diff --git a/README.md b/README.md
index 71b8a98..7e25304 100644
--- a/README.md
+++ b/README.md
@@ -137,6 +137,16 @@ flow video "a dragon flying over mountains" --resolution 1080p
 flow video "a dragon flying over mountains" --resolution 4k -o /absolute/path/dragon.mp4
 ```
 
+Draft quickly, upscale for quality. `--draft` generates at 360p, the faster,
+lower-credit mode Google Flow added with Gemini Omni 1.1 Flash (~half the
+credits of the 720p default: 4s=4, 6s=5, 8s=6, 10s=7 vs 7/10/12/15):
+
+```bash
+flow video "a dragon flying over mountains" --draft
+```
+
+Draft a concept, then regenerate the keeper with `--resolution 1080p` or `--resolution 4k`.
+
 The upsampled file is written to `--output`; the 720p original stays in the
 output directory and in `history.json`. If the upsample pass fails, the 720p
 video is still delivered and the CLI prints a note.
diff --git a/flow-agent/flow_server/models.py b/flow-agent/flow_server/models.py
index 9eea3cb..6e9f3f0 100644
--- a/flow-agent/flow_server/models.py
+++ b/flow-agent/flow_server/models.py
@@ -39,6 +39,7 @@ class VideoGenerationRequest(BaseModel):
             "the same step behind the Flow UI's high-resolution download."
         ),
     )
+    draft: Optional[bool] = Field(False, description="Generate a faster, lower-credit 360p draft (about half the credits)")
 
 
 class VideoUpsampleRequest(BaseModel):
diff --git a/flow-agent/flow_server/routes/generation.py b/flow-agent/flow_server/routes/generation.py
index 3d81da5..30fd500 100644
--- a/flow-agent/flow_server/routes/generation.py
+++ b/flow-agent/flow_server/routes/generation.py
@@ -46,6 +46,11 @@ def _request_payload(req, operation: str, x_client_id: Optional[str]) -> Dict[st
     return {"operation": operation, "request": request_data, "client_id": x_client_id}
 
 
+def _video_model_key(duration: int, draft: bool, video_model: Optional[str] = None) -> str:
+    """Video model key; _360p suffix requests a fast, lower-credit 360p draft."""
+    return video_model or f"abra_t2v_{duration}s" + ("_360p" if draft else "")
+
+
 def _header_string(value: Any) -> Optional[str]:
     """FastAPI Header defaults remain Header objects in direct MCP calls."""
     if not isinstance(value, str):
@@ -789,6 +794,7 @@ async def _generate_video(req: VideoGenerationRequest, x_client_id: Optional[str
 
     try:
         # Submit generation
+        video_model = _video_model_key(req.duration, req.draft, req.video_model)
         if is_video_input and image_media_id:
             from flow_engine.generators.v2v import edit_video
             media_ids = await edit_video(active_bridge, req.prompt, aspect_key, project_id, image_media_id, duration=req.duration, ref_media_ids=ref_media_ids or None)
@@ -804,17 +810,17 @@ async def _generate_video(req: VideoGenerationRequest, x_client_id: Optional[str
                 duration=req.duration,
                 count=generation_count,
                 seed=req.seed,
-                video_model=req.video_model,
+                video_model=video_model,
             )
         elif ref_media_ids:
             from flow_engine.generators.i2v import generate_video_r2v
-            media_ids = await generate_video_r2v(active_bridge, req.prompt, aspect_key, project_id, ref_media_ids, duration=req.duration, count=generation_count, seed=req.seed, video_model=req.video_model)
+            media_ids = await generate_video_r2v(active_bridge, req.prompt, aspect_key, project_id, ref_media_ids, duration=req.duration, count=generation_count, seed=req.seed, video_model=video_model)
         elif image_media_id:
             from flow_engine.generators.i2v import generate_video_i2v
-            media_ids = await generate_video_i2v(active_bridge, req.prompt, aspect_key, project_id, image_media_id, duration=req.duration, count=generation_count, seed=req.seed, video_model=req.video_model)
+            media_ids = await generate_video_i2v(active_bridge, req.prompt, aspect_key, project_id, image_media_id, duration=req.duration, count=generation_count, seed=req.seed, video_model=video_model)
         else:
             from flow_engine.generators.t2v import generate_video
-            media_ids = await generate_video(active_bridge, req.prompt, aspect_key, project_id, duration=req.duration, count=generation_count, seed=req.seed, video_model=req.video_model)
+            media_ids = await generate_video(active_bridge, req.prompt, aspect_key, project_id, duration=req.duration, count=generation_count, seed=req.seed, video_model=video_model)
     except Exception as e:
         if temp_img_path and os.path.exists(temp_img_path):
             try:
diff --git a/flow-agent/main.py b/flow-agent/main.py
index f0d1a7d..514b8a8 100644
--- a/flow-agent/main.py
+++ b/flow-agent/main.py
@@ -558,6 +558,7 @@ def cmd_video(argv):
     parser.add_argument("--end", metavar="IMAGE", help="End image path or media ID (use with --start)")
     parser.add_argument("--ref", "-r", nargs="+", metavar="IMAGE", help="Reference image paths or media IDs")
     parser.add_argument("--project-id", "-p", help="Deprecated; configure DEFAULT_PROJECT on the backend")
+    parser.add_argument("--draft", action="store_true", help="Generate a faster, lower-credit 360p draft (~half the credits)")
     parser.add_argument(
         "--resolution",
         "-R",
@@ -586,6 +587,8 @@ def cmd_video(argv):
         "duration": args.duration,
         "n": args.count,
     }
+    if args.draft:
+        payload["draft"] = True
     if args.resolution != "720p":
         payload["resolution"] = args.resolution
     if args.edit:
diff --git a/flow-agent/tests/test_video_draft.py b/flow-agent/tests/test_video_draft.py
new file mode 100644
index 0000000..5fb930b
--- /dev/null
+++ b/flow-agent/tests/test_video_draft.py
@@ -0,0 +1,57 @@
+"""Draft-at-360p generation: --draft / draft:true request the _360p model key."""
+
+import io
+
+import main
+from flow_server.routes.generation import _video_model_key
+
+
+def test_draft_appends_360p_model_key():
+    assert _video_model_key(4, True) == "abra_t2v_4s_360p"
+    assert _video_model_key(6, False) == "abra_t2v_6s"
+    assert _video_model_key(10, True, "custom_model") == "custom_model"
+
+
+READY = {"status": "healthy", "extension_connected": True, "has_flow_key": True}
+
+
+class DownloadResponse(io.BytesIO):
+    status = 200
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc, traceback):
+        self.close()
+
+
+def test_cli_draft_flag_sends_payload(monkeypatch, tmp_path):
+    requested = tmp_path / "draft.mp4"
+    captured = {}
+    media_bytes = bytes([0, 0, 0, 24]) + b"ftypisomvideo"
+    monkeypatch.setattr(main, "_wait_for_generation_ready", lambda: READY)
+
+    def fake_post(path, payload, key, timeout):
+        captured.update(path=path, payload=payload, key=key)
+        return {"job_id": "job-draft", "status": "processing", "data": []}
+
+    def fake_request(path, **kwargs):
+        return {
+            "job_id": "job-draft",
+            "status": "succeeded",
+            "data": [{"url": "https://media.invalid/video", "media_id": "video-1"}],
+        }, 200
+
+    monkeypatch.setattr(main, "_post_generation", fake_post)
+    monkeypatch.setattr(main, "_request_json", fake_request)
+    monkeypatch.setattr(main.time, "sleep", lambda _seconds: None)
+    monkeypatch.setattr(
+        main.urllib.request,
+        "urlopen",
+        lambda *args, **kwargs: DownloadResponse(media_bytes),
+    )
+
+    main.cmd_video(["draft it", "--draft", "--output", str(requested)])
+
+    assert captured["payload"]["draft"] is True
+    assert requested.read_bytes() == media_bytes