Revert layout to original output card; keep timing order only
Codex+1: Claude Sonnet 4.6
d61efd6
book_display and coloring_display back inside the right output column.
The output card layout is restored to: audio → PDF buttons → book pages.
The three-step TIMING remains: audio appears first, then PDF download
button, then book pages — just all within the same column as before.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move book pages below audio: audio always visible first
Codex+1: Claude Sonnet 4.6
e49e280
Audio player + download buttons stay in right column at eye level.
Illustrated storybook pages move to a full-width row below both columns,
so users see and use the narration before scrolling to the book pages.
Added a subtle scroll hint arrow between the two sections.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove char name field + page slider; illustrated PDF cover
Codex+1: Claude Sonnet 4.6
786a8ec
- Removed unused 'Character name' input field — only hero name is needed
- Removed 'Number of pages' slider — pages 7-10 generated identical filler
because the model reliably writes 6 pages; fixed at 6 for a complete book
- PDF cover now uses the first FLUX illustration as a full-bleed background
with dark banner overlays for kicker + title + badge, matching the in-app
HTML cover; fallback cream design kept for when no illustration is available
- Both story PDF and coloring PDF use the same illustrated cover system
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fix voice recording page freeze: upload-only for custom voice
Codex+1: Claude Sonnet 4.6
7301102
Gradio's in-browser microphone freezes the page on stop-recording due to
client-side audio encoding. Switch to upload-only so users record on their
phone/voice memo app and upload the file — same cloning result, no freeze.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fix voice freeze, cover design, output order, story continuity
Codex+1: Claude Sonnet 4.6
ef904db
- Voice cloning (my_voice) now runs sequentially AFTER FLUX images to avoid
ZeroGPU concurrency freeze: two concurrent @spaces.GPU threads caused the
browser to hang whenever the user recorded their voice
- Book cover now shows the first page illustration as a framed artwork above
the title, replacing the plain cream-background placeholder
- PDF downloads moved to the output card (below audio, above book pages) so
the output reads: audio → PDFs → storybook
- Removed 'Story spark' input — the 1B model couldn't reliably follow it
and it produced inconsistent results; input wiring updated (8 → 7 inputs)
- Story prompt rules tightened: explicit continuity requirement (each page
flows from the previous), distinctive hero trait anchoring, and required
lesson phrasing on the final page
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Single Storybook tab: remove Bedtime/Kannada, add Custom Voice + page count + 4 new themes
Codex+1: Codex
8d51e7b
- Remove Bedtime Voice tab and all Kannada/translation code (NLLB-200, IndicF5,
MMS-TTS, gTTS) — single Storybook tab now covers all features
- Add Custom Voice (My Voice) option: VoxCPM2 clones user's recorded/uploaded
voice via reference_wav_path; shown only when My Voice is selected
- Add page-count slider 6–10 pages (story, images, and narration all scale)
- Add optional Story Spark input for child-driven themes
- 4 new story themes: kindness to animals, the magic of imagination,
celebrating who you are, a rainy day adventure — each with 3 rich arc variants
- Expand FEW_SHOT exemplar rules for variable page counts
- Raise TTS budget to 180s (preset voices for 10 pages + voice cloning)
- Raise image budget to 200s (up to 10 pages × FLUX)
- Better header with feature strip; remove bedtime CSS (~250 lines)
- Update README with hackathon validator checklist, sponsor model table,
parameter counts (7B total, all under 32B cap), demo/social placeholders
- Remove gtts from requirem…
Fix gated-model startup crash and bedtime UI contrast
Codex
cdeff8b
app.py:
- Remove load_translation + load_kannada_tts from module-scope ZeroGPU
loading — both ai4bharat models are gated and crash startup without
HF_TOKEN; now loaded lazily at inference time only
- generate_kannada_gpu: surface clear user message when HF_TOKEN missing
instead of opaque GatedRepoError
ui/layout.py:
- Swap bedtime tab palette from dark navy (invisible text on Gradio light
base) to twilight lavender (#ece4ff / #dfd6f8) — all text readable
- Cards: light purple gradient with indigo ink text
- Chips, fields, audio inputs: white-over-lavender with visible labels
- Button: gold/moon on lavender (unchanged)
- Story display: indigo-ink text on semi-transparent white panels
- .db-star: changed from white (invisible on paper bg) to purple
- Intro tagline + empty-state text: #4a2a9a / #7b5fc4 (was #9090c0)
Built by Codex
- ui/layout.py: full two-tab layout (Storybook + Bedtime Voice) with
night-sky dark palette, kid-friendly sparkle/star animations, and
moon-gold CSS for the bedtime panel
- app.py: BEDTIME_TEMPLATES expanded to 3 random arc variants per genre
(6 themes x 3 arcs), richer few-shot exemplars that enforce character
consistency, temperature sampling (0.85/0.92) for story variety,
generate_bedtime_story_gpu / generate_tts_cloned_gpu / generate_kannada_gpu,
create_bedtime generator wired to UI
- config.py: TRANSLATION_MODEL, KANNADA_TTS_MODEL, KANNADA_FINETUNE,
BEDTIME_GENRES, BEDTIME_MOODS added
- requirements.txt: librosa, IndicTransToolkit added
- audio_postprocess.py, audio_utils.py: voice reference prep helpers
- indic_text.py: English→Kannada via IndicTrans2
- indic_tts.py: Kannada narration via IndicF5 + fine-tuned bedtime checkpoint
- README.md: ai4bharat/IndicF5 and indictrans2-en-indic-1B added to models
Co-authored-by: Codex <noreply@codex.ai>
Fix mobile layout: stack columns vertically, restore scroll
Codex+1: Claude Sonnet 4.6
166350a
Two-column row was clipping off-screen on phones. Columns now wrap to
full width on ≤820px. Also fixes iOS Safari scroll by switching
background-attachment to scroll and adding overflow-y: auto.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Upgrade Space to gradio 6 so voxcpm==2.0.3 (VoxCPM2) installs; move theme/css/js/head to launch(); fix removed show_copy_button
