File size: 3,165 Bytes
15d68eb | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 | # Examples — Indic Heritage Studio v2
This directory contains pre-baked demo outputs for the project profile PDF
and the demo video. Run `python scripts/generate_demo_outputs.py` to
generate the full gallery on your dev box before recording the AMD demo.
## Structure
```
examples/
├── README.md ← this file
├── inputs/ ← source images used for style transfer / video / inpainting
│ ├── portrait.jpg
│ ├── landscape.jpg
│ └── portrait_mask.png
├── 1_text_to_image/ ← 5 styles × 3 prompts × 3 seeds = 45 images
│ ├── madhubani/
│ ├── warli/
│ ├── pattachitra/
│ ├── mughal/
│ └── tanjore/
├── 2_style_transfer/ ← 5 styles × 2 input images = 10 styled images
│ ├── madhubani/
│ ├── warli/
│ └── …
├── 3_image_to_video/ ← 5 styles × 25-frame MP4 = 5 videos
│ ├── madhubani.mp4
│ └── …
├── 5_inpainting/ ← 1 inpainting demo
└── 6_controlnet/ ← 1 ControlNet demo (canny + mughal)
```
## Demo video script (3-5 min)
1. **0:00 — Title card + intro (15s)**
- "Indic Heritage Studio — multimodal heritage art generation on AMD Radeon + ROCm"
2. **0:15 — GPU verification (15s)**
- `rocm-smi` visible in terminal
- `python scripts/verify_rocm.py` shows ROCm ready
3. **0:30 — Text → Image demo (60s)**
- Run: `python -m core.text_to_image --prompt "..." --style madhubani --size 1024`
- Show output PNG
- Mention: "SDXL 1.0 + per-style LoRA, 1024×1024 native"
4. **1:30 — Style transfer demo (45s)**
- Run: `python -m core.style_transfer --image examples/inputs/portrait.jpg --style mughal`
- Show before/after
- Mention: "IP-Adapter XL on SDXL, 30 steps"
5. **2:15 — Image → Video demo (45s)**
- Run: `python -m core.image_to_video --image outputs/demo_styled_mughal.png --out outputs/demo.mp4`
- Play the 25-frame MP4
- Mention: "Stable Video Diffusion XT 1.1, 25 frames at 8 fps"
6. **3:00 — Inpainting + ControlNet demo (45s)**
- Quick show of inpainting output
- Quick show of ControlNet (canny → mughal)
7. **3:45 — Multi-GPU benchmark (30s)**
- `python scripts/benchmark.py --configs 1 2 4 8 --samples 4`
- Show scaling chart
8. **4:15 — UI walkthrough (30s)**
- `python app.py` → walk through all 6 tabs in browser
9. **4:45 — Closing (15s)**
- "Track 1, TeamIndicForge, Indic Heritage Studio — submitted."
## Pre-baking strategy
Run `python scripts/generate_demo_outputs.py` on your 8×80GB NVIDIA dev box
in Week 2. This takes ~30 minutes and produces every demo asset. The AMD
demo recording (Aug 4) then only needs to *re-run* a few commands live for
the camera — the actual outputs already exist as fallback if AMD
generation fails.
## Curating the gallery for the PDF
After generating, curate:
- Pick 1 best image per style for the PDF cover
- Pick 1 best video for the PDF embedded figure
- Pick the most visually striking ControlNet result
- Pick a 2×2 grid showing 4 styles for the architecture section
|