File size: 3,235 Bytes
c62a863 4a80ff1 | 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 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 | ---
license: creativeml-openrail-m
base_model: runwayml/stable-diffusion-v1-5
tags:
- stable-diffusion
- lora
- cinematic
- interstellar
- nolan
- christopher-nolan
- text-to-image
---
# Absolute-Cinema-AI-Tiny
A cinematic LoRA fine-tuned on *Interstellar (2014)* stills,
trained on 910 curated frames to capture Hoyte van Hoytema's
iconic cinematography style — heavily inspired by the visual storytelling of Christopher Nolan.
## Disclaimer
This is a **tiny experimental LoRA**, not a production-grade model.
Results can be inconsistent, and in many cases, quality may fall short compared to larger or better-trained models. Think of this more as a **proof-of-concept / early iteration** rather than a polished release.
That said — it can still produce some surprisingly cinematic shots if prompted well
## Usage
**Trigger word:** `ISTELLAR`
## Example Prompt
ISTELLAR, cinematic masterpiece, hoyte van hoytema,
anamorphic lens flare, 70mm film grain, IMAX format,
dramatic natural lighting, shallow depth of field,
teal and orange color grading, [your scene here]
---
## Training
- **Base model:** SD 1.5
- **Images:** 910
- **Epochs:** 8
- **Network:** LoCon (dim 32, alpha 16)
---
## Performance
- **Size: 1.8GB**
- Strong cinematic composition out-of-the-box
- Handles lighting, contrast, and framing exceptionally well
- Works best with detailed prompts (camera terms = boost)
- Minimal overfitting despite stylization
---
## Generation Time (Estimated on T4)
*(No upscaling unless stated otherwise)*
| Resolution | Aspect Ratio | Label | Time (Approx) |
|------------------|-------------|---------------|--------------|
| 512×512 | 1:1 | Square | ~8–12 sec |
| 768×768 | 1:1 | Square HD | ~20–36 sec |
| 1024×1024 | 1:1 | Square HQ | ~32–45 sec |
| 768×432 | 16:9 | Cinematic | ~10–16 sec |
| 1024×576 | 16:9 | Cinematic HD | ~18–23 sec |
| 1280×720 | 16:9 | HD (720p) | ~25–35 sec |
| 1920×1080 | 16:9 | Full HD | ~45–60 sec |
| 2048×858 | ~2.39:1 | Ultrawide | ~65–110 sec |
| 3840×2160* | 16:9 | IMAX / 4K | ~120–240 sec |
> *4K/IMAX typically achieved via 4× upscaling pipeline*
> Times vary based on steps, guidance scale, and VRAM optimization settings
> **NOTE: Upscaler is buit in!**
---
## Notes
- Designed for cinematic realism, not anime/cartoon styles
- Best paired with camera/film keywords (IMAX, 70mm, anamorphic, etc.)
- Works insanely well for space, landscapes, and emotional close-ups
- Exeptions are humans and anything human body related unless specified very much
---
## Future Work
| Version | Base Model | Dataset | Status |
|---|---|---|---|
| **Absolute-Cinema-AI-Medium** | SDXL 1.0 | Interstellar + Oppenheimer | Next |
| **Absolute-Cinema-AI-Large** | SDXL 1.0 | All 4 movies | Planned |
| **Absolute-Cinema-AI-XL** | Flux.1-dev | All 4 + extra Nolan | Someday |
## Credits
- Cinematography inspiration: Hoyte van Hoytema
- Style influence: Christopher Nolan
- Creator: Andy-ML-And-AI |