feat: admin image generation via Cloudflare Workers AI

- Add cfai/image.go: ImageGenClient with GenerateImage, GenerateImageFromReference, AllImageModels (9 models)
- Add handlers_image.go: GET /api/admin/image-gen/models + POST /api/admin/image-gen (JSON + multipart)
- Wire ImageGen client in main.go + server.go Dependencies
- Add admin/image-gen SvelteKit page: type toggle, model selector, prompt, reference img2img, advanced options, result panel, history, save-as-cover, download
- Add SvelteKit proxy route api/admin/image-gen forwarding to Go backend
- Add admin_nav_image_gen message key to all 5 locale files
- Add Image Gen nav link to admin layout

backend/cmd/backend/main.go

@@ -26,6 +26,7 @@ import (
 	"github.com/hibiken/asynq"
 	"github.com/libnovel/backend/internal/asynqqueue"
 	"github.com/libnovel/backend/internal/backend"
+	"github.com/libnovel/backend/internal/cfai"
 	"github.com/libnovel/backend/internal/config"
 	"github.com/libnovel/backend/internal/kokoro"
 	"github.com/libnovel/backend/internal/meili"
@@ -114,6 +115,24 @@ func run() error {
 		log.Info("POCKET_TTS_URL not set — pocket-tts voices unavailable in backend")
 	}
 
+	// ── Cloudflare Workers AI (voice sample generation + audio-stream live TTS) ──
+	var cfaiClient cfai.Client
+	if cfg.CFAI.AccountID != "" && cfg.CFAI.APIToken != "" {
+		cfaiClient = cfai.New(cfg.CFAI.AccountID, cfg.CFAI.APIToken, cfg.CFAI.Model)
+		log.Info("cloudflare AI TTS enabled", "model", cfg.CFAI.Model)
+	} else {
+		log.Info("CFAI_ACCOUNT_ID/CFAI_API_TOKEN not set — CF AI voices unavailable in backend")
+	}
+
+	// ── Cloudflare Workers AI Image Generation ────────────────────────────────
+	var imageGenClient cfai.ImageGenClient
+	if cfg.CFAI.AccountID != "" && cfg.CFAI.APIToken != "" {
+		imageGenClient = cfai.NewImageGen(cfg.CFAI.AccountID, cfg.CFAI.APIToken)
+		log.Info("cloudflare AI image generation enabled")
+	} else {
+		log.Info("CFAI_ACCOUNT_ID/CFAI_API_TOKEN not set — image generation unavailable")
+	}
+
 	// ── Meilisearch (search reads only; indexing is the runner's job) ────────
 	var searchIndex meili.Client
 	if cfg.Meilisearch.URL != "" {
@@ -163,6 +182,8 @@ func run() error {
 			SearchIndex:      searchIndex,
 			Kokoro:           kokoroClient,
 			PocketTTS:        pocketTTSClient,
+			CFAI:             cfaiClient,
+			ImageGen:         imageGenClient,
 			Log:              log,
 		},
 	)

backend/cmd/runner/main.go

@@ -23,6 +23,7 @@ import (
 	"github.com/getsentry/sentry-go"
 	"github.com/libnovel/backend/internal/asynqqueue"
 	"github.com/libnovel/backend/internal/browser"
+	"github.com/libnovel/backend/internal/cfai"
 	"github.com/libnovel/backend/internal/config"
 	"github.com/libnovel/backend/internal/kokoro"
 	"github.com/libnovel/backend/internal/libretranslate"
@@ -130,6 +131,15 @@ func run() error {
 		log.Warn("POCKET_TTS_URL not set — pocket-tts voice tasks will fail")
 	}
 
+	// ── Cloudflare Workers AI ────────────────────────────────────────────────
+	var cfaiClient cfai.Client
+	if cfg.CFAI.AccountID != "" && cfg.CFAI.APIToken != "" {
+		cfaiClient = cfai.New(cfg.CFAI.AccountID, cfg.CFAI.APIToken, cfg.CFAI.Model)
+		log.Info("cloudflare AI TTS enabled", "model", cfg.CFAI.Model)
+	} else {
+		log.Info("CFAI_ACCOUNT_ID/CFAI_API_TOKEN not set — CF AI voice tasks will fail")
+	}
+
 	// ── LibreTranslate ──────────────────────────────────────────────────────
 	ltClient := libretranslate.New(cfg.LibreTranslate.URL, cfg.LibreTranslate.APIKey)
 	if ltClient != nil {
@@ -191,6 +201,7 @@ func run() error {
 		Novel:            novel,
 		Kokoro:           kokoroClient,
 		PocketTTS:        pocketTTSClient,
+		CFAI:             cfaiClient,
 		LibreTranslate:   ltClient,
 		Log:              log,
 	}

backend/internal/backend/handlers.go

@@ -44,6 +44,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/libnovel/backend/internal/cfai"
 	"github.com/libnovel/backend/internal/domain"
 	"github.com/libnovel/backend/internal/kokoro"
 	"github.com/libnovel/backend/internal/meili"
@@ -774,7 +775,13 @@ func (s *Server) handleAudioStream(w http.ResponseWriter, r *http.Request) {
 	// Open the TTS stream (WAV or MP3 depending on format param).
 	var audioStream io.ReadCloser
 	if format == "wav" {
-		if pockettts.IsPocketTTSVoice(voice) {
+		if cfai.IsCFAIVoice(voice) {
+			if s.deps.CFAI == nil {
+				jsonError(w, http.StatusServiceUnavailable, "cloudflare AI TTS not configured")
+				return
+			}
+			audioStream, err = s.deps.CFAI.StreamAudioWAV(r.Context(), text, voice)
+		} else if pockettts.IsPocketTTSVoice(voice) {
 			if s.deps.PocketTTS == nil {
 				jsonError(w, http.StatusServiceUnavailable, "pocket-tts not configured")
 				return
@@ -788,7 +795,13 @@ func (s *Server) handleAudioStream(w http.ResponseWriter, r *http.Request) {
 			audioStream, err = s.deps.Kokoro.StreamAudioWAV(r.Context(), text, voice)
 		}
 	} else {
-		if pockettts.IsPocketTTSVoice(voice) {
+		if cfai.IsCFAIVoice(voice) {
+			if s.deps.CFAI == nil {
+				jsonError(w, http.StatusServiceUnavailable, "cloudflare AI TTS not configured")
+				return
+			}
+			audioStream, err = s.deps.CFAI.StreamAudioMP3(r.Context(), text, voice)
+		} else if pockettts.IsPocketTTSVoice(voice) {
 			if s.deps.PocketTTS == nil {
 				jsonError(w, http.StatusServiceUnavailable, "pocket-tts not configured")
 				return
@@ -1343,6 +1356,9 @@ func (s *Server) handlePresignVoiceSample(w http.ResponseWriter, r *http.Request
 	}
 
 	key := kokoro.VoiceSampleKey(voice)
+	if cfai.IsCFAIVoice(voice) {
+		key = cfai.VoiceSampleKey(voice)
+	}
 
 	// Generate sample on demand when it is not in MinIO yet.
 	if !s.deps.AudioStore.AudioExists(r.Context(), key) {
@@ -1352,7 +1368,13 @@ func (s *Server) handlePresignVoiceSample(w http.ResponseWriter, r *http.Request
 			mp3 []byte
 			err error
 		)
-		if pockettts.IsPocketTTSVoice(voice) {
+		if cfai.IsCFAIVoice(voice) {
+			if s.deps.CFAI == nil {
+				jsonError(w, http.StatusServiceUnavailable, "cloudflare AI TTS not configured")
+				return
+			}
+			mp3, err = s.deps.CFAI.GenerateAudio(r.Context(), voiceSampleText, voice)
+		} else if pockettts.IsPocketTTSVoice(voice) {
 			if s.deps.PocketTTS == nil {
 				jsonError(w, http.StatusServiceUnavailable, "pocket-tts not configured")
 				return

backend/internal/backend/handlers_image.go

@@ -0,0 +1,234 @@
+package backend
+
+import (
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+
+	"github.com/libnovel/backend/internal/cfai"
+)
+
+// handleAdminImageGenModels handles GET /api/admin/image-gen/models.
+// Returns the list of supported Cloudflare AI image generation models.
+func (s *Server) handleAdminImageGenModels(w http.ResponseWriter, r *http.Request) {
+	if s.deps.ImageGen == nil {
+		jsonError(w, http.StatusServiceUnavailable, "image generation not configured (CFAI_ACCOUNT_ID/CFAI_API_TOKEN missing)")
+		return
+	}
+	models := s.deps.ImageGen.Models()
+	writeJSON(w, 0, map[string]any{"models": models})
+}
+
+// imageGenRequest is the JSON body for POST /api/admin/image-gen.
+type imageGenRequest struct {
+	// Prompt is the text description of the desired image.
+	Prompt string `json:"prompt"`
+
+	// Model is the CF Workers AI model ID (e.g. "@cf/black-forest-labs/flux-2-dev").
+	// Defaults to the recommended model for the given type.
+	Model string `json:"model"`
+
+	// Type is either "cover" or "chapter".
+	Type string `json:"type"`
+
+	// Slug is the book slug. Required for cover; required for chapter.
+	Slug string `json:"slug"`
+
+	// Chapter number (1-based). Required when type == "chapter".
+	Chapter int `json:"chapter"`
+
+	// ReferenceImageB64 is an optional base64-encoded PNG/JPEG reference image.
+	// When present the img2img path is used.
+	ReferenceImageB64 string `json:"reference_image_b64"`
+
+	// NumSteps overrides inference steps (default 20).
+	NumSteps int `json:"num_steps"`
+
+	// Width / Height override output dimensions (0 = model default).
+	Width  int `json:"width"`
+	Height int `json:"height"`
+
+	// Guidance overrides prompt guidance scale (0 = model default).
+	Guidance float64 `json:"guidance"`
+
+	// Strength for img2img: 0.0–1.0, default 0.75.
+	Strength float64 `json:"strength"`
+
+	// SaveToCover when true stores the result as the book cover in MinIO
+	// (overwriting any existing cover) and sets the book's cover URL.
+	// Only valid when type == "cover".
+	SaveToCover bool `json:"save_to_cover"`
+}
+
+// imageGenResponse is the JSON body returned by POST /api/admin/image-gen.
+type imageGenResponse struct {
+	// ImageB64 is the generated image as a base64-encoded PNG string.
+	ImageB64 string `json:"image_b64"`
+	// ContentType is "image/png" or "image/jpeg".
+	ContentType string `json:"content_type"`
+	// Saved indicates whether the image was persisted to MinIO.
+	Saved bool `json:"saved"`
+	// CoverURL is the URL the cover is now served from (only set when Saved==true).
+	CoverURL string `json:"cover_url,omitempty"`
+	// Model is the model that was used.
+	Model string `json:"model"`
+	// Bytes is the raw image size in bytes.
+	Bytes int `json:"bytes"`
+}
+
+// handleAdminImageGen handles POST /api/admin/image-gen.
+//
+// Generates an image using Cloudflare Workers AI and optionally stores it.
+// Multipart/form-data is also accepted so the reference image can be uploaded
+// directly; otherwise the reference is expected as base64 JSON.
+func (s *Server) handleAdminImageGen(w http.ResponseWriter, r *http.Request) {
+	if s.deps.ImageGen == nil {
+		jsonError(w, http.StatusServiceUnavailable, "image generation not configured (CFAI_ACCOUNT_ID/CFAI_API_TOKEN missing)")
+		return
+	}
+
+	var req imageGenRequest
+	var refImageData []byte
+
+	ct := r.Header.Get("Content-Type")
+	if strings.HasPrefix(ct, "multipart/form-data") {
+		// Multipart: parse JSON fields from a "json" part + optional "reference" file part.
+		if err := r.ParseMultipartForm(32 << 20); err != nil {
+			jsonError(w, http.StatusBadRequest, "parse multipart: "+err.Error())
+			return
+		}
+		if jsonPart := r.FormValue("json"); jsonPart != "" {
+			if err := json.Unmarshal([]byte(jsonPart), &req); err != nil {
+				jsonError(w, http.StatusBadRequest, "parse json field: "+err.Error())
+				return
+			}
+		}
+		if f, _, err := r.FormFile("reference"); err == nil {
+			defer f.Close()
+			refImageData, _ = io.ReadAll(f)
+		}
+	} else {
+		if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+			jsonError(w, http.StatusBadRequest, "parse body: "+err.Error())
+			return
+		}
+		if req.ReferenceImageB64 != "" {
+			var decErr error
+			refImageData, decErr = base64.StdEncoding.DecodeString(req.ReferenceImageB64)
+			if decErr != nil {
+				// Try std without padding
+				refImageData, decErr = base64.RawStdEncoding.DecodeString(req.ReferenceImageB64)
+				if decErr != nil {
+					jsonError(w, http.StatusBadRequest, "decode reference_image_b64: "+decErr.Error())
+					return
+				}
+			}
+		}
+	}
+
+	if strings.TrimSpace(req.Prompt) == "" {
+		jsonError(w, http.StatusBadRequest, "prompt is required")
+		return
+	}
+	if req.Type != "cover" && req.Type != "chapter" {
+		jsonError(w, http.StatusBadRequest, `type must be "cover" or "chapter"`)
+		return
+	}
+	if req.Slug == "" {
+		jsonError(w, http.StatusBadRequest, "slug is required")
+		return
+	}
+	if req.Type == "chapter" && req.Chapter <= 0 {
+		jsonError(w, http.StatusBadRequest, "chapter must be > 0 when type is chapter")
+		return
+	}
+
+	// Resolve model
+	model := cfai.ImageModel(req.Model)
+	if model == "" {
+		if req.Type == "cover" {
+			model = cfai.DefaultImageModel
+		} else {
+			model = cfai.ImageModelFlux2Klein4B
+		}
+	}
+
+	imgReq := cfai.ImageRequest{
+		Prompt:   req.Prompt,
+		Model:    model,
+		NumSteps: req.NumSteps,
+		Width:    req.Width,
+		Height:   req.Height,
+		Guidance: req.Guidance,
+		Strength: req.Strength,
+	}
+
+	s.deps.Log.Info("admin: image gen requested",
+		"type", req.Type, "slug", req.Slug, "chapter", req.Chapter,
+		"model", model, "has_reference", len(refImageData) > 0)
+
+	var imgData []byte
+	var genErr error
+	if len(refImageData) > 0 {
+		imgData, genErr = s.deps.ImageGen.GenerateImageFromReference(r.Context(), imgReq, refImageData)
+	} else {
+		imgData, genErr = s.deps.ImageGen.GenerateImage(r.Context(), imgReq)
+	}
+	if genErr != nil {
+		s.deps.Log.Error("admin: image gen failed", "err", genErr)
+		jsonError(w, http.StatusBadGateway, "image generation failed: "+genErr.Error())
+		return
+	}
+
+	contentType := sniffImageContentType(imgData)
+
+	// ── Optional persistence ──────────────────────────────────────────────────
+	var saved bool
+	var coverURL string
+
+	if req.SaveToCover && req.Type == "cover" && s.deps.CoverStore != nil {
+		if err := s.deps.CoverStore.PutCover(r.Context(), req.Slug, imgData, contentType); err != nil {
+			s.deps.Log.Error("admin: save generated cover failed", "slug", req.Slug, "err", err)
+			// Non-fatal: still return the image
+		} else {
+			saved = true
+			coverURL = fmt.Sprintf("/api/cover/local/%s", req.Slug)
+			s.deps.Log.Info("admin: generated cover saved", "slug", req.Slug, "bytes", len(imgData))
+		}
+	}
+
+	// Encode result as base64
+	b64 := base64.StdEncoding.EncodeToString(imgData)
+
+	writeJSON(w, 0, imageGenResponse{
+		ImageB64:    b64,
+		ContentType: contentType,
+		Saved:       saved,
+		CoverURL:    coverURL,
+		Model:       string(model),
+		Bytes:       len(imgData),
+	})
+}
+
+// sniffImageContentType returns the MIME type of the image bytes.
+func sniffImageContentType(data []byte) string {
+	if len(data) >= 4 {
+		// PNG: 0x89 P N G
+		if data[0] == 0x89 && data[1] == 0x50 && data[2] == 0x4e && data[3] == 0x47 {
+			return "image/png"
+		}
+		// JPEG: FF D8 FF
+		if data[0] == 0xFF && data[1] == 0xD8 && data[2] == 0xFF {
+			return "image/jpeg"
+		}
+		// WebP: RIFF....WEBP
+		if len(data) >= 12 && data[0] == 'R' && data[1] == 'I' && data[2] == 'F' && data[3] == 'F' &&
+			data[8] == 'W' && data[9] == 'E' && data[10] == 'B' && data[11] == 'P' {
+			return "image/webp"
+		}
+	}
+	return "image/png"
+}

backend/internal/backend/server.go

@@ -30,6 +30,7 @@ import (
 
 	sentryhttp "github.com/getsentry/sentry-go/http"
 	"github.com/libnovel/backend/internal/bookstore"
+	"github.com/libnovel/backend/internal/cfai"
 	"github.com/libnovel/backend/internal/domain"
 	"github.com/libnovel/backend/internal/kokoro"
 	"github.com/libnovel/backend/internal/meili"
@@ -69,6 +70,12 @@ type Dependencies struct {
 	// PocketTTS is the pocket-tts client (used for voice list only in the backend;
 	// audio generation is done by the runner).
 	PocketTTS pockettts.Client
+	// CFAI is the Cloudflare Workers AI TTS client (used for voice sample
+	// generation and audio-stream live TTS; audio task generation is done by the runner).
+	CFAI cfai.Client
+	// ImageGen is the Cloudflare Workers AI image generation client.
+	// If nil, image generation endpoints return 503.
+	ImageGen cfai.ImageGenClient
 	// Log is the structured logger.
 	Log *slog.Logger
 }
@@ -179,6 +186,10 @@ func (s *Server) ListenAndServe(ctx context.Context) error {
 	mux.HandleFunc("POST /api/admin/audio/bulk", s.handleAdminAudioBulk)
 	mux.HandleFunc("POST /api/admin/audio/cancel-bulk", s.handleAdminAudioCancelBulk)
 
+	// Admin image generation endpoints
+	mux.HandleFunc("GET /api/admin/image-gen/models", s.handleAdminImageGenModels)
+	mux.HandleFunc("POST /api/admin/image-gen", s.handleAdminImageGen)
+
 	// Voices list
 	mux.HandleFunc("GET /api/voices", s.handleVoices)
 
@@ -338,6 +349,23 @@ func (s *Server) voices(ctx context.Context) []domain.Voice {
 		}
 	}
 
+	// ── Cloudflare AI voices ──────────────────────────────────────────────────
+	if s.deps.CFAI != nil {
+		for _, speaker := range cfai.Speakers() {
+			gender := "m"
+			if cfai.IsFemale(speaker) {
+				gender = "f"
+			}
+			result = append(result, domain.Voice{
+				ID:     cfai.VoiceID(speaker),
+				Engine: "cfai",
+				Lang:   "en",
+				Gender: gender,
+			})
+		}
+		s.deps.Log.Info("backend: loaded CF AI voices", "count", len(cfai.Speakers()))
+	}
+
 	s.voiceMu.Lock()
 	s.cachedVoices = result
 	s.voiceMu.Unlock()

backend/internal/cfai/client.go

@@ -0,0 +1,214 @@
+// Package cfai provides a client for Cloudflare Workers AI Text-to-Speech models.
+//
+// The Cloudflare Workers AI REST API is used to run TTS models:
+//
+//	POST https://api.cloudflare.com/client/v4/accounts/{accountID}/ai/run/{model}
+//	Authorization: Bearer {apiToken}
+//	Content-Type:  application/json
+//	{ "text": "...", "speaker": "luna" }
+//
+//	→ 200 audio/mpeg — raw MP3 bytes
+//
+// Currently supported model: @cf/deepgram/aura-2-en (40 English speakers).
+// Voice IDs are prefixed with "cfai:" to distinguish them from Kokoro/pocket-tts
+// voices (e.g. "cfai:luna", "cfai:orion").
+//
+// The API is batch-only (no streaming), so GenerateAudio waits for the full
+// response. There is no 100-second Cloudflare proxy timeout because we are
+// calling the Cloudflare API directly, not routing through a Cloudflare-proxied
+// homelab tunnel.
+package cfai
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+)
+
+const (
+	// DefaultModel is the Cloudflare Workers AI TTS model used by default.
+	DefaultModel = "@cf/deepgram/aura-2-en"
+
+	// voicePrefix is the prefix used to namespace CF AI voice IDs.
+	voicePrefix = "cfai:"
+)
+
+// aura2Speakers is the exhaustive list of speakers supported by aura-2-en.
+var aura2Speakers = []string{
+	"amalthea", "andromeda", "apollo", "arcas", "aries", "asteria",
+	"athena", "atlas", "aurora", "callista", "cora", "cordelia",
+	"delia", "draco", "electra", "harmonia", "helena", "hera",
+	"hermes", "hyperion", "iris", "janus", "juno", "jupiter",
+	"luna", "mars", "minerva", "neptune", "odysseus", "ophelia",
+	"orion", "orpheus", "pandora", "phoebe", "pluto", "saturn",
+	"thalia", "theia", "vesta", "zeus",
+}
+
+// femaleSpeakers is the set of aura-2-en speaker names that are female voices.
+var femaleSpeakers = map[string]struct{}{
+	"amalthea": {}, "andromeda": {}, "aries": {}, "asteria": {},
+	"athena": {}, "aurora": {}, "callista": {}, "cora": {},
+	"cordelia": {}, "delia": {}, "electra": {}, "harmonia": {},
+	"helena": {}, "hera": {}, "iris": {}, "juno": {},
+	"luna": {}, "minerva": {}, "ophelia": {}, "pandora": {},
+	"phoebe": {}, "thalia": {}, "theia": {}, "vesta": {},
+}
+
+// IsCFAIVoice reports whether voice is served by the Cloudflare AI client.
+// CF AI voices use the "cfai:" prefix, e.g. "cfai:luna".
+func IsCFAIVoice(voice string) bool {
+	return strings.HasPrefix(voice, voicePrefix)
+}
+
+// SpeakerName strips the "cfai:" prefix and returns the bare speaker name.
+// If voice is not a CF AI voice the original string is returned unchanged.
+func SpeakerName(voice string) string {
+	return strings.TrimPrefix(voice, voicePrefix)
+}
+
+// VoiceID returns the full voice ID (with prefix) for a bare speaker name.
+func VoiceID(speaker string) string {
+	return voicePrefix + speaker
+}
+
+// VoiceSampleKey returns the MinIO object key for a CF AI voice sample MP3.
+func VoiceSampleKey(voice string) string {
+	safe := strings.Map(func(r rune) rune {
+		if (r >= 'a' && r <= 'z') || (r >= 'A' && r <= 'Z') ||
+			(r >= '0' && r <= '9') || r == '_' || r == '-' {
+			return r
+		}
+		return '_'
+	}, voice)
+	return fmt.Sprintf("_voice-samples/%s.mp3", safe)
+}
+
+// IsFemale reports whether the given CF AI voice ID (with or without prefix)
+// is a female speaker.
+func IsFemale(voice string) bool {
+	speaker := SpeakerName(voice)
+	_, ok := femaleSpeakers[speaker]
+	return ok
+}
+
+// Speakers returns all available bare speaker names for aura-2-en.
+func Speakers() []string {
+	out := make([]string, len(aura2Speakers))
+	copy(out, aura2Speakers)
+	return out
+}
+
+// Client is the interface for interacting with Cloudflare Workers AI TTS.
+type Client interface {
+	// GenerateAudio synthesises text using the given voice (e.g. "cfai:luna")
+	// and returns raw MP3 bytes.
+	GenerateAudio(ctx context.Context, text, voice string) ([]byte, error)
+
+	// StreamAudioMP3 is not natively supported by the CF AI batch API.
+	// It buffers the full response and returns an io.ReadCloser over the bytes,
+	// so callers can use it like a stream without special-casing.
+	StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error)
+
+	// StreamAudioWAV is not natively supported; the CF AI model returns MP3.
+	// This method returns the same MP3 bytes wrapped as an io.ReadCloser.
+	StreamAudioWAV(ctx context.Context, text, voice string) (io.ReadCloser, error)
+
+	// ListVoices returns all available voice IDs (with the "cfai:" prefix).
+	ListVoices(ctx context.Context) ([]string, error)
+}
+
+// httpClient is the concrete CF AI HTTP client.
+type httpClient struct {
+	accountID string
+	apiToken  string
+	model     string
+	http      *http.Client
+}
+
+// New returns a Client for the given Cloudflare account and API token.
+// model defaults to DefaultModel when empty.
+func New(accountID, apiToken, model string) Client {
+	if model == "" {
+		model = DefaultModel
+	}
+	return &httpClient{
+		accountID: accountID,
+		apiToken:  apiToken,
+		model:     model,
+		http:      &http.Client{Timeout: 5 * time.Minute},
+	}
+}
+
+// GenerateAudio calls the Cloudflare Workers AI TTS endpoint and returns MP3 bytes.
+func (c *httpClient) GenerateAudio(ctx context.Context, text, voice string) ([]byte, error) {
+	if text == "" {
+		return nil, fmt.Errorf("cfai: empty text")
+	}
+	speaker := SpeakerName(voice)
+	if speaker == "" {
+		speaker = "luna"
+	}
+
+	body, err := json.Marshal(map[string]any{
+		"text":    text,
+		"speaker": speaker,
+	})
+	if err != nil {
+		return nil, fmt.Errorf("cfai: marshal request: %w", err)
+	}
+
+	url := fmt.Sprintf("https://api.cloudflare.com/client/v4/accounts/%s/ai/run/%s",
+		c.accountID, c.model)
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("cfai: build request: %w", err)
+	}
+	req.Header.Set("Authorization", "Bearer "+c.apiToken)
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("cfai: request: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		body, _ := io.ReadAll(resp.Body)
+		return nil, fmt.Errorf("cfai: server returned %d: %s", resp.StatusCode, strings.TrimSpace(string(body)))
+	}
+
+	mp3, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("cfai: read response: %w", err)
+	}
+	return mp3, nil
+}
+
+// StreamAudioMP3 generates audio and wraps the MP3 bytes as an io.ReadCloser.
+func (c *httpClient) StreamAudioMP3(ctx context.Context, text, voice string) (io.ReadCloser, error) {
+	mp3, err := c.GenerateAudio(ctx, text, voice)
+	if err != nil {
+		return nil, err
+	}
+	return io.NopCloser(bytes.NewReader(mp3)), nil
+}
+
+// StreamAudioWAV generates audio (MP3) and wraps it as an io.ReadCloser.
+// Note: the CF AI aura-2-en model returns MP3 regardless of the method name.
+func (c *httpClient) StreamAudioWAV(ctx context.Context, text, voice string) (io.ReadCloser, error) {
+	return c.StreamAudioMP3(ctx, text, voice)
+}
+
+// ListVoices returns all available CF AI voice IDs (with the "cfai:" prefix).
+func (c *httpClient) ListVoices(_ context.Context) ([]string, error) {
+	ids := make([]string, len(aura2Speakers))
+	for i, s := range aura2Speakers {
+		ids[i] = VoiceID(s)
+	}
+	return ids, nil
+}

backend/internal/cfai/image.go

@@ -0,0 +1,312 @@
+// Image generation via Cloudflare Workers AI text-to-image models.
+//
+// API reference:
+//
+//	POST https://api.cloudflare.com/client/v4/accounts/{accountID}/ai/run/{model}
+//	Authorization: Bearer {apiToken}
+//	Content-Type:  application/json
+//
+// Text-only request (all models):
+//
+//	{ "prompt": "...", "num_steps": 20 }
+//
+// Reference-image request:
+//   - FLUX models: { "prompt": "...", "image_b64": "<base64>" }
+//   - SD img2img:  { "prompt": "...", "image": [r,g,b,a,...], "strength": 0.75 }
+//
+// All models return raw PNG bytes on success (Content-Type: image/png).
+//
+// Recommended models for LibNovel:
+//   - Book covers (no reference): flux-2-dev, flux-2-klein-9b, lucid-origin
+//   - Chapter images (speed):     flux-2-klein-4b, flux-1-schnell
+//   - With reference image:       flux-2-dev, flux-2-klein-9b, sd-v1-5-img2img
+package cfai
+
+import (
+	"bytes"
+	"context"
+	"encoding/base64"
+	"encoding/json"
+	"fmt"
+	"image"
+	_ "image/jpeg" // register JPEG decoder
+	_ "image/png"  // register PNG decoder
+	"io"
+	"net/http"
+	"time"
+)
+
+// ImageModel identifies a Cloudflare Workers AI text-to-image model.
+type ImageModel string
+
+const (
+	// ImageModelFlux2Dev — best quality, multi-reference. Recommended for covers.
+	ImageModelFlux2Dev ImageModel = "@cf/black-forest-labs/flux-2-dev"
+	// ImageModelFlux2Klein9B — 9B params, multi-reference. Good for covers.
+	ImageModelFlux2Klein9B ImageModel = "@cf/black-forest-labs/flux-2-klein-9b"
+	// ImageModelFlux2Klein4B — ultra-fast, unified gen+edit. Recommended for chapters.
+	ImageModelFlux2Klein4B ImageModel = "@cf/black-forest-labs/flux-2-klein-4b"
+	// ImageModelFlux1Schnell — fastest, text-only. Good for quick illustrations.
+	ImageModelFlux1Schnell ImageModel = "@cf/black-forest-labs/flux-1-schnell"
+	// ImageModelSDXLLightning — fast 1024px generation.
+	ImageModelSDXLLightning ImageModel = "@cf/bytedance/stable-diffusion-xl-lightning"
+	// ImageModelSD15Img2Img — explicit img2img with flat RGBA reference.
+	ImageModelSD15Img2Img ImageModel = "@cf/runwayml/stable-diffusion-v1-5-img2img"
+	// ImageModelSDXLBase — Stability AI SDXL base.
+	ImageModelSDXLBase ImageModel = "@cf/stabilityai/stable-diffusion-xl-base-1.0"
+	// ImageModelLucidOrigin — Leonardo AI; strong prompt adherence.
+	ImageModelLucidOrigin ImageModel = "@cf/leonardo/lucid-origin"
+	// ImageModelPhoenix10 — Leonardo AI; accurate text rendering.
+	ImageModelPhoenix10 ImageModel = "@cf/leonardo/phoenix-1.0"
+
+	// DefaultImageModel is the default model for book-cover generation.
+	DefaultImageModel = ImageModelFlux2Dev
+)
+
+// ImageModelInfo describes a single image generation model.
+type ImageModelInfo struct {
+	ID             string   `json:"id"`
+	Label          string   `json:"label"`
+	Provider       string   `json:"provider"`
+	SupportsRef    bool     `json:"supports_ref"`
+	RecommendedFor []string `json:"recommended_for"` // "cover" and/or "chapter"
+	Description    string   `json:"description"`
+}
+
+// AllImageModels returns metadata about every supported image model.
+func AllImageModels() []ImageModelInfo {
+	return []ImageModelInfo{
+		{
+			ID: string(ImageModelFlux2Dev), Label: "FLUX.2 Dev", Provider: "Black Forest Labs",
+			SupportsRef: true, RecommendedFor: []string{"cover"},
+			Description: "Best quality; multi-reference editing. Recommended for book covers.",
+		},
+		{
+			ID: string(ImageModelFlux2Klein9B), Label: "FLUX.2 Klein 9B", Provider: "Black Forest Labs",
+			SupportsRef: true, RecommendedFor: []string{"cover"},
+			Description: "9B parameters with multi-reference support.",
+		},
+		{
+			ID: string(ImageModelFlux2Klein4B), Label: "FLUX.2 Klein 4B", Provider: "Black Forest Labs",
+			SupportsRef: true, RecommendedFor: []string{"chapter"},
+			Description: "Ultra-fast unified gen+edit. Recommended for chapter images.",
+		},
+		{
+			ID: string(ImageModelFlux1Schnell), Label: "FLUX.1 Schnell", Provider: "Black Forest Labs",
+			SupportsRef: false, RecommendedFor: []string{"chapter"},
+			Description: "Fastest inference. Good for quick chapter illustrations.",
+		},
+		{
+			ID: string(ImageModelSDXLLightning), Label: "SDXL Lightning", Provider: "ByteDance",
+			SupportsRef: false, RecommendedFor: []string{"chapter"},
+			Description: "Lightning-fast 1024px images in a few steps.",
+		},
+		{
+			ID: string(ImageModelSD15Img2Img), Label: "SD 1.5 img2img", Provider: "RunwayML",
+			SupportsRef: true, RecommendedFor: []string{"cover", "chapter"},
+			Description: "Explicit img2img: generates from a reference image + prompt.",
+		},
+		{
+			ID: string(ImageModelSDXLBase), Label: "SDXL Base 1.0", Provider: "Stability AI",
+			SupportsRef: false, RecommendedFor: []string{"cover"},
+			Description: "Stable Diffusion XL base model.",
+		},
+		{
+			ID: string(ImageModelLucidOrigin), Label: "Lucid Origin", Provider: "Leonardo AI",
+			SupportsRef: false, RecommendedFor: []string{"cover"},
+			Description: "Highly prompt-responsive; strong graphic design and HD renders.",
+		},
+		{
+			ID: string(ImageModelPhoenix10), Label: "Phoenix 1.0", Provider: "Leonardo AI",
+			SupportsRef: false, RecommendedFor: []string{"cover"},
+			Description: "Exceptional prompt adherence; accurate text rendering.",
+		},
+	}
+}
+
+// ImageRequest is the input to GenerateImage / GenerateImageFromReference.
+type ImageRequest struct {
+	// Prompt is the text description of the desired image.
+	Prompt string
+	// Model is the CF Workers AI model. Defaults to DefaultImageModel when empty.
+	Model ImageModel
+	// NumSteps controls inference quality (default 20). Range: 1–20.
+	NumSteps int
+	// Width and Height in pixels. 0 = model default (typically 1024x1024).
+	Width, Height int
+	// Guidance controls prompt adherence (default 7.5).
+	Guidance float64
+	// Strength for img2img: 0.0 = copy reference, 1.0 = ignore reference (default 0.75).
+	Strength float64
+}
+
+// ImageGenClient generates images via Cloudflare Workers AI.
+type ImageGenClient interface {
+	// GenerateImage creates an image from a text prompt only.
+	// Returns raw PNG bytes.
+	GenerateImage(ctx context.Context, req ImageRequest) ([]byte, error)
+
+	// GenerateImageFromReference creates an image from a text prompt + reference image.
+	// refImage should be PNG or JPEG bytes. Returns raw PNG bytes.
+	GenerateImageFromReference(ctx context.Context, req ImageRequest, refImage []byte) ([]byte, error)
+
+	// Models returns metadata about all supported image models.
+	Models() []ImageModelInfo
+}
+
+// imageGenHTTPClient is the concrete CF AI image generation client.
+type imageGenHTTPClient struct {
+	accountID string
+	apiToken  string
+	http      *http.Client
+}
+
+// NewImageGen returns an ImageGenClient for the given Cloudflare account.
+func NewImageGen(accountID, apiToken string) ImageGenClient {
+	return &imageGenHTTPClient{
+		accountID: accountID,
+		apiToken:  apiToken,
+		http:      &http.Client{Timeout: 5 * time.Minute},
+	}
+}
+
+// GenerateImage generates an image from text only.
+func (c *imageGenHTTPClient) GenerateImage(ctx context.Context, req ImageRequest) ([]byte, error) {
+	req = applyImageDefaults(req)
+	body := map[string]any{
+		"prompt":    req.Prompt,
+		"num_steps": req.NumSteps,
+	}
+	if req.Width > 0 {
+		body["width"] = req.Width
+	}
+	if req.Height > 0 {
+		body["height"] = req.Height
+	}
+	if req.Guidance > 0 {
+		body["guidance"] = req.Guidance
+	}
+	return c.callImageAPI(ctx, req.Model, body)
+}
+
+// GenerateImageFromReference generates an image from a text prompt + reference image.
+func (c *imageGenHTTPClient) GenerateImageFromReference(ctx context.Context, req ImageRequest, refImage []byte) ([]byte, error) {
+	if len(refImage) == 0 {
+		return c.GenerateImage(ctx, req)
+	}
+	req = applyImageDefaults(req)
+
+	var body map[string]any
+	if req.Model == ImageModelSD15Img2Img {
+		pixels, err := decodeImageToRGBA(refImage)
+		if err != nil {
+			return nil, fmt.Errorf("cfai/image: decode reference: %w", err)
+		}
+		strength := req.Strength
+		if strength <= 0 {
+			strength = 0.75
+		}
+		body = map[string]any{
+			"prompt":    req.Prompt,
+			"image":     pixels,
+			"strength":  strength,
+			"num_steps": req.NumSteps,
+		}
+	} else {
+		b64 := base64.StdEncoding.EncodeToString(refImage)
+		body = map[string]any{
+			"prompt":    req.Prompt,
+			"image_b64": b64,
+			"num_steps": req.NumSteps,
+		}
+		if req.Strength > 0 {
+			body["strength"] = req.Strength
+		}
+	}
+	if req.Width > 0 {
+		body["width"] = req.Width
+	}
+	if req.Height > 0 {
+		body["height"] = req.Height
+	}
+	if req.Guidance > 0 {
+		body["guidance"] = req.Guidance
+	}
+	return c.callImageAPI(ctx, req.Model, body)
+}
+
+// Models returns all supported image model metadata.
+func (c *imageGenHTTPClient) Models() []ImageModelInfo {
+	return AllImageModels()
+}
+
+func (c *imageGenHTTPClient) callImageAPI(ctx context.Context, model ImageModel, body map[string]any) ([]byte, error) {
+	encoded, err := json.Marshal(body)
+	if err != nil {
+		return nil, fmt.Errorf("cfai/image: marshal: %w", err)
+	}
+	url := fmt.Sprintf("https://api.cloudflare.com/client/v4/accounts/%s/ai/run/%s",
+		c.accountID, string(model))
+	req, err := http.NewRequestWithContext(ctx, http.MethodPost, url, bytes.NewReader(encoded))
+	if err != nil {
+		return nil, fmt.Errorf("cfai/image: build request: %w", err)
+	}
+	req.Header.Set("Authorization", "Bearer "+c.apiToken)
+	req.Header.Set("Content-Type", "application/json")
+
+	resp, err := c.http.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("cfai/image: http: %w", err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		errBody, _ := io.ReadAll(resp.Body)
+		msg := string(errBody)
+		if len(msg) > 300 {
+			msg = msg[:300]
+		}
+		return nil, fmt.Errorf("cfai/image: model %s returned %d: %s", model, resp.StatusCode, msg)
+	}
+
+	data, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("cfai/image: read response: %w", err)
+	}
+	return data, nil
+}
+
+func applyImageDefaults(req ImageRequest) ImageRequest {
+	if req.Model == "" {
+		req.Model = DefaultImageModel
+	}
+	if req.NumSteps <= 0 {
+		req.NumSteps = 20
+	}
+	return req
+}
+
+// decodeImageToRGBA decodes PNG/JPEG bytes to a flat []uint8 RGBA pixel array
+// required by the stable-diffusion-v1-5-img2img model.
+func decodeImageToRGBA(data []byte) ([]uint8, error) {
+	img, _, err := image.Decode(bytes.NewReader(data))
+	if err != nil {
+		return nil, fmt.Errorf("decode image: %w", err)
+	}
+	bounds := img.Bounds()
+	w := bounds.Max.X - bounds.Min.X
+	h := bounds.Max.Y - bounds.Min.Y
+	pixels := make([]uint8, w*h*4)
+	idx := 0
+	for y := bounds.Min.Y; y < bounds.Max.Y; y++ {
+		for x := bounds.Min.X; x < bounds.Max.X; x++ {
+			r, g, b, a := img.At(x, y).RGBA()
+			pixels[idx] = uint8(r >> 8)
+			pixels[idx+1] = uint8(g >> 8)
+			pixels[idx+2] = uint8(b >> 8)
+			pixels[idx+3] = uint8(a >> 8)
+			idx += 4
+		}
+	}
+	return pixels, nil
+}

backend/internal/config/config.go

@@ -66,6 +66,18 @@ type PocketTTS struct {
 	URL string
 }
 
+// CFAI holds credentials for Cloudflare Workers AI TTS.
+type CFAI struct {
+	// AccountID is the Cloudflare account ID.
+	// An empty string disables CF AI generation.
+	AccountID string
+	// APIToken is a Workers AI API token with Workers AI Read+Edit permissions.
+	APIToken string
+	// Model is the Workers AI TTS model ID.
+	// Defaults to "@cf/deepgram/aura-2-en" when empty.
+	Model string
+}
+
 // LibreTranslate holds connection settings for a self-hosted LibreTranslate instance.
 type LibreTranslate struct {
 	// URL is the base URL of the LibreTranslate instance, e.g. https://translate.libnovel.cc
@@ -153,6 +165,7 @@ type Config struct {
 	MinIO          MinIO
 	Kokoro         Kokoro
 	PocketTTS      PocketTTS
+	CFAI           CFAI
 	LibreTranslate LibreTranslate
 	HTTP           HTTP
 	Runner         Runner
@@ -203,6 +216,12 @@ func Load() Config {
 			URL: envOr("POCKET_TTS_URL", ""),
 		},
 
+		CFAI: CFAI{
+			AccountID: envOr("CFAI_ACCOUNT_ID", ""),
+			APIToken:  envOr("CFAI_API_TOKEN", ""),
+			Model:     envOr("CFAI_TTS_MODEL", ""),
+		},
+
 		LibreTranslate: LibreTranslate{
 			URL:    envOr("LIBRETRANSLATE_URL", ""),
 			APIKey: envOr("LIBRETRANSLATE_API_KEY", ""),

backend/internal/runner/runner.go

@@ -27,6 +27,7 @@ import (
 	"go.opentelemetry.io/otel/codes"
 
 	"github.com/libnovel/backend/internal/bookstore"
+	"github.com/libnovel/backend/internal/cfai"
 	"github.com/libnovel/backend/internal/domain"
 	"github.com/libnovel/backend/internal/kokoro"
 	"github.com/libnovel/backend/internal/libretranslate"
@@ -112,6 +113,9 @@ type Dependencies struct {
 	// PocketTTS is the pocket-tts client (CPU, kyutai voices: alba, marius, etc.).
 	// If nil, pocket-tts voice tasks will fail with a clear error.
 	PocketTTS pockettts.Client
+	// CFAI is the Cloudflare Workers AI TTS client (cfai:* prefixed voices).
+	// If nil, CF AI voice tasks will fail with a clear error.
+	CFAI cfai.Client
 	// LibreTranslate is the machine translation client.
 	// If nil, translation tasks will fail with a clear error.
 	LibreTranslate libretranslate.Client
@@ -555,6 +559,18 @@ func (r *Runner) runAudioTask(ctx context.Context, task domain.AudioTask) {
 			return
 		}
 		log.Info("runner: audio generated via pocket-tts", "voice", task.Voice)
+	} else if cfai.IsCFAIVoice(task.Voice) {
+		if r.deps.CFAI == nil {
+			fail("cloudflare AI client not configured (CFAI_ACCOUNT_ID/CFAI_API_TOKEN empty)")
+			return
+		}
+		var genErr error
+		audioData, genErr = r.deps.CFAI.GenerateAudio(ctx, text, task.Voice)
+		if genErr != nil {
+			fail(fmt.Sprintf("cfai generate: %v", genErr))
+			return
+		}
+		log.Info("runner: audio generated via cloudflare AI", "voice", task.Voice)
 	} else {
 		if r.deps.Kokoro == nil {
 			fail("kokoro client not configured (KOKORO_URL is empty)")

homelab/runner/docker-compose.yml

@@ -13,6 +13,11 @@
 #   - RUNNER_SKIP_INITIAL_CATALOGUE_REFRESH=true
 #   - REDIS_ADDR              → rediss://redis.libnovel.cc:6380 (prod Redis via Caddy TLS proxy)
 #   - LibreTranslate service for machine translation (internal network only)
+#
+# extra_hosts pins storage.libnovel.cc and pb.libnovel.cc to the prod server IP
+# (165.22.70.138) so that large PutObject uploads and PocketBase writes bypass
+# Cloudflare's 100-second proxy timeout entirely. TLS still terminates at Caddy
+# on prod; the TLS certificate is valid for the domain names so SNI works fine.
 
 services:
   libretranslate:
@@ -35,6 +40,12 @@ services:
     stop_grace_period: 135s
     depends_on:
       - libretranslate
+    # Pin prod subdomains to the prod server IP to bypass Cloudflare's 100s
+    # proxy timeout. Large MP3 PutObject uploads and PocketBase writes go
+    # directly to Caddy on prod; TLS and SNI still work normally.
+    extra_hosts:
+      - "storage.libnovel.cc:165.22.70.138"
+      - "pb.libnovel.cc:165.22.70.138"
     environment:
       # ── PocketBase ──────────────────────────────────────────────────────────
       POCKETBASE_URL: "https://pb.libnovel.cc"
@@ -63,6 +74,10 @@ services:
       # ── Pocket TTS ──────────────────────────────────────────────────────────
       POCKET_TTS_URL: "${POCKET_TTS_URL}"
 
+      # ── Cloudflare Workers AI TTS ────────────────────────────────────────────
+      CFAI_ACCOUNT_ID: "${CFAI_ACCOUNT_ID}"
+      CFAI_API_TOKEN: "${CFAI_API_TOKEN}"
+
       # ── LibreTranslate (internal Docker network) ────────────────────────────
       LIBRETRANSLATE_URL: "http://libretranslate:5000"
       LIBRETRANSLATE_API_KEY: "${LIBRETRANSLATE_API_KEY}"

ui/messages/en.json

@@ -362,6 +362,7 @@
   "admin_nav_audio": "Audio",
   "admin_nav_translation": "Translation",
   "admin_nav_changelog": "Changelog",
+  "admin_nav_image_gen": "Image Gen",
   "admin_nav_feedback": "Feedback",
   "admin_nav_errors": "Errors",
   "admin_nav_analytics": "Analytics",

ui/messages/fr.json

@@ -362,6 +362,7 @@
   "admin_nav_audio": "Audio",
   "admin_nav_translation": "Traduction",
   "admin_nav_changelog": "Modifications",
+  "admin_nav_image_gen": "Image Gen",
   "admin_nav_feedback": "Retours",
   "admin_nav_errors": "Erreurs",
   "admin_nav_analytics": "Analytique",

ui/messages/id.json

@@ -362,6 +362,7 @@
   "admin_nav_audio": "Audio",
   "admin_nav_translation": "Terjemahan",
   "admin_nav_changelog": "Perubahan",
+  "admin_nav_image_gen": "Image Gen",
   "admin_nav_feedback": "Masukan",
   "admin_nav_errors": "Kesalahan",
   "admin_nav_analytics": "Analitik",

ui/messages/pt.json

@@ -362,6 +362,7 @@
   "admin_nav_audio": "Áudio",
   "admin_nav_translation": "Tradução",
   "admin_nav_changelog": "Alterações",
+  "admin_nav_image_gen": "Image Gen",
   "admin_nav_feedback": "Feedback",
   "admin_nav_errors": "Erros",
   "admin_nav_analytics": "Análise",

ui/messages/ru.json

@@ -362,6 +362,7 @@
   "admin_nav_audio": "Аудио",
   "admin_nav_translation": "Перевод",
   "admin_nav_changelog": "Изменения",
+  "admin_nav_image_gen": "Image Gen",
   "admin_nav_feedback": "Отзывы",
   "admin_nav_errors": "Ошибки",
   "admin_nav_analytics": "Аналитика",

ui/src/lib/components/AudioPlayer.svelte

@@ -86,6 +86,7 @@
 	// ── Derived: voices grouped by engine ──────────────────────────────────
 	const kokoroVoices = $derived(voices.filter((v) => v.engine === 'kokoro'));
 	const pocketVoices = $derived(voices.filter((v) => v.engine === 'pocket-tts'));
+	const cfaiVoices = $derived(voices.filter((v) => v.engine === 'cfai'));
 
 	// ── Voice selector state ────────────────────────────────────────────────
 	let showVoicePanel = $state(false);
@@ -98,6 +99,7 @@
 	 * Human-readable label for a voice.
 	 * Kokoro: "af_bella" → "Bella (US F)"
 	 * Pocket-TTS: "alba" → "Alba (EN F)"
+	 * CF AI: "cfai:luna" → "Luna (EN F)"
 	 * Falls back gracefully if called with a bare string (e.g. from the store default).
 	 */
 	function voiceLabel(v: Voice | string): string {
@@ -110,6 +112,14 @@
 			return kokoroLabelFromId(v);
 		}
 
+		if (v.engine === 'cfai') {
+			// "cfai:luna" → "Luna (EN F)"
+			const speaker = v.id.startsWith('cfai:') ? v.id.slice(5) : v.id;
+			const name = speaker.replace(/\b\w/g, (c) => c.toUpperCase());
+			const genderLabel = v.gender.toUpperCase();
+			return `${name} (EN ${genderLabel})`;
+		}
+
 		if (v.engine === 'pocket-tts') {
 			const langLabel = v.lang.toUpperCase().replace('-', '');
 			const genderLabel = v.gender.toUpperCase();
@@ -844,6 +854,16 @@
 						{@render voiceRow(v)}
 					{/each}
 				{/if}
+
+				<!-- Cloudflare AI section -->
+				{#if cfaiVoices.length > 0}
+					<div class="px-3 py-1.5 bg-(--color-surface-2)/70 border-b border-(--color-border)/50 {kokoroVoices.length > 0 || pocketVoices.length > 0 ? 'border-t border-(--color-border)' : ''}">
+						<span class="text-[10px] font-semibold text-(--color-muted) uppercase tracking-widest">Cloudflare AI</span>
+					</div>
+					{#each cfaiVoices as v (v.id)}
+						{@render voiceRow(v)}
+					{/each}
+				{/if}
 			</div>
 			<div class="px-3 py-2 border-t border-(--color-border) bg-(--color-surface-2)/50">
 				<p class="text-xs text-(--color-muted)">

ui/src/lib/types.ts

@@ -12,7 +12,7 @@
 export interface Voice {
 	/** Voice identifier passed to TTS clients (e.g. "af_bella", "alba"). */
 	id: string;
-	/** TTS engine: "kokoro" | "pocket-tts". */
+	/** TTS engine: "kokoro" | "pocket-tts" | "cfai". */
 	engine: string;
 	/** Primary language tag (e.g. "en-us", "en-gb", "en", "es", "fr"). */
 	lang: string;

ui/src/routes/admin/+layout.svelte

@@ -6,7 +6,8 @@
 		{ href: '/admin/scrape', label: () => m.admin_nav_scrape() },
 		{ href: '/admin/audio', label: () => m.admin_nav_audio() },
 		{ href: '/admin/translation', label: () => m.admin_nav_translation() },
-		{ href: '/admin/changelog', label: () => m.admin_nav_changelog() }
+		{ href: '/admin/changelog', label: () => m.admin_nav_changelog() },
+		{ href: '/admin/image-gen', label: () => m.admin_nav_image_gen() }
 	];
 
 	const externalLinks = [

ui/src/routes/admin/audio/+page.svelte

@@ -57,6 +57,12 @@
 		return `${m}m ${s % 60}s`;
 	}
 
+	function engineLabel(voice: string): string {
+		if (voice.startsWith('cfai:')) return 'CF AI';
+		if (!voice.includes('_')) return 'Pocket TTS';
+		return 'Kokoro';
+	}
+
 	// ── Audio jobs stats + filter ────────────────────────────────────────────────
 	let jobsQ = $state('');
 	let filteredJobs = $derived(
@@ -160,6 +166,7 @@
 							<th class="px-4 py-3 text-left">Book</th>
 							<th class="px-4 py-3 text-right">Ch.</th>
 							<th class="px-4 py-3 text-left">Voice</th>
+							<th class="px-4 py-3 text-left">Engine</th>
 							<th class="px-4 py-3 text-left">Status</th>
 							<th class="px-4 py-3 text-left">Started</th>
 							<th class="px-4 py-3 text-left">Duration</th>
@@ -173,6 +180,7 @@
 								</td>
 								<td class="px-4 py-3 text-right text-(--color-muted)">{job.chapter}</td>
 								<td class="px-4 py-3 text-(--color-muted) font-mono text-xs">{job.voice}</td>
+								<td class="px-4 py-3 text-(--color-muted) text-xs">{engineLabel(job.voice)}</td>
 								<td class="px-4 py-3">
 									<span class="font-medium {jobStatusColor(job.status)}">{job.status}</span>
 								</td>
@@ -181,7 +189,7 @@
 							</tr>
 							{#if job.error_message}
 								<tr class="bg-(--color-danger)/10">
-									<td colspan="6" class="px-4 py-2 text-xs text-(--color-danger) font-mono">{job.error_message}</td>
+									<td colspan="7" class="px-4 py-2 text-xs text-(--color-danger) font-mono">{job.error_message}</td>
 								</tr>
 							{/if}
 						{/each}
@@ -202,6 +210,7 @@
 						<div class="grid grid-cols-2 gap-1 text-xs">
 							<span class="text-(--color-muted)">Chapter</span><span class="text-(--color-muted) text-right">{job.chapter}</span>
 							<span class="text-(--color-muted)">Voice</span><span class="text-(--color-muted) font-mono text-right truncate">{job.voice}</span>
+							<span class="text-(--color-muted)">Engine</span><span class="text-(--color-muted) text-right">{engineLabel(job.voice)}</span>
 							<span class="text-(--color-muted)">Started</span><span class="text-(--color-muted) text-right">{fmtDate(job.started)}</span>
 							<span class="text-(--color-muted)">Duration</span><span class="text-(--color-muted) text-right">{duration(job.started, job.finished)}</span>
 						</div>
@@ -236,6 +245,7 @@
 							<th class="px-4 py-3 text-left">Book</th>
 							<th class="px-4 py-3 text-left">Chapter</th>
 							<th class="px-4 py-3 text-left">Voice</th>
+							<th class="px-4 py-3 text-left">Engine</th>
 							<th class="px-4 py-3 text-left">Filename</th>
 							<th class="px-4 py-3 text-left">Updated</th>
 						</tr>
@@ -249,6 +259,7 @@
 								</td>
 								<td class="px-4 py-3 text-(--color-muted)">{parts.chapter}</td>
 								<td class="px-4 py-3 text-(--color-muted) font-mono text-xs">{parts.voice}</td>
+								<td class="px-4 py-3 text-(--color-muted) text-xs">{engineLabel(parts.voice)}</td>
 								<td class="px-4 py-3 text-(--color-muted) font-mono text-xs truncate max-w-[14rem]" title={entry.filename}>
 									{entry.filename}
 								</td>
@@ -267,11 +278,12 @@
 						<a href="/books/{parts.slug}" class="text-(--color-text) font-medium hover:text-(--color-brand) transition-colors block truncate">
 							{parts.slug}
 						</a>
-						<div class="grid grid-cols-2 gap-1 text-xs">
-							<span class="text-(--color-muted)">Chapter</span><span class="text-(--color-muted) text-right">{parts.chapter}</span>
-							<span class="text-(--color-muted)">Voice</span><span class="text-(--color-muted) font-mono text-right truncate">{parts.voice}</span>
-							<span class="text-(--color-muted)">Updated</span><span class="text-(--color-muted) text-right">{fmtDate(entry.updated)}</span>
-						</div>
+					<div class="grid grid-cols-2 gap-1 text-xs">
+						<span class="text-(--color-muted)">Chapter</span><span class="text-(--color-muted) text-right">{parts.chapter}</span>
+						<span class="text-(--color-muted)">Voice</span><span class="text-(--color-muted) font-mono text-right truncate">{parts.voice}</span>
+						<span class="text-(--color-muted)">Engine</span><span class="text-(--color-muted) text-right">{engineLabel(parts.voice)}</span>
+						<span class="text-(--color-muted)">Updated</span><span class="text-(--color-muted) text-right">{fmtDate(entry.updated)}</span>
+					</div>
 						{#if entry.filename}
 							<p class="text-xs text-(--color-muted) font-mono truncate" title={entry.filename}>{entry.filename}</p>
 						{/if}

ui/src/routes/admin/image-gen/+page.server.ts

@@ -0,0 +1,28 @@
+import type { PageServerLoad } from './$types';
+import { backendFetch } from '$lib/server/scraper';
+import { log } from '$lib/server/logger';
+
+export interface ImageModelInfo {
+	id: string;
+	label: string;
+	provider: string;
+	supports_ref: boolean;
+	recommended_for: string[]; // "cover" | "chapter"
+	description: string;
+}
+
+export const load: PageServerLoad = async () => {
+	// parent layout already guards admin role
+	try {
+		const res = await backendFetch('/api/admin/image-gen/models');
+		if (!res.ok) {
+			log.warn('admin/image-gen', 'failed to load models', { status: res.status });
+			return { models: [] as ImageModelInfo[] };
+		}
+		const data = await res.json();
+		return { models: (data.models ?? []) as ImageModelInfo[] };
+	} catch (e) {
+		log.warn('admin/image-gen', 'backend unreachable', { err: String(e) });
+		return { models: [] as ImageModelInfo[] };
+	}
+};

ui/src/routes/admin/image-gen/+page.svelte

@@ -0,0 +1,676 @@
+<script lang="ts">
+	import type { PageData } from './$types';
+	import type { ImageModelInfo } from './+page.server';
+
+	let { data }: { data: PageData } = $props();
+
+	// ── Form state ───────────────────────────────────────────────────────────────
+	type ImageType = 'cover' | 'chapter';
+	let imageType = $state<ImageType>('cover');
+	let slug = $state('');
+	let chapter = $state<number>(1);
+	let selectedModel = $state('');
+	let prompt = $state('');
+	let referenceFile = $state<File | null>(null);
+	let referencePreviewUrl = $state('');
+
+	// Advanced
+	let showAdvanced = $state(false);
+	let numSteps = $state(20);
+	let guidance = $state(7.5);
+	let strength = $state(0.75);
+	let width = $state(1024);
+	let height = $state(1024);
+
+	// ── Generation state ─────────────────────────────────────────────────────────
+	let generating = $state(false);
+	let genError = $state('');
+	let elapsedMs = $state(0);
+	let elapsedInterval: ReturnType<typeof setInterval> | null = null;
+
+	// ── Result state ─────────────────────────────────────────────────────────────
+	interface GenResult {
+		imageSrc: string;
+		model: string;
+		bytes: number;
+		contentType: string;
+		saved: boolean;
+		coverUrl: string;
+		elapsedMs: number;
+		slug: string;
+		imageType: ImageType;
+		chapter: number;
+	}
+
+	let result = $state<GenResult | null>(null);
+	let history = $state<GenResult[]>([]);
+
+	let saving = $state(false);
+	let saveError = $state('');
+	let saveSuccess = $state(false);
+
+	// ── Model helpers ────────────────────────────────────────────────────────────
+	const models = data.models as ImageModelInfo[];
+
+	let filteredModels = $derived(
+		referenceFile
+			? models // show all; warn on ones without ref support
+			: models
+	);
+
+	let coverModels = $derived(filteredModels.filter((m) => m.recommended_for.includes('cover')));
+	let chapterModels = $derived(filteredModels.filter((m) => m.recommended_for.includes('chapter')));
+	let otherModels = $derived(
+		filteredModels.filter(
+			(m) => !m.recommended_for.includes('cover') && !m.recommended_for.includes('chapter')
+		)
+	);
+
+	// ── Auto-select default model when type changes ──────────────────────────────
+	$effect(() => {
+		const preferred = imageType === 'cover' ? coverModels : chapterModels;
+		if (!selectedModel && preferred.length > 0) {
+			selectedModel = preferred[0].id;
+		}
+	});
+
+	// Reset model selection when type changes if current selection no longer fits
+	$effect(() => {
+		void imageType; // track
+		const preferred = imageType === 'cover' ? coverModels : chapterModels;
+		if (preferred.length > 0) {
+			// only auto-switch if current model isn't in preferred list for this type
+			const current = models.find((m) => m.id === selectedModel);
+			if (!current || !current.recommended_for.includes(imageType)) {
+				selectedModel = preferred[0].id;
+			}
+		}
+	});
+
+	// ── Prompt templates ────────────────────────────────────────────────────────
+	let promptTemplate = $derived(
+		imageType === 'cover'
+			? `Book cover for "${slug || 'untitled novel'}", a fantasy adventure novel. Epic scene with dramatic lighting, professional book cover art, cinematic composition, highly detailed, 4K.`
+			: `Illustration for chapter ${chapter} of "${slug || 'untitled novel'}". Dramatic moment, vivid colors, anime-inspired style, detailed background, cinematic lighting.`
+	);
+
+	function applyTemplate() {
+		prompt = promptTemplate;
+	}
+
+	// ── Reference image handling ─────────────────────────────────────────────────
+	let dragOver = $state(false);
+
+	function handleReferenceFile(file: File | null) {
+		referenceFile = file;
+		if (referencePreviewUrl) URL.revokeObjectURL(referencePreviewUrl);
+		referencePreviewUrl = file ? URL.createObjectURL(file) : '';
+	}
+
+	function onFileInput(e: Event) {
+		const input = e.target as HTMLInputElement;
+		handleReferenceFile(input.files?.[0] ?? null);
+	}
+
+	function onDrop(e: DragEvent) {
+		e.preventDefault();
+		dragOver = false;
+		const file = e.dataTransfer?.files[0];
+		if (file && file.type.startsWith('image/')) handleReferenceFile(file);
+	}
+
+	function clearReference() {
+		handleReferenceFile(null);
+		const input = document.getElementById('ref-file-input') as HTMLInputElement | null;
+		if (input) input.value = '';
+	}
+
+	// ── Selected model info ──────────────────────────────────────────────────────
+	let selectedModelInfo = $derived(models.find((m) => m.id === selectedModel) ?? null);
+	let refWarning = $derived(
+		referenceFile && selectedModelInfo && !selectedModelInfo.supports_ref
+			? `${selectedModelInfo.label} does not support reference images. The reference will be ignored.`
+			: ''
+	);
+
+	// ── Generate ────────────────────────────────────────────────────────────────
+	let canGenerate = $derived(prompt.trim().length > 0 && slug.trim().length > 0 && !generating);
+
+	async function generate() {
+		if (!canGenerate) return;
+		generating = true;
+		genError = '';
+		result = null;
+		elapsedMs = 0;
+		saveSuccess = false;
+		saveError = '';
+
+		const startTs = Date.now();
+		elapsedInterval = setInterval(() => {
+			elapsedMs = Date.now() - startTs;
+		}, 200);
+
+		try {
+			const payload = {
+				prompt: prompt.trim(),
+				model: selectedModel,
+				type: imageType,
+				slug: slug.trim(),
+				chapter: imageType === 'chapter' ? chapter : 0,
+				num_steps: numSteps,
+				guidance,
+				strength,
+				width,
+				height
+			};
+
+			let res: Response;
+			if (referenceFile && selectedModelInfo?.supports_ref) {
+				const fd = new FormData();
+				fd.append('json', JSON.stringify(payload));
+				fd.append('reference', referenceFile);
+				res = await fetch('/api/admin/image-gen', { method: 'POST', body: fd });
+			} else {
+				res = await fetch('/api/admin/image-gen', {
+					method: 'POST',
+					headers: { 'Content-Type': 'application/json' },
+					body: JSON.stringify(payload)
+				});
+			}
+
+			const body = await res.json().catch(() => ({}));
+			if (!res.ok) {
+				genError = body.error ?? body.message ?? `Error ${res.status}`;
+				return;
+			}
+
+			const totalMs = Date.now() - startTs;
+			const newResult: GenResult = {
+				imageSrc: `data:${body.content_type};base64,${body.image_b64}`,
+				model: body.model,
+				bytes: body.bytes,
+				contentType: body.content_type,
+				saved: body.saved ?? false,
+				coverUrl: body.cover_url ?? '',
+				elapsedMs: totalMs,
+				slug: slug.trim(),
+				imageType,
+				chapter
+			};
+
+			result = newResult;
+			history = [newResult, ...history].slice(0, 5);
+		} catch {
+			genError = 'Network error.';
+		} finally {
+			generating = false;
+			if (elapsedInterval) {
+				clearInterval(elapsedInterval);
+				elapsedInterval = null;
+			}
+		}
+	}
+
+	// ── Save as cover ────────────────────────────────────────────────────────────
+	async function saveAsCover() {
+		if (!result || saving) return;
+		saving = true;
+		saveError = '';
+		saveSuccess = false;
+
+		try {
+			const payload = {
+				prompt: prompt.trim(),
+				model: result.model,
+				type: 'cover',
+				slug: result.slug,
+				num_steps: numSteps,
+				guidance,
+				strength,
+				width,
+				height,
+				save_to_cover: true
+			};
+
+			// Re-generate with save_to_cover=true (backend saves atomically)
+			// Alternatively, we could add a separate save endpoint.
+			// For now we pass the same prompt + model to re-generate and save.
+			// TODO: A lighter approach would be a dedicated save endpoint that accepts
+			// the base64 payload. For now re-gen is acceptable given admin-only usage.
+			const res = await fetch('/api/admin/image-gen', {
+				method: 'POST',
+				headers: { 'Content-Type': 'application/json' },
+				body: JSON.stringify(payload)
+			});
+			const body = await res.json().catch(() => ({}));
+			if (!res.ok) {
+				saveError = body.error ?? body.message ?? `Error ${res.status}`;
+				return;
+			}
+			if (body.saved) {
+				saveSuccess = true;
+				result = { ...result, saved: true, coverUrl: body.cover_url ?? result.coverUrl };
+			} else {
+				saveError = 'Backend did not save the cover.';
+			}
+		} catch {
+			saveError = 'Network error.';
+		} finally {
+			saving = false;
+		}
+	}
+
+	// ── Download ─────────────────────────────────────────────────────────────────
+	function download() {
+		if (!result) return;
+		const a = document.createElement('a');
+		a.href = result.imageSrc;
+		const ext = result.contentType === 'image/jpeg' ? 'jpg' : 'png';
+		a.download =
+			result.imageType === 'cover'
+				? `${result.slug}-cover.${ext}`
+				: `${result.slug}-ch${result.chapter}.${ext}`;
+		a.click();
+	}
+
+	// ── Formatting helpers ───────────────────────────────────────────────────────
+	function fmtElapsed(ms: number) {
+		if (ms < 1000) return `${ms}ms`;
+		return `${(ms / 1000).toFixed(1)}s`;
+	}
+
+	function fmtBytes(b: number) {
+		if (b < 1024) return `${b} B`;
+		if (b < 1024 * 1024) return `${(b / 1024).toFixed(1)} KB`;
+		return `${(b / 1024 / 1024).toFixed(2)} MB`;
+	}
+</script>
+
+<svelte:head>
+	<title>Image Gen — Admin</title>
+</svelte:head>
+
+<div class="space-y-6 max-w-6xl">
+	<!-- Header -->
+	<div>
+		<h1 class="text-2xl font-bold text-(--color-text)">Image Generation</h1>
+		<p class="text-(--color-muted) text-sm mt-1">
+			Generate book covers and chapter images using Cloudflare Workers AI.
+		</p>
+	</div>
+
+	<!-- Layout: form + result side by side on large screens -->
+	<div class="grid grid-cols-1 lg:grid-cols-2 gap-6 items-start">
+		<!-- ── Left: Form panel ──────────────────────────────────────────────────── -->
+		<div class="space-y-4">
+			<!-- Type toggle -->
+			<div class="flex gap-1 bg-(--color-surface-2) rounded-lg p-1 w-fit border border-(--color-border)">
+				{#each (['cover', 'chapter'] as const) as t}
+					<button
+						onclick={() => (imageType = t)}
+						class="px-4 py-1.5 rounded-md text-sm font-medium transition-colors
+							{imageType === t
+								? 'bg-(--color-surface-3) text-(--color-text)'
+								: 'text-(--color-muted) hover:text-(--color-text)'}"
+					>
+						{t === 'cover' ? 'Cover' : 'Chapter Image'}
+					</button>
+				{/each}
+			</div>
+
+			<!-- Slug + chapter -->
+			<div class="flex gap-3">
+				<div class="flex-1 min-w-0 space-y-1">
+					<label class="text-xs font-medium text-(--color-muted) uppercase tracking-wide" for="slug-input">
+						Book slug
+					</label>
+					<input
+						id="slug-input"
+						type="text"
+						bind:value={slug}
+						placeholder="e.g. shadow-slave"
+						class="w-full bg-(--color-surface-2) border border-(--color-border) rounded-lg px-3 py-2 text-(--color-text) text-sm placeholder-zinc-500 focus:outline-none focus:ring-2 focus:ring-(--color-brand)"
+					/>
+				</div>
+
+				{#if imageType === 'chapter'}
+					<div class="w-24 space-y-1 shrink-0">
+						<label class="text-xs font-medium text-(--color-muted) uppercase tracking-wide" for="chapter-input">
+							Chapter
+						</label>
+						<input
+							id="chapter-input"
+							type="number"
+							bind:value={chapter}
+							min="1"
+							class="w-full bg-(--color-surface-2) border border-(--color-border) rounded-lg px-3 py-2 text-(--color-text) text-sm focus:outline-none focus:ring-2 focus:ring-(--color-brand)"
+						/>
+					</div>
+				{/if}
+			</div>
+
+			<!-- Model selector -->
+			<div class="space-y-1">
+				<label class="text-xs font-medium text-(--color-muted) uppercase tracking-wide" for="model-select">
+					Model
+				</label>
+				<select
+					id="model-select"
+					bind:value={selectedModel}
+					class="w-full bg-(--color-surface-2) border border-(--color-border) rounded-lg px-3 py-2 text-(--color-text) text-sm focus:outline-none focus:ring-2 focus:ring-(--color-brand)"
+				>
+					{#if coverModels.length > 0}
+						<optgroup label="Recommended for covers">
+							{#each coverModels as m}
+								<option value={m.id}>
+									{m.label} — {m.provider}{m.supports_ref ? ' ★ref' : ''}
+								</option>
+							{/each}
+						</optgroup>
+					{/if}
+					{#if chapterModels.length > 0}
+						<optgroup label="Recommended for chapters">
+							{#each chapterModels as m}
+								<option value={m.id}>
+									{m.label} — {m.provider}{m.supports_ref ? ' ★ref' : ''}
+								</option>
+							{/each}
+						</optgroup>
+					{/if}
+					{#if otherModels.length > 0}
+						<optgroup label="All models">
+							{#each otherModels as m}
+								<option value={m.id}>
+									{m.label} — {m.provider}{m.supports_ref ? ' ★ref' : ''}
+								</option>
+							{/each}
+						</optgroup>
+					{/if}
+				</select>
+				{#if selectedModelInfo}
+					<p class="text-xs text-(--color-muted)">{selectedModelInfo.description}</p>
+				{/if}
+				{#if refWarning}
+					<p class="text-xs text-amber-400">{refWarning}</p>
+				{/if}
+			</div>
+
+			<!-- Prompt -->
+			<div class="space-y-1">
+				<div class="flex items-center justify-between">
+					<label class="text-xs font-medium text-(--color-muted) uppercase tracking-wide" for="prompt-input">
+						Prompt
+					</label>
+					<button
+						onclick={applyTemplate}
+						class="text-xs text-(--color-brand) hover:text-(--color-brand-dim) transition-colors"
+					>
+						Use template
+					</button>
+				</div>
+				<textarea
+					id="prompt-input"
+					bind:value={prompt}
+					rows="5"
+					placeholder="Describe the image to generate…"
+					class="w-full bg-(--color-surface-2) border border-(--color-border) rounded-lg px-3 py-2 text-(--color-text) text-sm placeholder-zinc-500 focus:outline-none focus:ring-2 focus:ring-(--color-brand) resize-y"
+				></textarea>
+			</div>
+
+			<!-- Reference image drop zone -->
+			<div class="space-y-1">
+				<p class="text-xs font-medium text-(--color-muted) uppercase tracking-wide">
+					Reference image <span class="normal-case font-normal text-(--color-muted)">(optional, img2img)</span>
+				</p>
+				{#if referenceFile && referencePreviewUrl}
+					<div class="flex items-start gap-3 p-3 bg-(--color-surface-2) rounded-lg border border-(--color-border)">
+						<img
+							src={referencePreviewUrl}
+							alt="Reference"
+							class="w-16 h-16 object-cover rounded-md shrink-0 border border-(--color-border)"
+						/>
+						<div class="min-w-0 flex-1 space-y-0.5">
+							<p class="text-sm text-(--color-text) truncate">{referenceFile.name}</p>
+							<p class="text-xs text-(--color-muted)">{fmtBytes(referenceFile.size)}</p>
+						</div>
+						<button
+							onclick={clearReference}
+							class="text-(--color-muted) hover:text-(--color-text) transition-colors shrink-0"
+							aria-label="Remove reference image"
+						>
+							<svg class="w-4 h-4" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+								<path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M6 18L18 6M6 6l12 12" />
+							</svg>
+						</button>
+					</div>
+				{:else}
+					<!-- Drop zone -->
+					<label
+						class="flex flex-col items-center justify-center gap-2 p-4 border-2 border-dashed rounded-lg cursor-pointer transition-colors
+							{dragOver
+								? 'border-(--color-brand) bg-(--color-brand)/5'
+								: 'border-(--color-border) hover:border-(--color-brand)/50 hover:bg-(--color-surface-2)'}"
+						ondragover={(e) => { e.preventDefault(); dragOver = true; }}
+						ondragleave={() => { dragOver = false; }}
+						ondrop={onDrop}
+					>
+						<svg class="w-6 h-6 text-(--color-muted)" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+							<path stroke-linecap="round" stroke-linejoin="round" stroke-width="1.5"
+								d="M4 16l4.586-4.586a2 2 0 012.828 0L16 16m-2-2l1.586-1.586a2 2 0 012.828 0L20 14m-6-6h.01M6 20h12a2 2 0 002-2V6a2 2 0 00-2-2H6a2 2 0 00-2 2v12a2 2 0 002 2z" />
+						</svg>
+						<span class="text-xs text-(--color-muted)">Drop image or <span class="text-(--color-brand)">click to browse</span></span>
+						<input
+							id="ref-file-input"
+							type="file"
+							accept="image/png,image/jpeg,image/webp"
+							onchange={onFileInput}
+							class="sr-only"
+						/>
+					</label>
+				{/if}
+			</div>
+
+			<!-- Advanced collapsible -->
+			<div class="border border-(--color-border) rounded-lg overflow-hidden">
+				<button
+					onclick={() => (showAdvanced = !showAdvanced)}
+					class="w-full flex items-center justify-between px-4 py-2.5 bg-(--color-surface-2) text-sm font-medium text-(--color-muted) hover:text-(--color-text) transition-colors"
+				>
+					Advanced options
+					<svg
+						class="w-4 h-4 transition-transform {showAdvanced ? 'rotate-180' : ''}"
+						fill="none" stroke="currentColor" viewBox="0 0 24 24"
+					>
+						<path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M19 9l-7 7-7-7" />
+					</svg>
+				</button>
+
+				{#if showAdvanced}
+					<div class="px-4 py-4 bg-(--color-surface) space-y-4">
+						<!-- num_steps -->
+						<div class="space-y-1">
+							<div class="flex justify-between">
+								<label class="text-xs text-(--color-muted)">Steps</label>
+								<span class="text-xs text-(--color-text) font-mono">{numSteps}</span>
+							</div>
+							<input type="range" min="1" max="20" step="1" bind:value={numSteps}
+								class="w-full accent-(--color-brand)" />
+						</div>
+
+						<!-- guidance -->
+						<div class="space-y-1">
+							<div class="flex justify-between">
+								<label class="text-xs text-(--color-muted)">Guidance</label>
+								<span class="text-xs text-(--color-text) font-mono">{guidance.toFixed(1)}</span>
+							</div>
+							<input type="range" min="1" max="20" step="0.5" bind:value={guidance}
+								class="w-full accent-(--color-brand)" />
+						</div>
+
+						<!-- strength (only when reference present) -->
+						{#if referenceFile}
+							<div class="space-y-1">
+								<div class="flex justify-between">
+									<label class="text-xs text-(--color-muted)">Strength</label>
+									<span class="text-xs text-(--color-text) font-mono">{strength.toFixed(2)}</span>
+								</div>
+								<input type="range" min="0" max="1" step="0.05" bind:value={strength}
+									class="w-full accent-(--color-brand)" />
+								<p class="text-xs text-(--color-muted)">0 = copy reference · 1 = ignore reference</p>
+							</div>
+						{/if}
+
+						<!-- width × height -->
+						<div class="grid grid-cols-2 gap-3">
+							<div class="space-y-1">
+								<label class="text-xs text-(--color-muted)" for="width-input">Width</label>
+								<input id="width-input" type="number" min="256" max="2048" step="64" bind:value={width}
+									class="w-full bg-(--color-surface-2) border border-(--color-border) rounded-md px-3 py-1.5 text-(--color-text) text-sm focus:outline-none focus:ring-1 focus:ring-(--color-brand)" />
+							</div>
+							<div class="space-y-1">
+								<label class="text-xs text-(--color-muted)" for="height-input">Height</label>
+								<input id="height-input" type="number" min="256" max="2048" step="64" bind:value={height}
+									class="w-full bg-(--color-surface-2) border border-(--color-border) rounded-md px-3 py-1.5 text-(--color-text) text-sm focus:outline-none focus:ring-1 focus:ring-(--color-brand)" />
+							</div>
+						</div>
+					</div>
+				{/if}
+			</div>
+
+			<!-- Generate button -->
+			<button
+				onclick={generate}
+				disabled={!canGenerate}
+				class="w-full py-2.5 rounded-lg bg-(--color-brand) text-(--color-surface) font-semibold text-sm
+					hover:bg-(--color-brand-dim) transition-colors disabled:opacity-50 disabled:cursor-not-allowed
+					flex items-center justify-center gap-2"
+			>
+				{#if generating}
+					<!-- Spinner -->
+					<svg class="w-4 h-4 animate-spin" fill="none" viewBox="0 0 24 24">
+						<circle class="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" stroke-width="4" />
+						<path class="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8v8H4z" />
+					</svg>
+					Generating… {fmtElapsed(elapsedMs)}
+				{:else}
+					Generate
+				{/if}
+			</button>
+
+			{#if genError}
+				<p class="text-sm text-(--color-danger) bg-(--color-danger)/10 rounded-lg px-3 py-2">{genError}</p>
+			{/if}
+		</div>
+
+		<!-- ── Right: Result panel ────────────────────────────────────────────────── -->
+		<div class="space-y-4">
+			{#if result}
+				<div class="bg-(--color-surface) border border-(--color-border) rounded-xl overflow-hidden">
+					<!-- Image -->
+					<img
+						src={result.imageSrc}
+						alt="Generated image"
+						class="w-full object-contain max-h-[36rem] bg-zinc-950"
+					/>
+
+					<!-- Meta bar -->
+					<div class="px-4 py-3 border-t border-(--color-border) space-y-3">
+						<div class="grid grid-cols-3 gap-2 text-xs">
+							<div>
+								<p class="text-(--color-muted)">Model</p>
+								<p class="text-(--color-text) font-mono truncate" title={result.model}>
+									{models.find((m) => m.id === result!.model)?.label ?? result.model}
+								</p>
+							</div>
+							<div>
+								<p class="text-(--color-muted)">Size</p>
+								<p class="text-(--color-text)">{fmtBytes(result.bytes)}</p>
+							</div>
+							<div>
+								<p class="text-(--color-muted)">Time</p>
+								<p class="text-(--color-text)">{fmtElapsed(result.elapsedMs)}</p>
+							</div>
+						</div>
+
+						{#if result.saved}
+							<p class="text-xs text-green-400">
+								Cover saved &rarr;
+								<a href={result.coverUrl} target="_blank" rel="noopener noreferrer"
+									class="underline hover:text-green-300">{result.coverUrl}</a>
+							</p>
+						{/if}
+
+						{#if saveSuccess && !result.saved}
+							<p class="text-xs text-green-400">Cover saved successfully.</p>
+						{/if}
+						{#if saveError}
+							<p class="text-xs text-(--color-danger)">{saveError}</p>
+						{/if}
+
+						<!-- Actions -->
+						<div class="flex gap-2 flex-wrap">
+							<button
+								onclick={download}
+								class="flex-1 px-3 py-1.5 rounded-md bg-(--color-surface-3) text-(--color-text) text-xs font-medium hover:bg-zinc-600 transition-colors"
+							>
+								Download
+							</button>
+
+							{#if result.imageType === 'cover'}
+								<button
+									onclick={saveAsCover}
+									disabled={saving || result.saved}
+									class="flex-1 px-3 py-1.5 rounded-md bg-(--color-brand) text-(--color-surface) text-xs font-semibold
+										hover:bg-(--color-brand-dim) transition-colors disabled:opacity-50"
+								>
+									{saving ? 'Saving…' : result.saved ? 'Saved ✓' : 'Save as cover'}
+								</button>
+							{/if}
+						</div>
+					</div>
+				</div>
+			{:else if generating}
+				<!-- Placeholder while generating -->
+				<div class="flex items-center justify-center bg-(--color-surface) border border-(--color-border) rounded-xl h-80">
+					<div class="text-center space-y-3">
+						<svg class="w-8 h-8 animate-spin mx-auto text-(--color-brand)" fill="none" viewBox="0 0 24 24">
+							<circle class="opacity-25" cx="12" cy="12" r="10" stroke="currentColor" stroke-width="4" />
+							<path class="opacity-75" fill="currentColor" d="M4 12a8 8 0 018-8v8H4z" />
+						</svg>
+						<p class="text-sm text-(--color-muted)">Generating… {fmtElapsed(elapsedMs)}</p>
+					</div>
+				</div>
+			{:else}
+				<!-- Empty state -->
+				<div class="flex items-center justify-center bg-(--color-surface) border border-(--color-border) border-dashed rounded-xl h-80">
+					<p class="text-sm text-(--color-muted)">Generated image will appear here</p>
+				</div>
+			{/if}
+
+			<!-- History thumbnails -->
+			{#if history.length > 0}
+				<div class="space-y-2">
+					<p class="text-xs font-semibold text-(--color-muted) uppercase tracking-widest">Session history</p>
+					<div class="flex gap-2 flex-wrap">
+						{#each history as h, i}
+							<button
+								onclick={() => result = h}
+								class="relative rounded-md overflow-hidden border transition-colors shrink-0
+									{result === h ? 'border-(--color-brand)' : 'border-(--color-border) hover:border-(--color-brand)/50'}"
+							>
+								<img
+									src={h.imageSrc}
+									alt="History {i + 1}"
+									class="w-16 h-16 object-cover"
+								/>
+								{#if h.saved}
+									<span class="absolute bottom-0.5 right-0.5 w-2.5 h-2.5 rounded-full bg-green-500 border border-(--color-surface)"></span>
+								{/if}
+							</button>
+						{/each}
+					</div>
+				</div>
+			{/if}
+		</div>
+	</div>
+</div>

ui/src/routes/api/admin/image-gen/+server.ts

@@ -0,0 +1,46 @@
+/**
+ * POST /api/admin/image-gen
+ *
+ * Admin-only proxy to the Go backend's image generation endpoint.
+ * Transparently forwards the request body (JSON or multipart/form-data)
+ * and returns the JSON response containing the base64-encoded image.
+ */
+
+import { json, error } from '@sveltejs/kit';
+import type { RequestHandler } from './$types';
+import { log } from '$lib/server/logger';
+import { backendFetch } from '$lib/server/scraper';
+
+export const POST: RequestHandler = async ({ request, locals }) => {
+	if (!locals.user || locals.user.role !== 'admin') {
+		throw error(403, 'Forbidden');
+	}
+
+	const ct = request.headers.get('content-type') ?? '';
+
+	let res: Response;
+	try {
+		if (ct.includes('multipart/form-data')) {
+			// Forward raw body bytes; let the backend parse multipart
+			const body = await request.arrayBuffer();
+			res = await backendFetch('/api/admin/image-gen', {
+				method: 'POST',
+				headers: { 'content-type': ct },
+				body
+			});
+		} else {
+			const body = await request.text();
+			res = await backendFetch('/api/admin/image-gen', {
+				method: 'POST',
+				headers: { 'content-type': 'application/json' },
+				body
+			});
+		}
+	} catch (e) {
+		log.error('admin/image-gen', 'backend proxy error', { err: String(e) });
+		throw error(502, 'Could not reach backend');
+	}
+
+	const data = await res.json().catch(() => ({}));
+	return json(data, { status: res.status });
+};

ui/src/routes/profile/+page.svelte

@@ -93,6 +93,30 @@
 
 	const kokoroVoices = $derived(voices.filter((v) => v.engine === 'kokoro'));
 	const pocketVoices = $derived(voices.filter((v) => v.engine === 'pocket-tts'));
+	const cfaiVoices = $derived(voices.filter((v) => v.engine === 'cfai'));
+
+	function voiceLabel(v: Voice): string {
+		if (v.engine === 'cfai') {
+			const speaker = v.id.startsWith('cfai:') ? v.id.slice(5) : v.id;
+			return speaker.replace(/\b\w/g, (c) => c.toUpperCase()) + (v.gender ? ` (EN ${v.gender.toUpperCase()})` : '');
+		}
+		if (v.engine === 'pocket-tts') {
+			const name = v.id.replace(/_/g, ' ').replace(/\b\w/g, (c) => c.toUpperCase());
+			return name + (v.gender ? ` (EN ${v.gender.toUpperCase()})` : '');
+		}
+		// Kokoro: "af_bella" → "Bella (US F)"
+		const langMap: Record<string, string> = {
+			af: 'US', am: 'US', bf: 'UK', bm: 'UK',
+			ef: 'ES', em: 'ES', ff: 'FR',
+			hf: 'IN', hm: 'IN', 'if': 'IT', im: 'IT',
+			jf: 'JP', jm: 'JP', pf: 'PT', pm: 'PT', zf: 'ZH', zm: 'ZH',
+		};
+		const prefix = v.id.slice(0, 2);
+		const name = v.id.slice(3).replace(/^v0/, '').replace(/^([a-z])/, (c) => c.toUpperCase());
+		const lang = langMap[prefix] ?? prefix.toUpperCase();
+		const gender = v.gender ? v.gender.toUpperCase() : '?';
+		return `${name} (${lang} ${gender})`;
+	}
 
 	$effect(() => {
 		fetch('/api/voices')
@@ -492,12 +516,17 @@
 					class="w-full bg-(--color-surface-3) border border-(--color-border) rounded-lg px-3 py-2 text-(--color-text) text-sm focus:outline-none focus:ring-2 focus:ring-(--color-brand)">
 					{#if kokoroVoices.length > 0}
 						<optgroup label="Kokoro (GPU)">
-							{#each kokoroVoices as v}<option value={v.id}>{v.id}</option>{/each}
+							{#each kokoroVoices as v}<option value={v.id}>{voiceLabel(v)}</option>{/each}
 						</optgroup>
 					{/if}
 					{#if pocketVoices.length > 0}
 						<optgroup label="Pocket TTS (CPU)">
-							{#each pocketVoices as v}<option value={v.id}>{v.id}</option>{/each}
+							{#each pocketVoices as v}<option value={v.id}>{voiceLabel(v)}</option>{/each}
+						</optgroup>
+					{/if}
+					{#if cfaiVoices.length > 0}
+						<optgroup label="Cloudflare AI">
+							{#each cfaiVoices as v}<option value={v.id}>{voiceLabel(v)}</option>{/each}
 						</optgroup>
 					{/if}
 				</select>