Fix gen UX: elapsed-time feedback + disable broken Qwen3 voice default

Root cause of 'stuck loading / frontend not updating': generation WORKS but is slow on
ZeroGPU (~9-13s) and serializes — the static 'painting…' message with no progress read as
stuck, and Qwen3-TTS (no DashScope key) 503'd as the saved default.
- web/elapsed.js: a live elapsed-seconds counter; wired into the portrait + voice paint
(heroCreator) and skill-icon (skillForge) statuses → 'painting via klein… 8s' ticks, so
slow ≠ stuck. Verified the portrait blob does render (1024x1024) after ~12s.
- Qwen3-TTS now reports available()=false (needs a server key / ?tts= bridge), so it's never
the auto-selected default and a saved-on-qwen3 hero falls back to VoxCPM (which works).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

web/elapsed.js

@@ -0,0 +1,10 @@
+// Show a live elapsed-seconds counter in a status while a slow call runs, so a 12s ZeroGPU
+// generation reads as "working (8s)" instead of a frozen "painting…" that looks stuck. The
+// counter ticks while the promise is pending and clears when it settles.
+export function withElapsed(setText, prefix, promise) {
+  const t0 = Date.now()
+  const tick = () => { try { setText(`${prefix} ${Math.round((Date.now() - t0) / 1000)}s`) } catch { /* ignore */ } }
+  tick()
+  const iv = setInterval(tick, 500)
+  return Promise.resolve(promise).finally(() => clearInterval(iv))
+}

web/heroCreator.js

@@ -17,6 +17,7 @@ import {
 } from '/web/tts.js'
 import { listPersonas, savePersona, removePersona, onRosterChange, putAudio, getAudio, putPortrait, getPortrait } from '/web/personaStore.js'
 import { generatePortrait, imageBackendLabel, imageNeedsDownload, ensureImage } from '/web/imagen.js'
+import { withElapsed } from '/web/elapsed.js'
 
 const CLASSES = ['Warrior', 'Ranger', 'Monk', 'Assassin', 'Mage', 'Paladin', 'Cleric', 'Knight']
 const MAX_TOKENS = 200 // persona JSON + a voice line + a quote
@@ -357,8 +358,8 @@ export function mountHeroCreator(host, opts = {}) {
         portraitStatus.textContent = 'loading model…'
         await ensureImage((f) => { portraitStatus.textContent = `downloading model… ${Math.round(f * 100)}%` })
       }
-      portraitStatus.textContent = `painting via ${imageBackendLabel()}…`
-      const blob = await generatePortrait(`${appearance}. ${PORTRAIT_STYLE}`, { seed: 42 })
+      // Ticking elapsed counter so a slow (~12s) ZeroGPU paint reads as "working", not stuck.
+      const blob = await withElapsed((t) => { portraitStatus.textContent = t }, `painting via ${imageBackendLabel()}…`, generatePortrait(`${appearance}. ${PORTRAIT_STYLE}`, { seed: 42 }))
       await putPortrait(savedId, blob)
       lastPersona.appearance = appearance; lastPersona.portraitUsed = appearance
       hasPortrait = true; setPortrait(blob); autosave()
@@ -410,16 +411,17 @@ export function mountHeroCreator(host, opts = {}) {
     const reclone = design && hasVoice && !designChanged()
     working = true; playBtn.classList.add('busy'); playBtn.disabled = true
     const verb = reclone ? 'updating' : (design ? 'designing' : 'generating')
-    voiceStatus.textContent = `${verb} voice via ${ttsBackendLabel()}…`
+    const prefix = `${verb} voice via ${ttsBackendLabel()}…`
+    const setVS = (t) => { voiceStatus.textContent = t }
     let wav = null
     try {
       if (design && reclone) {
         const blob = await getAudio(savedId)
-        wav = await cloneVoiceWav(await blob.arrayBuffer(), lastPersona.voiceQuote || '', line, lastPersona.voice || '')
+        wav = await withElapsed(setVS, prefix, cloneVoiceWav(await blob.arrayBuffer(), lastPersona.voiceQuote || '', line, lastPersona.voice || ''))
       } else if (design) {
-        wav = await createVoiceWav(lastPersona.voice, line)
+        wav = await withElapsed(setVS, prefix, createVoiceWav(lastPersona.voice, line))
       } else {
-        wav = await synthVoiceWav(lastPersona.voiceId || '', line)
+        wav = await withElapsed(setVS, prefix, synthVoiceWav(lastPersona.voiceId || '', line))
       }
       await putAudio(savedId, new Blob([wav], { type: 'audio/wav' }))
       lastPersona.voiceQuote = line

web/skillForge.js

@@ -9,6 +9,7 @@ import { validateSkill, SAFE_OPS, SAFE_CONDITIONS, ELEMENTS, SHAPES } from '/web
 import { generatePortrait } from '/web/imagen.js'
 import { getPersona, patchPersona, putSkillIcon } from '/web/personaStore.js'
 import { appendEvent } from '/web/progression.js'
+import { withElapsed } from '/web/elapsed.js'
 
 export const FORGE_SYSTEM = [
   'You are the Skill Forge for a fantasy auto-battler. Author ONE POWERFUL signature skill for a',
@@ -81,10 +82,9 @@ export async function forgeSkillForHero(personaId, request, { onStatus, onToken,
   if (!v.ok) return { ok: false, error: v.errors.join('; ') }
 
   const skill = v.skill
-  onStatus && onStatus('painting icon…')
   let iconBlob = null
   try {
-    iconBlob = await generatePortrait(iconPrompt(p, skill), { seed: skill.id })
+    iconBlob = await withElapsed((t) => onStatus && onStatus(t), 'painting icon…', generatePortrait(iconPrompt(p, skill), { seed: skill.id }))
     if (iconBlob) await putSkillIcon(skill.id, iconBlob)
   } catch (e) { console.warn('[tinyarmy:skill] icon generation failed (skill still learned):', (e && e.message) || e) } // logged via diag; art is optional
 

web/ttsQwen3.js

@@ -79,7 +79,10 @@ export const engine = {
   ...common,
   id: 'qwen3',
   label: 'Qwen3-TTS · Voice Design (cloud)',
-  available: () => true,
+  // Needs a server-side DashScope key (or a `?tts=` local bridge). Without one the backend 503s,
+  // so don't claim availability — that keeps it from being the auto-selected default and lets a
+  // saved-on-qwen3 hero fall back to a working engine (VoxCPM). Pick it explicitly if you have a key.
+  available: () => { try { return new URLSearchParams(location.search).has('tts') } catch { return false } },
   synth: (text, voiceId) => postSynth(ttsBase(), text, voiceId),
   synthWav: (text, voiceId) => postSynthWav(ttsBase(), text, voiceId),
   cloneWav: (text, refAb, refText, instruct) => postClone(ttsBase(), text, refAb, refText, instruct),