diff --git a/studio/backend/core/training/worker.py b/studio/backend/core/training/worker.py
index 6915bedd4b..d0cba0c85b 100644
--- a/studio/backend/core/training/worker.py
+++ b/studio/backend/core/training/worker.py
@@ -1211,7 +1211,23 @@ def _run_mlx_training(event_queue, stop_queue, config):
     is_vlm = bool(is_dataset_image and getattr(model, "_is_vlm_model", False))
     model._is_vlm_model = is_vlm
     vision_image_size = config.get("vision_image_size")
-    if is_vlm and vision_image_size is not None:
+    # Mirror the Torch trainer.py exclusion: DeepSeek OCR's preset is a tuple
+    # (image_size, base_size, crop_mode), so resizing dataset images outside
+    # that preset desyncs the token grid. Skip the resize on MLX too.
+    _model_name_lower = str(config.get("model_name", "")).lower()
+    _is_deepseek_ocr = (
+        "deepseek" in _model_name_lower and "ocr" in _model_name_lower
+    )
+    if is_vlm and vision_image_size is not None and _is_deepseek_ocr:
+        _send(
+            "status",
+            status_message = (
+                "MLX vision image resize ignored for DeepSeek OCR "
+                "(uses fixed Gundam preset)."
+            ),
+        )
+        vision_image_size = None
+    elif is_vlm and vision_image_size is not None:
         vision_image_size = int(vision_image_size)
         _send(
             "status",
diff --git a/studio/frontend/src/features/studio/sections/params-section.tsx b/studio/frontend/src/features/studio/sections/params-section.tsx
index d6489d677c..799ae4e059 100644
--- a/studio/frontend/src/features/studio/sections/params-section.tsx
+++ b/studio/frontend/src/features/studio/sections/params-section.tsx
@@ -133,6 +133,14 @@ export function ParamsSection(): ReactElement {
   const isCpt = store.trainingMethod === "cpt";
   const isRawText = isRawTextDatasetFormat(store.datasetFormat);
   const showVisionLora = store.isVisionModel && store.isDatasetImage === true;
+  // DeepSeek OCR's preset is a coupled tuple, so the backend ignores any
+  // user-selected image size for it. Hide the control rather than offer a
+  // setting that silently has no effect.
+  const _selectedModelLower = (store.selectedModel ?? "").toLowerCase();
+  const isDeepseekOcr =
+    _selectedModelLower.includes("deepseek") &&
+    _selectedModelLower.includes("ocr");
+  const showVisionImageSize = showVisionLora && !isDeepseekOcr;
   const [loraOpen, setLoraOpen] = useState(false);
   const [hyperOpen, setHyperOpen] = useState(false);
   const needsExpandedHeight = isCpt || (isLora && loraOpen) || hyperOpen;
@@ -917,7 +925,7 @@ export function ParamsSection(): ReactElement {
                 </TabsContent>
 
                 <TabsContent value="memory" className="mt-3 flex flex-col gap-3">
-                  {showVisionLora && (
+                  {showVisionImageSize && (
                     <Row
                       label="Image Size"
                       tooltip={
diff --git a/studio/frontend/src/features/training/api/mappers.ts b/studio/frontend/src/features/training/api/mappers.ts
index 75ce8ccdb5..1223888b11 100644
--- a/studio/frontend/src/features/training/api/mappers.ts
+++ b/studio/frontend/src/features/training/api/mappers.ts
@@ -23,7 +23,12 @@ export function buildTrainingStartPayload(
   const isCpt = config.trainingMethod === "cpt";
   const adapterMethod = config.trainingMethod !== "full";
   const isQloraMethod = config.trainingMethod === "qlora";
-  const isFourBitModel = (config.selectedModel ?? "").toLowerCase().includes("4bit");
+  const _selectedModelLower = (config.selectedModel ?? "").toLowerCase();
+  const isFourBitModel = _selectedModelLower.includes("4bit");
+  // DeepSeek OCR ignores user-selected image size; do not send it.
+  const isDeepseekOcr =
+    _selectedModelLower.includes("deepseek") &&
+    _selectedModelLower.includes("ocr");
   const isEmbedding = config.isEmbeddingModel;
   const isRawText = isRawTextDatasetFormat(config.datasetFormat);
   const hfDataset = config.datasetSource === "huggingface" ? config.dataset : null;
@@ -56,7 +61,7 @@ export function buildTrainingStartPayload(
     load_in_4bit: (adapterMethod && isQloraMethod) || (isCpt && isFourBitModel),
     max_seq_length: config.contextLength,
     vision_image_size:
-      config.isVisionModel && config.isDatasetImage === true
+      config.isVisionModel && config.isDatasetImage === true && !isDeepseekOcr
         ? config.visionImageSize
         : null,
     trust_remote_code: config.trustRemoteCode ?? false,