From 4b48535a7d66b89a4314e087e70fb7051e54eaaa Mon Sep 17 00:00:00 2001
From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com>
Date: Mon, 1 Jun 2026 18:08:20 -0700
Subject: [PATCH] Do tripo dinov3 inference in fp32. (#14221)

---
 comfy/image_encoders/dino3.py    | 7 ++++---
 comfy_extras/nodes_triposplat.py | 3 +--
 2 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/comfy/image_encoders/dino3.py b/comfy/image_encoders/dino3.py
index 014d1d29a..ad29b06f8 100644
--- a/comfy/image_encoders/dino3.py
+++ b/comfy/image_encoders/dino3.py
@@ -3,6 +3,7 @@ import torch
 import torch.nn as nn
 import torch.nn.functional as F
 
+import comfy.ops
 from comfy.ldm.modules.attention import optimized_attention_for_device
 from comfy.image_encoders.dino2 import LayerScale as DINOv3ViTLayerScale
 
@@ -171,11 +172,11 @@ class DINOv3ViTEmbeddings(nn.Module):
         patch_embeddings = patch_embeddings.flatten(2).transpose(1, 2)
 
         if bool_masked_pos is not None:
-            mask_token = self.mask_token.to(patch_embeddings.dtype)
+            mask_token = comfy.ops.cast_to_input(self.mask_token, patch_embeddings)
             patch_embeddings = torch.where(bool_masked_pos.unsqueeze(-1), mask_token, patch_embeddings)
 
-        cls_token = self.cls_token.expand(batch_size, -1, -1).to(patch_embeddings.device)
-        register_tokens = self.register_tokens.expand(batch_size, -1, -1).to(patch_embeddings.device)
+        cls_token = comfy.ops.cast_to_input(self.cls_token.expand(batch_size, -1, -1), patch_embeddings)
+        register_tokens = comfy.ops.cast_to_input(self.register_tokens.expand(batch_size, -1, -1), patch_embeddings)
         embeddings = torch.cat([cls_token, register_tokens, patch_embeddings], dim=1)
         return embeddings
 
diff --git a/comfy_extras/nodes_triposplat.py b/comfy_extras/nodes_triposplat.py
index 5646d611b..1848ad31a 100644
--- a/comfy_extras/nodes_triposplat.py
+++ b/comfy_extras/nodes_triposplat.py
@@ -115,12 +115,11 @@ class TripoSplatConditioning(IO.ComfyNode):
         # feature1: DINOv3 token sequence (cls + registers + patches), ImageNet-normalized, with a final non-affine layer norm on top
         comfy.model_management.load_model_gpu(clip_vision.patcher)
         device = clip_vision.load_device
-        model_dtype = next(clip_vision.model.parameters()).dtype
         img = image.movedim(-1, 1).to(device)  # (B,3,H,W) in [0,1]
         mean = torch.tensor(_DINOV3_MEAN, device=device).view(1, 3, 1, 1)
         std = torch.tensor(_DINOV3_STD, device=device).view(1, 3, 1, 1)
         img = (img - mean) / std
-        seq = clip_vision.model(pixel_values=img.to(model_dtype))[0]
+        seq = clip_vision.model(pixel_values=img.float())[0]
         feature1 = F.layer_norm(seq.float(), seq.shape[-1:]).to(comfy.model_management.intermediate_device())
 
         # Second conditioning: the Flux2 VAE latent of the image, carried as a standard reference_latents entry