FIX: Use base64 encoded images in AI Image Caption via LLaVa (#693)

* FIX: Use base64 encoded images in AI Image Caption via LLaVa

This fixed a regression introduced in #646 where we started sending
schemaless URLs for our LLaVa service, which doesn't handle it well.

Moving to base64 encoded images solves:

- The service needing to download images
  Now the service running LLaVa doesn't need internet access

- Secure uploads compat
  Every image is treated the same, less branching for secure uploads

- Image Size problems
  Discourse is now responsible for ensure a max size for images

- Troublesome dev env
  Previously to this commit you would need a dev env that was internet
acessible to use llava image captions

lib/ai_helper/assistant.rb

@@ -130,11 +130,16 @@ module DiscourseAi
 
       def generate_image_caption(upload, user)
         if SiteSetting.ai_helper_image_caption_model == "llava"
-          image_url =
-            upload.secure? ? Discourse.store.url_for(upload) : UrlHelper.absolute(upload.url)
+          image_base64 =
+            DiscourseAi::Completions::UploadEncoder.encode(
+              upload_ids: [upload.id],
+              max_pixels: 1_048_576,
+            ).first[
+              :base64
+            ]
           parameters = {
             input: {
-              image: image_url,
+              image: "data:image/#{upload.extension};base64, #{image_base64}",
               top_p: 1,
               max_tokens: 1024,
               temperature: 0.2,

spec/requests/ai_helper/assistant_controller_spec.rb

@@ -104,7 +104,8 @@ RSpec.describe DiscourseAi::AiHelper::AssistantController do
   end
 
   describe "#caption_image" do
-    fab!(:upload)
+    let(:image) { plugin_file_from_fixtures("100x100.jpg") }
+    let(:upload) { UploadCreator.new(image, "image.jpg").create_for(Discourse.system_user.id) }
     let(:image_url) { "#{Discourse.base_url}#{upload.url}" }
     let(:caption) { "A picture of a cat sitting on a table" }
     let(:caption_with_attrs) do
@@ -197,15 +198,40 @@ RSpec.describe DiscourseAi::AiHelper::AssistantController do
       context "for secure uploads" do
         fab!(:group)
         fab!(:private_category) { Fabricate(:private_category, group: group) }
-        fab!(:post_in_secure_context) do
-          Fabricate(:post, topic: Fabricate(:topic, category: private_category))
-        end
-        fab!(:upload) { Fabricate(:secure_upload, access_control_post: post_in_secure_context) }
+        let(:image) { plugin_file_from_fixtures("100x100.jpg") }
+        let(:upload) { UploadCreator.new(image, "image.jpg").create_for(Discourse.system_user.id) }
         let(:image_url) { "#{Discourse.base_url}/secure-uploads/#{upload.url}" }
 
-        before { enable_secure_uploads }
+        before do
+          Jobs.run_immediately!
+
+          # this is done so the after_save callbacks for site settings to make
+          # UploadReference records works
+          @original_provider = SiteSetting.provider
+          SiteSetting.provider = SiteSettings::DbProvider.new(SiteSetting)
+          setup_s3
+          stub_s3_store
+          SiteSetting.secure_uploads = true
+          SiteSetting.ai_helper_allowed_groups = Group::AUTO_GROUPS[:trust_level_1]
+          SiteSetting.ai_llava_endpoint = "https://example.com"
+          Group.find(SiteSetting.ai_helper_allowed_groups_map.first).add(user)
+          user.reload
+
+          stub_request(
+            :get,
+            "http://s3-upload-bucket.s3.dualstack.us-west-1.amazonaws.com/original/1X/#{upload.sha1}.#{upload.extension}",
+          ).to_return(status: 200, body: "", headers: {})
+        end
+        after { SiteSetting.provider = @original_provider }
 
         it "returns a 403 error if the user cannot access the secure upload" do
+          create_post(
+            title: "Secure upload post",
+            raw: "This is a new post <img src=\"#{upload.url}\" />",
+            category: private_category,
+            user: Discourse.system_user,
+          )
+
           post "/discourse-ai/ai-helper/caption_image",
                params: {
                  image_url: image_url,
@@ -221,6 +247,7 @@ RSpec.describe DiscourseAi::AiHelper::AssistantController do
                  image_url: image_url,
                  image_url_type: "long_url",
                }
+
           expect(response.status).to eq(200)
           expect(response.parsed_body["caption"]).to eq(caption_with_attrs)
         end