diff --git a/.hgignore b/.hgignore
index a86b81104d..c851ae937c 100644
--- a/.hgignore
+++ b/.hgignore
@@ -12,6 +12,8 @@ rust/passgen/target/
 
 web/quotes/theme/static/
 
+py/tumblrcap/dl/
+
 syntax: regexp
 ^go/trains/.*/start.sh$
 ^go/trains/.*/lukegb-trains.json$
diff --git a/py/default.nix b/py/default.nix
index ddf9ef8849..5aabf5a247 100644
--- a/py/default.nix
+++ b/py/default.nix
@@ -5,4 +5,5 @@
 args: {
   valveindexinstock = import ./valveindexinstock args;
   icalfilter = import ./icalfilter args;
+  tumblrcap = import ./tumblrcap args;
 }
diff --git a/py/tumblrcap/__init__.py b/py/tumblrcap/__init__.py
new file mode 100644
index 0000000000..e69de29bb2
diff --git a/py/tumblrcap/__main__.py b/py/tumblrcap/__main__.py
new file mode 100644
index 0000000000..d6a581844a
--- /dev/null
+++ b/py/tumblrcap/__main__.py
@@ -0,0 +1,350 @@
+import copy
+import functools
+import hashlib
+import json
+import os
+import shutil
+from pathlib import Path
+from typing import Dict, List, Optional
+
+import attrs
+import bs4
+import pytumblr2
+import requests
+import requests_oauthlib
+from absl import app
+
+TUMBLR_REQUEST_TOKEN_URL = "https://www.tumblr.com/oauth/request_token"
+TUMBLR_AUTHORIZE_URL = "https://www.tumblr.com/oauth/authorize"
+TUMBLR_ACCESS_TOKEN_URL = "https://www.tumblr.com/oauth/access_token"
+
+
+@attrs.frozen
+class Credentials:
+    consumer_key: str
+    consumer_secret: str
+    oauth_token: Optional[str] = attrs.field(default=None)
+    oauth_token_secret: Optional[str] = attrs.field(default=None)
+
+    @staticmethod
+    def _path():
+        return Path.home() / ".config" / "tumblrcap" / "credentials.json"
+
+    @classmethod
+    def load(cls):
+        try:
+            with open(cls._path(), "rt") as f:
+                loaded = json.load(f)
+                return cls(**loaded)
+        except FileNotFoundError:
+            raise
+
+    def save(self):
+        with open(self._path(), "wt") as f:
+            f.write(json.dumps(attrs.asdict(self)))
+
+    def with_token(self, oauth_token, oauth_token_secret):
+        return attrs.evolve(
+            self, oauth_token=oauth_token, oauth_token_secret=oauth_token_secret
+        )
+
+
+def make_tumblr_client() -> pytumblr2.TumblrRestClient:
+    try:
+        creds = Credentials.load()
+    except FileNotFoundError:
+        consumer_key = input("Consumer key: ").strip()
+        consumer_secret = input("Consumer secret: ").strip()
+        creds = Credentials(consumer_key=consumer_key, consumer_secret=consumer_secret)
+        creds.save()
+
+    if not creds.oauth_token:
+        oauth_session = requests_oauthlib.OAuth1Session(
+            creds.consumer_key, client_secret=creds.consumer_secret
+        )
+        fetch_response = oauth_session.fetch_request_token(TUMBLR_REQUEST_TOKEN_URL)
+        resource_owner_key = fetch_response.get("oauth_token")
+        resource_owner_secret = fetch_response.get("oauth_token_secret")
+
+        full_authorize_url = oauth_session.authorization_url(TUMBLR_AUTHORIZE_URL)
+        print("\nAuthorize: {}".format(full_authorize_url))
+        redirect_response = input("Full redirect URL here:\n").strip()
+
+        oauth_response = oauth_session.parse_authorization_response(redirect_response)
+
+        verifier = oauth_response.get("oauth_verifier")
+
+        oauth_session = OAuth1Session(
+            creds.consumer_key,
+            client_secret=creds.consumer_secret,
+            resource_owner_key=resource_owner_key,
+            resource_owner_secret=resource_owner_secret,
+            verifier=verifier,
+        )
+        oauth_tokens = oauth_session.fetch_access_token(TUMBLR_ACCESS_TOKEN_URL)
+
+        creds = creds.with_token(
+            oauth_tokens.get("oauth_token"), oauth_tokens.get("oauth_token_secret")
+        )
+        creds.save()
+
+    return pytumblr2.TumblrRestClient(
+        creds.consumer_key,
+        creds.consumer_secret,
+        creds.oauth_token,
+        creds.oauth_token_secret,
+    )
+
+
+def all_likes(client):
+    kwargs = {}
+    while True:
+        this_likes = client.likes(**kwargs)
+        if "_links" in this_likes and "next" in this_likes["_links"]:
+            kwargs = this_likes["_links"]["next"]["query_params"]
+        else:
+            break
+        for post in this_likes.get("liked_posts", []):
+            yield pytumblr2.simulate_legacy_payload(post)
+
+
+def all_posts(client, blogname):
+    kwargs = {}
+    while True:
+        kwargs['blogname'] = blogname
+        this_posts = client.posts(**kwargs)
+        if "_links" in this_posts and "next" in this_posts["_links"]:
+            kwargs = this_posts["_links"]["next"]["query_params"]
+        else:
+            break
+        for post in this_posts.get("posts", []):
+            yield pytumblr2.simulate_legacy_payload(post)
+
+
+def dump_likes(client: pytumblr2.TumblrRestClient, out_path: Path) -> None:
+    out_path.mkdir(parents=True, exist_ok=True)
+    order = []
+    for post in all_likes(client):
+        post_path = out_path / ("{}.json".format(post["id_string"]))
+        with open(post_path, "wt") as f:
+            json.dump(post, f)
+            print(post_path)
+        order.append(post["id_string"])
+    with open(out_path / "_order.json", "wt") as f:
+        json.dump(order, f)
+
+
+def dump_blog(client: pytumblr2.TumblrRestClient, blog_id: str, out_path: Path) -> None:
+    out_path.mkdir(parents=True, exist_ok=True)
+    order = []
+    for post in all_posts(client, blog_id):
+        post_path = out_path / ("{}.json".format(post["id_string"]))
+        with open(post_path, "wt") as f:
+            json.dump(post, f)
+            print(post_path)
+        order.append(post["id_string"])
+    with open(out_path / "_order.json", "wt") as f:
+        json.dump(order, f)
+
+
+def fetch_image(
+    sess: requests.Session, known_images: Dict[str, str], output_dir: Path, src: str
+) -> str:
+    if src in known_images:
+        return known_images[src]
+
+    target_filename_hash = hashlib.sha256()
+    target_filename_hash.update(src.encode("utf8"))
+    src_suffix = src
+    if "." not in src_suffix:
+        src_suffix = ".raw"
+    else:
+        src_suffix = src_suffix[src_suffix.rfind(".") :]
+    target_filename = "{}{}".format(target_filename_hash.hexdigest(), src_suffix)
+    target_path = output_dir / target_filename[0:2] / target_filename[2:4]
+    target_path.mkdir(parents=True, exist_ok=True)
+    target_path = target_path / target_filename
+    if not target_path.is_file():
+        with sess.get(src, stream=True) as r:
+            r.raw.read = functools.partial(r.raw.read, decode_content=True)
+            with open(target_path, "wb") as f:
+                shutil.copyfileobj(r.raw, f)
+    else:
+        print(target_path, "already exists")
+
+    final_src = "_images/{}/{}/{}".format(
+        target_filename[0:2], target_filename[2:4], target_filename
+    )
+    known_images[src] = final_src
+    return final_src
+
+
+def mutate_attr(d, attr_path, f):
+    if len(attr_path) == 1:
+        if attr_path[0] == "*":
+            if isinstance(d, list):
+                for n in range(len(d)):
+                    d[n] = f(d[n])
+        else:
+            if attr_path[0] in d:
+                d[attr_path[0]] = f(d[attr_path[0]])
+        return
+
+    if attr_path[0] == "*":
+        if isinstance(d, list):
+            for n in range(len(d)):
+                mutate_attr(d[n], attr_path[1:], f)
+    else:
+        if attr_path[0] not in d:
+            return
+        mutate_attr(d[attr_path[0]], attr_path[1:], f)
+
+
+def fetch_image_for_post(
+    sess: requests.Session, known_images, output_dir, post
+) -> (Dict, int):
+    post = copy.copy(post)
+    converted = 0
+
+    def _process_attr(attr, attr_value: str) -> str:
+        nonlocal converted
+        if not isinstance(attr_value, str):
+            return attr_value
+        print(post['id'], attr, attr_value)
+        soup = bs4.BeautifulSoup(attr_value, features="lxml")
+        for img in soup.find_all("img"):
+            print(attr, "src", img["src"])
+            img["src"] = fetch_image(sess, known_images, output_dir, img["src"])
+            converted += 1
+        return str(soup)
+
+    def _process_url_attr(attr, attr_value: str) -> str:
+        nonlocal converted
+        print(post['id'], attr, attr_value)
+        new_src = fetch_image(sess, known_images, output_dir, attr_value)
+        converted += 1
+        return new_src
+
+    HTML_ATTRIBUTES = [
+        "body",
+        "caption",
+        "trail.*.content",
+        "trial.*.content_raw",
+        "reblog.tree_html",
+        "reblog.comment",
+        "body_abstract",
+        "trail.*.content_abstract",
+        "answer",
+        "summary",
+    ]
+    for attr in HTML_ATTRIBUTES:
+        mutate_attr(post, attr.split("."), functools.partial(_process_attr, attr))
+    URL_ATTRIBUTES = [
+        "trail.*.content.*.media.*.url",
+        "trail.*.content.*.media.url",
+        "photos.*.original_size.url",
+        "photos.*.panorama_size.url",
+    ]
+    for attr in URL_ATTRIBUTES:
+        mutate_attr(post, attr.split("."), functools.partial(_process_url_attr, attr))
+
+    # Try to find things that look like URLs.
+    def _traverse(path, vs):
+        if isinstance(vs, dict):
+            for k, v in vs.items():
+                for v in _traverse(path + [k], v):
+                    yield v
+        elif isinstance(vs, list):
+            for k, v in enumerate(vs):
+                for v in _traverse(path + ["*"], v):
+                    yield v
+        elif isinstance(vs, str):
+            if "http://" in vs or "https://" in vs:
+                yield ".".join(path)
+
+    KNOWN_NOT_IMAGE = {
+        "blog.url",
+        "post_url",
+        "short_url",
+        "trail.*.blog.theme.header_image",
+        "trail.*.blog.theme.header_image_focused",
+        "trail.*.blog.theme.header_image_scaled",
+        "photos.*.alt_sizes.*.url",
+        "image_permalink",
+        "blog.description",
+        "link_url",
+        "asking_avatar.*.url",
+        "url",
+        "asking_url",
+        "trail.*.blog.theme.header_image_poster",
+        "audio_source_url",
+        "audio_url",
+        "embed",
+        "player",
+        "source_url",
+    }
+    print(
+        "MAYBEHTML",
+        post["id"],
+        sorted(set(_traverse([], post)) - KNOWN_NOT_IMAGE - set(HTML_ATTRIBUTES)),
+    )
+
+    return post, converted
+
+
+def fetch_images(
+    client: pytumblr2.TumblrRestClient, from_path: Path, to_path: Path
+) -> None:
+    to_path.mkdir(parents=True, exist_ok=True)
+    images_path = to_path / "_images"
+    images_path.mkdir(parents=True, exist_ok=True)
+
+    known_images = {}
+
+    sess = requests.session()
+    for child in from_path.iterdir():
+        if not child.is_file():
+            continue
+        elif child.name.startswith("_"):
+            shutil.copyfile(child, to_path / child.name)
+            continue
+
+        with open(child, "rt") as f:
+            post = json.load(f)
+        print(child, post["type"])
+
+        new_post, post_converted = fetch_image_for_post(
+            sess, known_images, images_path, post
+        )
+
+        new_child = to_path / child.name
+        with open(new_child, "wt") as f:
+            json.dump(new_post, f)
+        print(child, new_child, post_converted)
+
+
+def main(argv: List[str]) -> None:
+    if len(argv) < 2:
+        raise app.UsageError("one arg, please")
+
+    client = make_tumblr_client()
+    client.legacy_conversion_on()
+
+    if argv[1] == "dump_likes":
+        if len(argv) != 3:
+            raise app.UsageError("{} dump_likes <path>".format(argv[0]))
+        dump_likes(client, Path(argv[2]))
+    elif argv[1] == "dump_blog":
+        if len(argv) != 4:
+            raise app.UsageError("{} dump_blog <blog name> <path>".format(argv[0]))
+        dump_blog(client, argv[2], Path(argv[3]))
+    elif argv[1] == "fetch_images":
+        if len(argv) != 4:
+            raise app.UsageError("{} fetch_images <in_path> <out_path>".format(argv[0]))
+        fetch_images(client, Path(argv[2]), Path(argv[3]))
+    else:
+        raise app.UsageError("unknown subcommand".format(argv[1]))
+
+
+if __name__ == "__main__":
+    app.run(main)
diff --git a/py/tumblrcap/default.nix b/py/tumblrcap/default.nix
new file mode 100644
index 0000000000..1d4be0883d
--- /dev/null
+++ b/py/tumblrcap/default.nix
@@ -0,0 +1,75 @@
+# SPDX-FileCopyrightText: 2023 Luke Granger-Brown <depot@lukegb.com>
+#
+# SPDX-License-Identifier: Apache-2.0
+
+{ depot, lib, pkgs, ... }@args:
+
+let
+  pytumblr2 = ps: ps.buildPythonPackage rec {
+    pname = "PyTumblr2";
+    version = "0.2.2";
+
+    src = ps.fetchPypi {
+      inherit pname version;
+      sha256 = "0xpl4v25kaywyr6fbanhsx9rpmbdvb7zs6hcj6m3md3ddkp9whkf";
+    };
+
+    propagatedBuildInputs = with ps; [
+      future
+      requests_oauthlib
+    ];
+
+    checkInputs = with ps; [
+      nose
+      nose-cov
+      mock
+    ];
+  };
+
+  python = pkgs.python3.withPackages (ps: with ps; [
+    absl-py
+    attrs
+    beautifulsoup4
+    requests
+    requests_oauthlib
+    (pytumblr2 ps)
+  ]);
+
+  filterSourcePred = (path: type: (type == "regular" &&
+      lib.hasSuffix ".py" path ||
+      lib.hasSuffix ".html" path
+    ) || (
+      type == "directory" &&
+      baseNameOf path != "__pycache__" &&
+      baseNameOf path != "node_modules" &&
+      baseNameOf path != "config" &&
+      baseNameOf path != "web" &&
+      true));
+
+  tumblrcap = pkgs.stdenvNoCC.mkDerivation rec {
+    name = "tumblrcap";
+
+    src = builtins.filterSource filterSourcePred ./.;
+
+    buildInputs = with pkgs; [ makeWrapper ];
+    propagatedBuildInputs = [ python ];
+
+    installPhase = ''
+      sitepkgdir="$out/lib/${python.libPrefix}/site-packages"
+      pkgdir="$sitepkgdir/tumblrcap"
+      mkdir -p $pkgdir
+      cp -R \
+        *.py \
+        $pkgdir
+
+      mkdir "$out/bin"
+      makeWrapper "${python}/bin/python" "$out/bin/tumblrcap" \
+        --add-flags "-m" \
+        --add-flags "tumblrcap" \
+        --suffix PYTHONPATH : "$sitepkgdir"
+    '';
+
+    passthru.pythonEnv = python;
+  };
+in
+tumblrcap
diff --git a/py/tumblrcap/shell.nix b/py/tumblrcap/shell.nix
new file mode 100644
index 0000000000..8fc9f7d52f
--- /dev/null
+++ b/py/tumblrcap/shell.nix
@@ -0,0 +1,13 @@
+{ depot ? import <depot> {} }:
+
+let 
+  inherit (depot.py) tumblrcap;
+  inherit (depot) pkgs;
+in pkgs.mkShell {
+  buildInputs = with pkgs; [
+    tumblrcap.pythonEnv
+
+    black
+    python3.pkgs.isort
+  ];
+}