask = self._prep_inpaint_mask(context, x)
 
@@ -672,6 +686,56 @@ class FluxDenoiseInvocation(BaseInvocation, WithMetadata, WithBoard):
         vae_info = context.models.load(self.controlnet_vae.vae)
         return FluxVaeEncodeInvocation.vae_encode(vae_info=vae_info, image_tensor=img_cond)
 
+    def _prep_flux_fill_img_cond(
+        self, context: InvocationContext, device: torch.device, dtype: torch.dtype
+    ) -> torch.Tensor | None:
+        """Prepare the FLUX Fill conditioning.
+
+        This logic is based on:
+        https://github.com/black-forest-labs/flux/blob/716724eb276d94397be99710a0a54d352664e23b/src/flux/sampling.py#L107-L157
+        """
+        if self.fill_conditioning is None:
+            return None
+
+        # TODO(ryand): We should probable rename controlnet_vae. It's used for more than just ControlNets.
+        if not self.controlnet_vae:
+            raise ValueError("controlnet_vae must be set when using a FLUX Fill conditioning.")
+
+        # Load the conditioning image and resize it to the target image size.
+        cond_img = context.images.get_pil(self.fill_conditioning.image.image_name, mode="RGB")
+        cond_img = cond_img.resize((self.width, self.height), Image.Resampling.BICUBIC)
+        cond_img = np.array(cond_img)
+        cond_img = torch.from_numpy(cond_img).float() / 127.5 - 1.0
+        cond_img = einops.rearrange(cond_img, "h w c -> 1 c h w")
+        cond_img = cond_img.to(device=device, dtype=dtype)
+
+        # Load the mask and resize it to the target image size.
+        mask = context.tensors.load(self.fill_conditioning.mask.tensor_name)
+        assert mask.dtype == torch.bool
+        mask = mask.to(device=device, dtype=dtype)
+        mask = einops.rearrange(mask, "h w -> 1 1 h w")
+
+        # Prepare image conditioning.
+        cond_img = cond_img * (1 - mask)
+        vae_info = context.models.load(self.controlnet_vae.vae)
+        cond_img = FluxVaeEncodeInvocation.vae_encode(vae_info=vae_info, image_tensor=cond_img)
+        cond_img = pack(cond_img)
+
+        # Prepare mask conditioning.
+        mask = mask[:, 0, :, :]
+        # Rearrange mask to a 16-channel representation that matches the shape of the VAE-encoded latent space.
+        mask = einops.rearrange(
+            mask,
+            "b (h ph) (w pw) -> b (ph pw) h w",
+            ph=8,
+            pw=8,
+        )
+        mask = pack(mask)
+
+        # Merge image and mask conditioning.
+        img_cond = torch.cat((cond_img, mask), dim=-1)
+        return img_cond
+
     def _normalize_ip_adapter_fields(self) -> list[IPAdapterField]:
         if self.ip_adapter is None:
             return []