VibeSliding ✨

Transform ideas into stunning presentations in seconds. Write your content, pick a style, and let AI handle the rest — no design skills required.

Why VibeSliding? 🚀

• Save Hours — What used to take hours now takes minutes
• Professional Results — Every slide looks polished and cohesive
• Zero Learning Curve — If you can write markdown, you can create presentations
• Iterate Fast — Generate multiple variants instantly and pick your favorite

Features

• 📝 Plan — A skill writes the deck plan from your idea and references, with no API calls
• 🎨 Two-Tone Style Plates — Dark curtain plates (cover/transition/ending) + light teaching plates (content), filled from a design brief
• 🖼️ Slide References — Attach slide-specific photos or image globs in the plan; they pass through to the script
• 🧠 Role-Aware Render — Routes each slide to the matching style plates by role (cover / transition / content / ending)
• 🔄 Multiple Variants — Generate several design options per slide
• 🎯 Selective Regeneration — Redo specific slides without starting over
• ⚡ Parallel Generation — Concurrent API calls with configurable concurrency
• 📽️ Auto PPTX — One 16:9 deck: slide images plus [Speech:] notes in presenter view
• 🎬 Narrated Video — TTS + ffmpeg turns slides and [Speech:] tags into MP4
• 🔌 Multi-Provider — OpenRouter (text + images) and Volcengine/Doubao Seedream (images)

Quick Start

pip install -r requirements.txt
cp .env.example .env   # add your API keys

Pipeline

Four stages — plan → write → design → render. Present is optional.

# 1. Plan (no API): idea.md + references → facts.md, plan_N.md, design_brief.md
#    Use the plan-deck skill, then check the plan:
python3 -m src.plan.cli --plan work/plan_16.md

# 2. Write: plan_N.md + facts.md (required) → one text call per slide → script_N.md
python3 -m src.write.cli --work work --plan work/plan_16.md

# 3. Design: image calls → work/style/*.png
python3 -m src.design.cli --work work --script work/script_16.md

# 4. Render: one image call per slide → image dir + PPTX
python3 -m src.render.cli --work work --script work/script_16.md

# Optional: slide PNGs + [Speech:] tags → narrated MP4 (requires ffmpeg)
python3 -m src.present.cli --output work/image_YYYYMMDD_HHMMSS

Then curate (delete variants you don't love) and polish (--page to regenerate individual slides).

💡 Plan writes facts.md (one - **F1 Name.** fact — citation bullet per fact), plan_N.md (slides cite fact IDs in Evidence:), and design_brief.md; src.plan.cli also checks that Section: k/total lines match the roadmap and that every cited ID exists in the facts.md beside the plan. Write requires facts.md, sends each slide its cited facts plus the most related others, expands each slide in parallel, saves prompts under work/write_prompts/, and then checks the script for structure, page numbers (Slide number: N on content slides only, never a total), text/visual balance (content slides within a 0.85-1.35 text-to-visual ratio; cover, roadmap, and ending slides more visual than text), and speech length (1-2 minutes on content slides, about 30 seconds on the rest). Design writes five plates under work/style/ (style_base_noncontent.png, style_base_content.png, style_cover.png, style_transition.png, style_content.png), each with a *_prompt.txt sidecar. Render routes plates by slide role on the full deck, then applies --page, and writes images to work/image_YYYYMMDD_HHMMSS/ plus a PPTX in work/. Present writes presentation_video_YYYYMMDD_HHMMSS.mp4 (install ffmpeg first).

Usage Examples

# Design: one plate per stage, or several candidates to pick from
python3 -m src.design.cli --work work --script work/script_16.md
python3 -m src.design.cli --work work --script work/script_16.md --candidates 4 --pick 1,1,2,1,3

# Write: redo a few slides
python3 -m src.write.cli --work work --plan work/plan_16.md --page "3,7-9"

# Render: multiple variants per slide
python3 -m src.render.cli --work work --script work/script_16.md --copy 4

# Render: selective regeneration
python3 -m src.render.cli --work work --script work/script_36.md --page "1,3,5-7"

# Render: rebuild PPTX after curating (no API calls)
python3 -m src.render.cli --work work --pptx-only --output work/image_20260520_220006
python3 -m src.render.cli --work work --pptx-only --output work/image_20260520_220006 --variant 1

# Present: TTS + video from curated slides (MiniMax TTS)
python3 -m src.present.cli --output work/image_20260520_220006 --variant 1

# Present: clone your voice from a short reference recording (clean WAV/MP3)
python3 -m src.present.cli --output work/image_20260520_220006 \
  --reference-audio work/voice_sample.wav

# Present: remux only after TTS MP3s already exist (no API calls)
python3 -m src.present.cli --output work/image_20260520_220006 --mux-only

CLI Reference

src.plan.cli

Option	Description
--plan	Deck plan to validate (default: work/plan_16.md); also checks Evidence: IDs against facts.md in the same folder

src.write.cli

Option	Description
--work	Work directory (default: work/); must contain facts.md, optional idea.md
--plan	Deck plan (plan_N.md), required
--page	Slides to rewrite, e.g. 3,7-9 (merges into an existing script)
--txt-model	Text model override
--api-key	OpenRouter API key override
--proxy	HTTP/HTTPS proxy for text calls

src.design.cli

Option	Description
--work	Work directory (default: work/)
--script	Script file (default: work/script_16.md)
--brief	Design brief (default: WORK/design_brief.md)
--style	Style directory (default: WORK/style/)
--candidates	Candidates per style image (default: 1, max: 12). 1 skips picking
--pick	Pre-select indices: base_noncontent,base_content,cover,transition,content
--api-key	API key override
--proxy	HTTP/HTTPS proxy (OpenRouter only)
--provider	openrouter or volcengine

src.render.cli

Option	Description
--work	Work directory (default: work/)
--script	Script file (default: work/script_16.md)
--style	Directory holding the style images (default: WORK/style when it exists, else style/)
--copy	Variants per slide (default: 1)
--output	Output dir (default: WORK/image_YYYYMMDD_HHMMSS/); PPTX goes to --work
--api-key	API key override
--page	Pages to generate, e.g. 1,3,5-7
--proxy	HTTP/HTTPS proxy (OpenRouter only)
--provider	openrouter or volcengine
--balance-only	Print OpenRouter credits and exit
--no-balance	Skip credits line after run
--pptx-only	Rebuild PPTX from existing PNGs in --output dir (no API calls); speaker notes from --script [Speech:] tags
--variant	With --pptx-only: variant filter for the PPTX, e.g. 1 or 1,2

Credits: After each OpenRouter run, remaining credits are printed unless --no-balance. Requires an OpenRouter Management API key — set OPENROUTER_MANAGEMENT_API_KEY or [openrouter] management_api_key in .env.

src.present.cli

Option	Description
--work	Work directory (default: work/)
--script	Script with [Speech:] tags (default: work/script_16.md). A named missing file is an error. When omitted and the default is missing, uses the newest script_*.md snapshot in --output
--output	Render image directory with slide_p##_v##.png (required)
--page	Slides to include, e.g. 1,3,5-7
--variant	Variant filter, e.g. 1 or 1,2
--api-key	MiniMax API key override
--tts-model	MiniMax TTS model (default: speech-2.8-hd)
--voice	TTS voice (default: Chinese (Mandarin)_Lyrical_Voice)
--reference-audio	WAV/MP3/FLAC of your voice for MiniMax voice cloning
--voice-id	Saved MiniMax voice ID (alternative to --reference-audio)
--mux-only	Skip TTS; mux existing slide_p##_v##.mp3 files (no API calls)
--silent-seconds	Duration for slides without speech (default: 3.0)

Prerequisite: ffmpeg and ffprobe must be on your PATH.

Voice clone: --reference-audio uploads reference audio and clones voice dynamically via MiniMax.

Input/Output

Work Directory

work/
├── idea.md                                    # seed: title, audience, language
├── facts.md                                   # verified facts with citations
├── source.md                                  # optional extra source material
├── design_brief.md                            # motif, palette language, on-plate labels
├── plan_16.md                                 # deck plan (content-slide count in the name)
├── script_16.md                               # full slide script
├── write_prompts/                             # one prompt file per slide
├── style/                                     # five style plates for this deck
│   ├── style_base_noncontent.png
│   ├── style_base_content.png
│   ├── style_cover.png
│   ├── style_transition.png
│   ├── style_content.png
│   ├── style_*_prompt.txt
│   └── style_candidates/
├── image_YYYYMMDD_HHMMSS/                     # render output (slide PNGs)
├── slides_YYYYMMDD.pptx                       # slide images + speaker notes
└── presentation_video_YYYYMMDD_HHMMSS.mp4     # narrated slide video

Outline Format

# PPT Outline: My Presentation

---

## Slide 1: Introduction
- **Key point:** Why this matters
- Core insight: One takeaway the audience should remember
[Style: style_content.png, style_base_content.png]
[Reference: photos/founder.png]
[Visual: Split-screen hero; light content background. Slide number: 1]
[Speech: Conversational presenter narration for this slide]

---

## Appendix: Global Visual Requirements
- **Theme:** Two-tone deck — white/light content slides; dark curtain for cover, transitions, ending. Accents: Primary #2563EB, Accent #10B981
- **Fonts:** Title / body families and sizes

Tag	Purpose
[Visual: ...]	Layout, composition, diagrams, icons, motifs
[Speech: ...]	Presenter narration (used by present + PPTX speaker notes)
[Style: ...]	Style plates for this slide — filenames only, looked up in --style; omit to inherit the plate for the slide's role
[Reference: ...]	Slide-specific image refs such as headshots or charts (place immediately before [Visual:]; leading @ accepted)
## Appendix: ...	Deck-wide text constraints only (theme/hex, fonts/sizes; state two-tone backgrounds)

Include 3–6 transition slides (titles prefixed Roadmap:) with progress bar k/total section markers. Content slides show Slide number: N — the page number only, never the total. Without a [Style: ...] tag, render attaches dark plates to cover/transition/ending and light plates to content slides:

Slide role	Plates attached
cover, ending	style_cover.png, style_base_noncontent.png
transition	style_transition.png, style_base_noncontent.png
content	style_content.png, style_base_content.png

Render Output

work/
├── image_YYYYMMDD_HHMMSS/
│   ├── script_16.md                     # script snapshot used for this run
│   ├── slide_p01_prompt.txt           # exact prompt used for slide 1
│   ├── slide_p01_v01.png              # Slide 1, variant 1
│   ├── slide_p01_v01.mp3              # TTS audio (after present)
│   ├── slide_p01_v02.png              # Slide 1, variant 2
│   └── slide_p02_v01.png              # Slide 2, variant 1
├── slides_YYYYMMDD.pptx                      # all slides + speaker notes in one deck
└── presentation_video_YYYYMMDD_HHMMSS.mp4    # narrated MP4

Configuration

Copy .env.example → .env and fill in your keys. INI-style sections:

Section	Used by	Key settings
(preamble)	All	provider, max_concurrent, proxy
[openrouter]	Write / Render (when provider = openrouter)	api_key, management_api_key, img_model, txt_model, use_proxy
[volcengine]	Design / Render (when provider = volcengine)	api_key, img_model, txt_model, use_proxy
[minimax]	Present (+ text/image when it is the active provider)	api_key, img_model, txt_model, use_proxy, optional tts_model, tts_voice

• Write → txt_model from the active provider section (falls back across [openrouter], [volcengine], [minimax])
• Design / Render → img_model from the active provider (src.design.cli + src.render.cli); OpenRouter image calls go through the unified Image API (POST /v1/images), so image-only models such as openai/gpt-image-2.5-sunburst work alongside chat-style ones like google/gemini-3-pro-image-preview
• Present → MiniMax TTS via [minimax] tts_model / tts_voice or --tts-model / --voice (defaults: speech-2.8-hd, Chinese (Mandarin)_Lyrical_Voice) with optional voice cloning
• Global proxy applies when a section's use_proxy = true; section-level proxy overrides are still supported
• ark_api_key in the preamble aliases [volcengine] api_key

License

See LICENSE file for details.

---

Happy Sliding! 🎨✨ Transform your markdown into beautiful presentations in minutes, not hours.