Instructions to use zulense/Z1_V0.1.0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use zulense/Z1_V0.1.0 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import export_to_video # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("rain1011/pyramid-flow-miniflux", dtype=torch.bfloat16, device_map="cuda") pipe.load_lora_weights("zulense/Z1_V0.1.0") prompt = "Line, you will remember that in the last match we made it, what was it, the graph was going up like this, it was going down, it was going through, it was going down, something like this," output = pipe(prompt=prompt).frames[0] export_to_video(output, "output.mp4") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Draw Things
z1

- Prompt
- Line, you will remember that in the last match we made it, what was it, the graph was going up like this, it was going down, it was going through, it was going down, something like this,
- Negative Prompt
- some math not found.
Model description
π§ Zulense Z1 (Base Model)
Zulense Z1 is a specialized text-to-video diffusion model designed specifically for educational visualization. Unlike general-purpose video generation models, Z1 has been fine-tuned on the Indian Mathematics Curriculum (Class 8 & 9) to generate accurate, whiteboard-style explanations for mathematical concepts.
π Model Details
Developed by: Zulense AI Research
Model Type: Latent Diffusion Model (LDM) with Educational VAE
Language(s): English
License: apache-2.0
Model Description: A lightweight video generation model capable of turning text prompts (e.g., "Line, you will remember that in the last match we made it, what was it, the graph was going up like this, it was going down, it was going through, it was going down, something like this,") into 2-4 minute visual animations.
Current Version: v0.1-alpha (Research Preview)
π Quantitative Evaluation (v0.1.0 Baseline)
Zulense Z1 (v0.1.0-alpha) has been evaluated on standard educational whiteboard generation sequences using optical flow motion analysis, frame-difference temporal tracking, and OCR recognition fidelity.
Empirical Results
| Metric Category | Evaluation Metric | Value | Target Direction | Status / Interpretation |
|---|---|---|---|---|
| Pedagogical Text | Mean OCR Confidence | 8.83% | $\uparrow$ (Higher) | Early-stage pseudo-glyph generation |
| Valid Text Legibility Rate | ~2.8% | $\uparrow$ (Higher) | Requires resolution upscaling | |
| Temporal Stability | Frame Flickering Score | 0.9554 | $\downarrow$ (Lower) | Stable global scene composition |
| Motion Quality | Motion Smoothness Index | 2.1761 | $\uparrow$ (Higher) | Consistent optical flow continuity |
π Qualitative Analysis & Roadmap to v0.2
- Current Baseline (v0.1-alpha): The model demonstrates high scene layout coherence (teacher presence, board borders, and coordinate plane placement) and smooth motion dynamics (2.18 smoothness index).
- Primary Optimization Target for v0.2: At $256 \times 256$ native resolution, whiteboard text is represented as pseudo-text/glyphs (OCR confidence of 0.088). Version 0.2 will integrate a dedicated high-resolution text VAE decoder and higher-resolution latent training to boost formula legibility.
π Intended Use
This model is designed for:
EdTech Developers: Building tools to visualize STEM concepts.
Content Creators: Generating assets for educational videos.
Research: Exploring the application of Gen-AI in pedagogy.
Out-of-Scope Use:
Generating photorealistic humans or non-educational content.
Generating content unrelated to STEM subjects.
πΎ Training Data
The Z1 model was trained on a proprietary dataset curated by Zulense, consisting of:
10+ Hours of "Whiteboard Style" math lectures.
Curriculum Text Pairs: Aligned with NCERT Class 8 math textbooks.
π How to Use (Python)
from diffusers import DiffusionPipeline
import torch
# Load the pipeline (Conceptual Example)
pipe = DiffusionPipeline.from_pretrained("zulense/z1", torch_dtype=torch.float16)
pipe.to("cuda")
prompt = "A 3D wireframe cylinder splitting open to show surface area formula"
video_frames = pipe(prompt, num_frames=24).frames
# Save video
video_frames[0].save("cylinder_explanation.gif", save_all=True, append_images=video_frames[1:], duration=100, loop=0)
β οΈ Limitations & Bias
Hallucinations: As an Alpha model, Z1 may occasionally generate incorrect text overlays or formulas. Always verify output.
Resolution: Currently limited to 256x256 resolution.
Scope: Strictly optimized for Geometry and Algebra; may perform poorly on Biology or History prompts.
π§ Contact
For partnership inquiries or access to the Zulense Pro (v2) checkpoints, please contact: info@zulense.com
Built with β€οΈ in new Delhi, India.
Trigger words
You should use text-to-video to trigger the image generation.
Download model
Download them in the Files & versions tab.
- Downloads last month
- 7
Model tree for zulense/Z1_V0.1.0
Base model
rain1011/pyramid-flow-miniflux