alt_text 0.1.1 → 0.2.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: a2ce46a5e7a525ed907991d2d50e35d8a2c002f73118e5502e7fe9305775d310
-  data.tar.gz: 564f906bdcebd926b85429d174e2c4cae9c66b1349973d7ec905cf6c74354bed
+  metadata.gz: 3b0d1d1ac9b9c9191b85d04a377a5583fc4a1b2ee54907151346a8483ef4b588
+  data.tar.gz: 49bcb6cb50d85b2af6b2145b00746d0927ec14b45969265b58593a754e437c8f
 SHA512:
-  metadata.gz: 70405efc438ae7ebb3b4ecf7cd4a9225256c44008c9dbed79958381ff48a13fc703cb73ba06c64387f053aa823139de47b18d12500b47c6c18211160d1691ac0
-  data.tar.gz: 56c6673daa0095dacc08edec163ac82ec9bae3de884e77f263b6b68ecf60501d439a9f7e1a0f25d3156fc91d6d0e61911ea316affc80d580db7b1652f87501b9
+  metadata.gz: 961d27e13b64d05ba621303df804df9cf6ba8baea87a1220da968722e6122d1c9eda20db2a32fd537652dc2e1ce984f820f3c998737b07ba045dd649c0606e1d
+  data.tar.gz: 8a91c62cfb49af32a0dde25422ed7fc8e75e2562ce64dc21acec851d673314f4323113383033ad35ee5c708db21dcc0931bdade6106aedccf826beefe7aa2f1a

data/README.md

@@ -2,6 +2,8 @@
 
 This uses Ruby's AWS SDK to send images and a prompt to an LLM in Amazon's Bedrock to generate Alt Text for the images.
 
+The client uses Bedrock's `converse` API and currently supports JPEG and PNG inputs.
+
 ### Ruby Client Usage
 
 This gem uses imagemagick to resize large images, so you will need to install imagemagick:
@@ -39,19 +41,29 @@ gem install alt_text
 Instantiate the client with injected AWS credentials:
 
 ```
-client = AltText::Client.new {
-  access_key_id: ENV['YOUR_ACCESS_KEY_ID'],
-  secret_access_key: ENV['YOUR_SECRET_ACCESS_KEY'],
+client = AltText::Client.new(
+  access_key: ENV['AWS_ACCESS_KEY_ID'],
+  secret_key: ENV['AWS_SECRET_ACCESS_KEY'],
   region: 'us-east-1'
-}
+)
 ```
 
 Call the `#process_image` method with the image path, prompt, and LLM ID as arguments:
 
 ```
-client.process_image('folder/image.png', 'Please generate alt text', 'sonnet3.51`)
+client.process_image(
+  'folder/image.png',
+  prompt: 'Please generate alt text',
+  model_id: 'default'
+)
 ```
 
+Supported image types:
+
+- `.jpg`
+- `.jpeg`
+- `.png`
+
 *Note: A sample prompt can be found in `prompt.txt`.*
 
 ### CLI Usage

data/lib/alt_text/client.rb

@@ -2,7 +2,7 @@
 
 require 'aws-sdk-bedrockruntime'
 require 'mini_magick'
-require 'base64'
+require 'marcel'
 
 module AltText
   class Client
@@ -16,36 +16,59 @@ module AltText
 
     def process_image(image_path, prompt:, model_id:)
       model_id = AltText::LLMRegistry.resolve(model_id)
+      image_format = image_format_for(image_path)
       tmp_image = resize_if_needed(image_path)
 
-      encoded_image = Base64.strict_encode64(File.binread(tmp_image))
+      image_bytes = File.binread(tmp_image)
       tmp_image.close! if tmp_image.is_a?(Tempfile)
 
-      payload = {
-        messages: [
-          { role: 'user',
-            content: [
-              { type: 'image',
-                source:
-                  { type: 'base64',
-                    media_type: 'image/jpeg',
-                    data: encoded_image } },
-              { type: 'text',
-                text: prompt }
-            ] }
-        ],
-        max_tokens: 10_000,
-        anthropic_version: 'bedrock-2023-05-31'
-      }
-
-      response = @client.invoke_model(model_id: model_id,
-                                      content_type: 'application/json',
-                                      body: payload.to_json)
-      JSON.parse(response.body.read)['content'][0]['text']
+      messages = [
+        {
+          role: 'user',
+          content: [
+            {
+              image: {
+                format: image_format,
+                source: {
+                  bytes: image_bytes
+                }
+              }
+            },
+            {
+              text: prompt
+            }
+          ]
+        }
+      ]
+
+      # The `converse` method of the Bedrock Ruby SDK is used to interact with
+      # LLM models in a standardized way, using a "messages" schema that supports
+      # text, images, and tool calls. Unlike `invoke_model`, which requires
+      # model-specific payloads.  Note that this prevents fine-grained control
+      # of image processing parameters that some models may support.
+      #
+      # Examples of supported models:
+      #   - Amazon Nova Pro (supports text and images)
+      #   - Amazon Nova Lite (supports text and images)
+      #   - Anthropic Claude / Opus (supports text and images)
+      response = @client.converse(model_id: model_id,
+                                  messages: messages)
+
+      response.output.message.content.first.text
     end
 
     private
 
+      def image_format_for(path)
+        content_type = Marcel::MimeType.for(Pathname.new(path))
+        case content_type
+        when 'image/jpeg' then 'jpeg'
+        when 'image/png'  then 'png'
+        else
+          raise ArgumentError, "Unsupported image type: #{content_type || 'unknown'}"
+        end
+      end
+
       def resize_if_needed(file)
         if File.size(file) < 4_000_000
           file

data/lib/alt_text/llm_registry.rb

@@ -1,12 +1,14 @@
 # frozen_string_literal: true
 
 module AltText
+  # Bedrock model IDs evolve over time, so this mapping is intentionally
+  # centralized and easy to update as models are added, renamed, or retired.
+  # Entries here are expected to work with the Bedrock Ruby SDK `converse` API.
   class LLMRegistry
     LLM_MAP = {
-      'default' => 'anthropic.claude-3-5-sonnet-20240620-v1:0',
-      'sonnet3.51' => 'anthropic.claude-3-5-sonnet-20240620-v1:0',
-      'sonnet3.52' => 'anthropic.claude-3-5-sonnet-20241022-v2:0',
-      'sonnet3.571' => 'us.anthropic.claude-3-7-sonnet-20250219-v1:0',
+      'default' => 'us.amazon.nova-pro-v1:0',
+      'novalite' => 'amazon.nova-lite-v1:0',
+      'sonnet4.5' => 'anthropic.claude-sonnet-4-5',
       'novapro' => 'us.amazon.nova-pro-v1:0'
     }.freeze
 

data/lib/alt_text/version.rb

@@ -1,5 +1,5 @@
 # frozen_string_literal: true
 
 module AltText
-  VERSION = '0.1.1'
+  VERSION = '0.2.0'
 end

metadata

@@ -1,13 +1,13 @@
 --- !ruby/object:Gem::Specification
 name: alt_text
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.2.0
 platform: ruby
 authors:
 - Alex Kiessling
 bindir: bin
 cert_chain: []
-date: 2025-11-18 00:00:00.000000000 Z
+date: 2026-03-11 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: aws-sdk-bedrockruntime
@@ -37,6 +37,20 @@ dependencies:
     - - "~>"
       - !ruby/object:Gem::Version
         version: 3.1.8
+- !ruby/object:Gem::Dependency
+  name: marcel
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.1'
+  type: :runtime
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - "~>"
+      - !ruby/object:Gem::Version
+        version: '1.1'
 - !ruby/object:Gem::Dependency
   name: mini_magick
   requirement: !ruby/object:Gem::Requirement