…s for PhotoBuilder

- Wire PhotoBuilder CTA (camera icon) into dist_files_controller for each page
- Add prefillMessage support to chat-based-content-editor controller for the
  "Embed generated images into content page" flow
- Register PhotoBuilder entities in doctrine.yaml and generate migration
  for photo_sessions and photo_images tables
- Add Vitest tests for photo_builder_controller (23 tests) and
  photo_image_controller (25 tests)
- Add tests for PhotoBuilder CTA in dist_files_controller (5 tests) and
  prefillMessage in chat_based_content_editor_controller (3 tests)

Co-authored-by: Cursor <cursoragent@cursor.com>

…ms, image serving

- Replace invalid placeholder strings (___SESSION_ID___) in Twig template
  with dummy UUIDs that satisfy Symfony route parameter requirements
- Use output_format instead of response_format for gpt-image-1 API
  (response_format is a dall-e-2/dall-e-3 parameter)
- Generate image URLs via Symfony router to include locale prefix,
  fixing broken image display due to missing /{_locale}/ in path
- Update vertical-wiring.md with PhotoBuilder facade dependencies
- Update corresponding unit and frontend tests

Co-authored-by: Cursor <cursoragent@cursor.com>

…dback, TestHarness

- Use etfswui-* styleguide classes on PhotoBuilder page (buttons, cards, forms)
- Add cursor-pointer to all CTAs via styleguide button classes
- Extract Remote Assets sidebar to @common.presentation/_remote_asset_browser_sidebar.html.twig
- Include shared partial in chat_based_content_editor and photo_builder
- Show 'Upload has been finished' banner on PhotoBuilder when upload completes (auto-hide 5s)
- Add PhotoBuilder TestHarness: FakePromptGenerator, FakeImageGenerator, env toggles
- PHOTO_BUILDER_SIMULATE_IMAGE_PROMPT_GENERATION and PHOTO_BUILDER_SIMULATE_IMAGE_GENERATION in .env
- Fix OpenAI image API (output_format for gpt-image-1), poll image URLs, Stimulus action wiring
- IMAGE_COUNT=1 for faster testing; docs/frontendbook.md and vertical-wiring.md updates

Co-authored-by: Cursor <cursoragent@cursor.com>

…er query params

- Show 'Upload has been finished' banner when image-card upload succeeds (not only sidebar)
- Regenerate prompts: overlay + spinner, clear unprotected prompt textareas on start
- Hide overlay when poll returns generating state; add regenerating_prompts translation
- Language switcher: preserve query string (page, conversationId) when switching locale on photo builder

Co-authored-by: Cursor <cursoragent@cursor.com>

- Add uploadedToMediaStoreAt to PhotoImage to track S3 uploads
- Persist upload state in uploadToMediaStore endpoint; idempotent when already uploaded
- Include uploadedToMediaStore in poll response
- Change embedIntoPage to async: upload non-uploaded images first, show
  'Uploading images, please wait...' overlay, then navigate on success
- Add translations for uploading_images (EN/DE)
- Reset uploadedToMediaStoreAt when image is regenerated

Co-authored-by: Cursor <cursoragent@cursor.com>

…prompt

- Add uploadedFileName to PhotoImage for hash-prefixed S3 names in embed message
- Pass keepImageIds from regenerate prompts to handler; skip regenerating kept images
- Only dispatch image generation for changed prompts, not kept ones
- Clear uploaded state when prompt is regenerated

Co-authored-by: Cursor <cursoragent@cursor.com>

Dispatch clearPromptIfNotKept event on each child card element instead
of the parent — DOM events bubble upward, so dispatching on the parent
never reached child controllers. Also show "Generating..." text with
pulse animation immediately on non-kept prompts, disable buttons during
regeneration, and document the parent-to-child event pattern in
frontendbook.

Co-authored-by: Cursor <cursoragent@cursor.com>

…ider configuration

Introduce a two-tier LLM configuration system: content editing (OpenAI-only)
and PhotoBuilder (OpenAI or Google Gemini). Projects can either reuse content
editing settings for image generation or configure a dedicated provider/key.

- Rename llmApiKey/llmModelProvider to contentEditing* scope across entity,
  DTOs, facades, controllers, templates, and tests
- Add nullable photoBuilder* LLM fields with fallback to content editing
- Extend LlmModelProvider enum with Google case and model selection methods
- Extend LlmModelName enum with gpt-image-1, gemini-3-pro-preview,
  gemini-3-pro-image-preview
- Implement GeminiImageGenerator adapter and ImageGeneratorFactory
- Parameterize ImagePromptAgent to support both OpenAI and Gemini providers
- Add Google API key verification via Gemini models endpoint
- Add PhotoBuilder LLM settings UI (Option A: reuse / Option B: dedicated)
  with provider selection, key input, verification, and one-click reuse
- Display active provider and model names on PhotoBuilder page
- Add docs/llm-usage-book.md documenting all LLM concerns and configuration

Co-authored-by: Cursor <cursoragent@cursor.com>

The Stimulus controller searched for the provider radio only within
its own element, missing sibling radios in the same fieldset.  This
caused Google Gemini keys to be verified against OpenAI, always
failing.  Widen the lookup scope to the closest fieldset/form ancestor.

Co-authored-by: Cursor <cursoragent@cursor.com>

…nly)

Lo-res mode (1K, default) enables faster iteration; hi-res mode (2K)
produces higher quality output. The toggle is only shown when the
effective PhotoBuilder provider is Google Gemini, since OpenAI always
generates 1024x1024. Switching modes re-generates all images client-side
using current prompts at the new resolution without a page reload.

Co-authored-by: Cursor <cursoragent@cursor.com>

Remove fixed container_name to allow scaling, add deploy.replicas: 5.

Co-authored-by: Cursor <cursoragent@cursor.com>

After uploading images to S3 via the "Embed into page" action, poll
the remote asset manifests until all uploaded filenames are confirmed
available before redirecting. This prevents the content editor from
referencing images that haven't propagated to the CDN yet.

- Add findAvailableFileNames() to RemoteContentAssetsFacade (basename
  matching against merged manifests) so the logic stays in the
  RemoteContentAssets vertical
- Add thin POST endpoint in PhotoBuilderController that delegates to
  the facade and returns { available, allAvailable }
- Frontend polls every 3s for up to 90s, showing a spinner overlay
- Includes PHP unit tests, frontend tests, and EN/DE translations

Co-authored-by: Cursor <cursoragent@cursor.com>

Use the faster and cheaper Flash model for generating image prompts
in PhotoBuilder when the Google provider is selected. Pro remains
the main text model for content editing.

Co-authored-by: Cursor <cursoragent@cursor.com>

NeuronAI's Gemini provider only checks parts[0] for functionCall,
but Gemini 3 models now return text/thought parts before functionCall
parts, causing all tool calls to be silently missed (0 prompts).

Introduce PatchedGemini provider that scans all parts and reindexes
the tools array after filtering out non-functionCall parts.

Co-authored-by: Cursor <cursoragent@cursor.com>

…from PhotoBuilder

- Add in-progress and success feedback near single-image Upload CTA
  (spinner + 'Uploading…', then checkmark + 'Uploaded'; dispatch
  uploadComplete/uploadFailed to card)
- Fix upload/success spans always visible: use wrapper spans so only
  'hidden' is toggled (no inline-flex vs hidden conflict)
- Find card from event target for reliable completion/failure delivery
- Add translations: uploading_to_media_store, uploaded_to_media_store
- Stack Regenerate and Upload buttons vertically (flex-col) to fit space

Co-authored-by: Cursor <cursoragent@cursor.com>

…ons textarea

- Track lastAppliedUserPrompt; only apply server userPrompt from poll when
  current value matches it (or first load) so local edits are not overwritten
- Add unit test: user edit preserved when poll runs with textarea unfocused

Co-authored-by: Cursor <cursoragent@cursor.com>

The change-detection optimization skipped dispatching stateChanged to
children when per-image data was unchanged, but children rely on that
event to re-read the parent's data-photo-builder-generating attribute
and enable/disable their Regenerate and Upload buttons. Now tracks
anyGenerating transitions and force-dispatches to all cards when it
changes.

Co-authored-by: Cursor <cursoragent@cursor.com>

…uster

The cache-buster is only applied when img.src is actually set (after a
regeneration cycle clears lastSetImageUrl), so repeated polls with the
same URL still skip re-assignment and avoid redundant fetches.

Co-authored-by: Cursor <cursoragent@cursor.com>

…poll

The promptAwaitingRegenerate logic only accepted new prompts when status
was "pending" or "generating". If image generation completed within one
poll cycle, status was already "completed" and the condition never
matched, leaving the textarea permanently stuck.

Now compares the incoming prompt against the saved pre-regeneration
prompt instead of checking status, correctly handling both fast
completions and stale old-data polls.

Co-authored-by: Cursor <cursoragent@cursor.com>

- Redesign Preview Pages and PhotoBuilder CTA as distinct styleguide cards
- Add translatable Edit HTML / Preview labels with proper icons
- Make filenames clickable links to preview URLs
- Move AI model info into Behind the Scenes section
- Translate embed prefill message for German locale
- Right-align Content Editor action buttons, use styleguide classes
- Use etfswui-card-back-link for PhotoBuilder back link
- Remove unused flex-row sidebar layout so content uses full width

Co-authored-by: Cursor <cursoragent@cursor.com>