diff --git a/.gitignore b/.gitignore
index ab2e70d..5f441bf 100644
--- a/.gitignore
+++ b/.gitignore
@@ -28,6 +28,8 @@ static/arthur/js/
 # the Django half's own state: the document database, the content-addressed blob
 # store (tiers 2 and 3), and collectstatic's output
 db.sqlite3
+db.sqlite3-shm
+db.sqlite3-wal
 /var/
 
 # vim swap files
diff --git a/clips/blobs.py b/clips/blobs.py
index 7469e66..2216fcb 100644
--- a/clips/blobs.py
+++ b/clips/blobs.py
@@ -16,11 +16,13 @@ addressing that answers questions about work not yet done.
 import hashlib
 import os
 import tempfile
+import zlib
 from pathlib import Path
 
 from django.conf import settings
 
 CHUNK = 1 << 20
+CROP_MEDIA_TYPE = "application/zlib"
 
 
 def digest_bytes(data: bytes) -> str:
@@ -86,6 +88,20 @@ def write_stream(chunks) -> tuple[str, int]:
     return digest.hexdigest(), size
 
 
+def write_compressed_stream(chunks) -> tuple[str, int]:
+    """Store a losslessly compressed stream; the digest names stored bytes."""
+    compressor = zlib.compressobj()
+
+    def compressed():
+        for chunk in chunks:
+            if part := compressor.compress(chunk):
+                yield part
+        if part := compressor.flush():
+            yield part
+
+    return write_stream(compressed())
+
+
 def adopt(source: Path) -> tuple[str, int]:
     """Store a file already on disk, by hard link where the filesystem allows it.
 
diff --git a/clips/management/commands/compress_crop_blocks.py b/clips/management/commands/compress_crop_blocks.py
new file mode 100644
index 0000000..6367b81
--- /dev/null
+++ b/clips/management/commands/compress_crop_blocks.py
@@ -0,0 +1,57 @@
+"""Compress existing raw mouth crop blocks without changing their public bytes."""
+
+import hashlib
+import zlib
+
+from django.core.management.base import BaseCommand, CommandError
+from django.db import transaction
+from django.db.models.deletion import ProtectedError
+
+from clips import blobs
+from clips.models import Blob, Block
+
+
+class Command(BaseCommand):
+    help = "Compress existing source/crops blobs and remove unreferenced raw copies"
+
+    def handle(self, *args, **options):
+        converted = 0
+        before = after = 0
+        for block in Block.objects.filter(role="source/crops").select_related("data"):
+            old = block.data
+            if old.media_type == blobs.CROP_MEDIA_TYPE:
+                continue
+            old_digest = old.digest
+            with open(blobs.path_for(old_digest), "rb") as source:
+                digest, size = blobs.write_compressed_stream(
+                    iter(lambda: source.read(blobs.CHUNK), b"")
+                )
+            check = hashlib.sha256()
+            decompressor = zlib.decompressobj()
+            with open(blobs.path_for(digest), "rb") as compressed:
+                while chunk := compressed.read(blobs.CHUNK):
+                    check.update(decompressor.decompress(chunk))
+                check.update(decompressor.flush())
+            if not decompressor.eof or check.hexdigest() != old_digest:
+                raise CommandError(f"crop compression failed verification: {block.key}")
+            with transaction.atomic():
+                new, _ = Blob.objects.get_or_create(
+                    digest=digest,
+                    defaults={"size": size, "media_type": blobs.CROP_MEDIA_TYPE},
+                )
+                changed = Block.objects.filter(key=block.key, data=old).update(data=new)
+            if not changed:
+                continue
+            converted += 1
+            before += old.size
+            after += size
+            if old_digest != new.digest:
+                try:
+                    old.delete()
+                except ProtectedError:
+                    pass
+                else:
+                    blobs.path_for(old_digest).unlink(missing_ok=True)
+        self.stdout.write(
+            f"Compressed {converted} crop blocks: {before:,} -> {after:,} bytes"
+        )
diff --git a/clips/tests/test_api.py b/clips/tests/test_api.py
index 411c0d6..da3b534 100644
--- a/clips/tests/test_api.py
+++ b/clips/tests/test_api.py
@@ -16,6 +16,7 @@ of the system rather than a convention in ClojureScript:
 The rest is the load/save round trip, the conditional write, and the footage
 manifest that makes the frames the backend's to serve.
 """
+import base64
 import hashlib
 import json
 import shutil
@@ -23,11 +24,13 @@ import struct
 import subprocess
 import tempfile
 import zlib
+from io import StringIO
 from pathlib import Path
 from unittest import skipUnless
 from unittest.mock import Mock, patch
 
 from django.core.files.uploadedfile import SimpleUploadedFile
+from django.core.management import call_command
 from django.test import TestCase, override_settings
 
 from clips import blobs, extraction
@@ -215,6 +218,47 @@ class Tier2Tests(TestCase):
         self.assertEqual("AAE=", fetched["state"])
         self.assertEqual(descriptor, fetched["descriptor"])
 
+    @override_settings(DATA_UPLOAD_MAX_MEMORY_SIZE=1024, FILE_UPLOAD_MAX_MEMORY_SIZE=1024)
+    def test_large_block_upload_streams_past_json_body_limit(self):
+        analysis = self.register_analysis()
+        descriptor = block_descriptor(analysis, role="source/crops")
+        key = key_for(descriptor)
+        payload = bytes(range(256)) * 16
+        response = self.client.post("/api/blocks", {
+            "key": key,
+            "descriptor": descriptor,
+            "data": SimpleUploadedFile("block.bin", payload),
+            "state": SimpleUploadedFile("state.bin", b"\x00\x01"),
+        })
+        self.assertEqual(201, response.status_code, response.content)
+        row = Block.objects.get(key=key)
+        self.assertEqual(blobs.CROP_MEDIA_TYPE, row.data.media_type)
+        self.assertLess(row.data.size, len(payload))
+        self.assertEqual(payload, zlib.decompress(blobs.read(row.data_id)))
+        self.assertEqual(base64.b64encode(payload).decode(),
+                         self.client.get(f"/api/blocks/{key}").json()["data"])
+        self.assertEqual(b"\x00\x01", blobs.read(row.state_id))
+
+    def test_existing_raw_crop_block_is_compressed_without_changing_its_key_or_read(self):
+        analysis = self.register_analysis()
+        descriptor = block_descriptor(analysis, role="source/crops")
+        key = key_for(descriptor)
+        payload = b"raw crop pixels" * 100
+        digest, size = blobs.write(payload)
+        old = Blob.objects.create(digest=digest, size=size)
+        Block.objects.create(key=key, descriptor=descriptor, role="source/crops",
+                             analysis_id=analysis, data=old)
+
+        call_command("compress_crop_blocks", stdout=StringIO())
+        row = Block.objects.select_related("data").get(key=key)
+        self.assertEqual(blobs.CROP_MEDIA_TYPE, row.data.media_type)
+        self.assertEqual(base64.b64encode(payload).decode(),
+                         self.client.get(f"/api/blocks/{key}").json()["data"])
+        self.assertFalse(blobs.path_for(digest).exists())
+        compressed_digest = row.data_id
+        call_command("compress_crop_blocks", stdout=StringIO())
+        self.assertEqual(compressed_digest, Block.objects.get(key=key).data_id)
+
     def test_a_block_whose_analysis_is_unknown_is_refused(self):
         descriptor = block_descriptor("sha256:" + "f" * 64)
         response = self.post("/api/blocks", {
diff --git a/clips/views.py b/clips/views.py
index b8a1c59..6203fe7 100644
--- a/clips/views.py
+++ b/clips/views.py
@@ -26,6 +26,7 @@ tool got worse", with no event to attach it to.
 import hashlib
 import json
 import re
+import zlib
 from functools import lru_cache
 from pathlib import Path
 from uuid import UUID
@@ -98,6 +99,22 @@ def _blob(b64, media_type="application/octet-stream"):
     return blob
 
 
+def _uploaded_blob(upload, media_type="application/octet-stream"):
+    digest, size = blobs.write_stream(upload.chunks())
+    blob, _ = Blob.objects.get_or_create(
+        digest=digest, defaults={"size": size, "media_type": media_type}
+    )
+    return blob
+
+
+def _crop_blob(chunks):
+    digest, size = blobs.write_compressed_stream(chunks)
+    blob, _ = Blob.objects.get_or_create(
+        digest=digest, defaults={"size": size, "media_type": blobs.CROP_MEDIA_TYPE}
+    )
+    return blob
+
+
 # ---------------------------------------------------------------------------
 # the page
 
@@ -456,7 +473,10 @@ def blocks(request):
     """Store one dense block: its bytes, its optional absence mask, and the
     descriptor its key is the hash of."""
     try:
-        data = _body(request)
+        multipart = request.content_type == "multipart/form-data"
+        data = request.POST if multipart else _body(request)
+        upload = request.FILES.get("data") if multipart else None
+        state_upload = request.FILES.get("state") if multipart else None
         key = data.get("key")
         descriptor = data.get("descriptor")
         parsed = _check_key(key, descriptor)
@@ -478,17 +498,27 @@ def blocks(request):
                 "version that produced it",
                 analysis=analysis_key,
             )
-        if not data.get("data"):
+        if not (upload and upload.size) and not data.get("data"):
             raise Bad("a block with no bytes")
         with transaction.atomic():
+            if role == "source/crops":
+                if upload:
+                    data_blob = _crop_blob(upload.chunks())
+                else:
+                    import base64
+
+                    data_blob = _crop_blob([base64.b64decode(data["data"])])
+            else:
+                data_blob = _uploaded_blob(upload) if upload else _blob(data["data"])
             row, created = Block.objects.get_or_create(
                 key=key,
                 defaults={
                     "descriptor": descriptor,
                     "role": role,
                     "analysis": analysis,
-                    "data": _blob(data["data"]),
-                    "state": _blob(data["state"]) if data.get("state") else None,
+                    "data": data_blob,
+                    "state": (_uploaded_blob(state_upload) if state_upload else
+                              _blob(data["state"]) if data.get("state") else None),
                 },
             )
         return JsonResponse({"key": row.key, "created": created}, status=201 if created else 200)
@@ -504,10 +534,13 @@ def block_detail(request, key):
         row = Block.objects.select_related("data", "state").get(key=key)
     except Block.DoesNotExist:
         return JsonResponse({"error": "no such block"}, status=404)
+    data = blobs.read(row.data.digest)
+    if row.data.media_type == blobs.CROP_MEDIA_TYPE:
+        data = zlib.decompress(data)
     out = {
         "key": row.key,
         "descriptor": row.descriptor,
-        "data": base64.b64encode(blobs.read(row.data.digest)).decode("ascii"),
+        "data": base64.b64encode(data).decode("ascii"),
     }
     if row.state_id:
         out["state"] = base64.b64encode(blobs.read(row.state.digest)).decode("ascii")
diff --git a/docs/architecture.md b/docs/architecture.md
index 86f3f6f..90d6536 100644
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -444,9 +444,12 @@ must never run while the transport is moving.
 
 The mouth crops are the non-obvious entry, and they are what makes remote work
 possible at all. `extractTeeth` reads source pixels, so without them a
-collaborator holding the analysis but not the 600 source PNGs cannot touch a
-single teeth knob. A 40×30 crop is about 1.2KB; a 600-frame take is under a
-megabyte against hundreds for the footage.
+collaborator holding the analysis but not the source video cannot touch a
+single teeth knob without decoding video again. These are RGBA crops: a 40×30
+crop is 4.8KB raw, and current 200-pixel-wide crops can total over 12MB for a
+take. The blob store compresses `source/crops` losslessly with zlib; block reads
+return the original pixels. Run `python manage.py compress_crop_blocks` once to
+convert existing raw crop blobs and remove their unreferenced copies.
 
 ### Bake B — resolved geometry. For scale.
 
@@ -573,7 +576,7 @@ POST /api/analyses              {key, descriptor}        idempotent
 GET  /api/analyses/<key>        metadata + source block keys
 PUT  /api/analyses/<key>        link dense landmarks, mask, crops
 POST /api/blocks/missing        {keys} -> {missing}
-POST /api/blocks                {key, descriptor, data, state}
+POST /api/blocks                multipart: key, descriptor, data file, optional state file (JSON also accepted)
 GET  /api/blocks/<key>
 GET  /api/footage/<id>          the manifest: video and stream URLs, audio, a URL per tracing still
 GET  /blob/<digest>             immutable bytes, with byte ranges for video playback
diff --git a/frontend/src/arthur/domain/wire.cljs b/frontend/src/arthur/domain/wire.cljs
index 272c673..f16e453 100644
--- a/frontend/src/arthur/domain/wire.cljs
+++ b/frontend/src/arthur/domain/wire.cljs
@@ -15,8 +15,8 @@
   a codec that returned a sorted map would work locally and stop working after one
   round trip, which is the failure the plain-map rule already prevents.
 
-  The bytes are separate and base64, because tier 2 is typed arrays and transit
-  has nothing to say about them. `channel/dense-at` reads a block as
+  The bytes are separate from transit: JSON block reads carry base64, while
+  uploads send binary file parts. `channel/dense-at` reads a block as
   `{:data <typed array> :state <Uint8Array>}` and both sides of the wire must hold
   byte-for-byte the same array — a handle that names a sha256 has to name the
   bytes you actually hold."
diff --git a/frontend/src/arthur/events/project.cljs b/frontend/src/arthur/events/project.cljs
index 7f776f7..d9e1439 100644
--- a/frontend/src/arthur/events/project.cljs
+++ b/frontend/src/arthur/events/project.cljs
@@ -23,6 +23,7 @@
   fx, which is the only thing in this namespace that is not pure."
   (:require [arthur.domain.clip :as clip]
             [arthur.domain.project :as project]
+            [arthur.domain.wire :as wire]
             [arthur.events.playback :as pb]
             [arthur.footage.store :as store]
             [arthur.flow.address :as address]
@@ -38,6 +39,20 @@
 (defn- block-keys [^js doc]
   (into-array (map #(.-key %) (array-seq (.-blocks doc)))))
 
+(defn- block-bytes [value]
+  (if (string? value)
+    (wire/bytes-of value)
+    (js/Uint8Array. (.-buffer value) (.-byteOffset value) (.-byteLength value))))
+
+(defn- block-form [^js block]
+  (let [form (js/FormData.)]
+    (.append form "key" (.-key block))
+    (.append form "descriptor" (.-descriptor block))
+    (.append form "data" (js/Blob. #js [(block-bytes (.-data block))]) "block.bin")
+    (when-let [state (.-state block)]
+      (.append form "state" (js/Blob. #js [(block-bytes state)]) "state.bin"))
+    form))
+
 (defn- upload-missing!
   "POST the blocks the server said it does not have, and nothing else.
 
@@ -45,9 +60,9 @@
   this and it made sqlite answer \"database is locked\" on a save — which reaches
   the page as a 500 with nothing wrong with the request. The backend was fixed too
   (WAL, and a busy timeout, in server/settings.py), and this stays sequential
-  anyway: the uploads are a few kilobytes each, nothing is waiting on them, and a
-  burst of parallel writes to buy nothing is how the same bug comes back the first
-  time a take has sixty blocks instead of eleven."
+  anyway: most uploads are small, and a burst of parallel writes to buy nothing
+  is how the same bug comes back the first time a take has sixty blocks instead
+  of eleven."
   [^js doc]
   (-> (http/POST "/api/blocks/missing" #js {:keys (block-keys doc)})
       (.then (fn [^js answer]
@@ -55,7 +70,9 @@
                      todo    (filterv #(contains? missing (.-key ^js %))
                                       (array-seq (.-blocks doc)))]
                  (-> (reduce (fn [chain block]
-                               (.then chain (fn [_] (http/POST "/api/blocks" block))))
+                               (.then chain
+                                      (fn [_]
+                                        (http/POST-form "/api/blocks" (block-form block)))))
                              (js/Promise.resolve nil)
                              todo)
                      (.then (fn [_] (count todo)))))))))
@@ -104,7 +121,7 @@
                       (.then (fn [_]
                                (when (seq source-blocks)
                                  (-> (upload-missing!
-                                      #js {:blocks (source/wire-blocks source-blocks)})
+                                      #js {:blocks (source/upload-blocks source-blocks)})
                                      (.then (fn [_]
                                               (http/PUT
                                                (str "/api/analyses/" (:id analysis))
diff --git a/frontend/src/arthur/flow/source.cljs b/frontend/src/arthur/flow/source.cljs
index 91a7190..c196577 100644
--- a/frontend/src/arthur/flow/source.cljs
+++ b/frontend/src/arthur/flow/source.cljs
@@ -75,6 +75,13 @@
             #js {:key key :descriptor descriptor :data (wire/base64 data)}))
         roles)))
 
+(defn upload-blocks [blocks]
+  (into-array
+   (map (fn [role]
+          (let [{:keys [key descriptor data]} (get blocks role)]
+            #js {:key key :descriptor descriptor :data data}))
+        roles)))
+
 (defn unpack
   "The three block-detail responses -> inputs for measurement and freeze."
   [^js responses [width height]]