[New Notebook] Biological Texture Generation (Iris) using LCM & OpenVINO

[humbeaniket2006-max]

Description:
This Pull Request adds a new notebook demonstrating the generation of high-fidelity biological textures, specifically iris patterns, using Latent Consistency Models (LCM) optimized with OpenVINO.

This notebook addresses the need for efficient, high-quality texture generation in medical imaging and biometric datasets, significantly reducing inference steps compared to traditional Latent Diffusion Models (LDMs).

Related Issue:
Closes #3241

Key Features:

• Model Implementation: Utilizes Latent Consistency Models (LCM) for rapid image generation.
• OpenVINO Optimization: Compiles the pipeline for Intel CPU/GPU inference.
• Specific Use Case: Synthetic biological data (iris textures).

Checklist:

• Dependencies installed via %pip install in the first cell.
• Code formatted with Black (line length 160).
• Telemetry and Scarf Pixel tags added.
• README.md created in the notebook folder.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[brmarkus]

Would you mind adding a README and (much) more documentation and descriptions, maybe including references, cites to papers or blogs, please, with example outputs with its example prompts (like reusing some from https://github.com/humbeaniket2006-max/Ocular_Core_Lite)?

Have a look into e.g. "https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/notebooks/hello-world/hello-world.ipynb" for how to typically install dependencies (e.g. %pip install -q ... ...)

Could you also add an inference device selection (CPU, GPU, VPU if supported), please?

Do you really want to completely rely on the thirdparty call for model compilation and the inference, instead of demonstrating OpenVINO APIs within this notebook? Completely delegating everything to a 3rdparty tool/library and then "just call one method to do everything" makes this whole notebook... not so... exciting... and demonstrative?

[humbeaniket2006-max]

Thank you for the detailed feedback! I appreciate the guidance on alignment with the repository standards.

I will start working on these changes immediately:

1.Adding a comprehensive README and more explanatory markdown/references in the notebook.

2.Adding the standard OpenVINO device selection widget.

3.Refactoring the pipeline to better demonstrate OpenVINO's specific capabilities (compilation and inference) rather than just abstracting it away.

I'll push the updates soon!

[brmarkus]

@jgespino do you want to comment as well?

[humbeaniket2006-max]

Thanks for the feedback! I have refactored the notebook to address your points:

1. Removed the 3rd-party wrapper: The notebook now uses optimum.intel.openvino explicitly within the cells, so users can see the full model loading and compilation pipeline.
2. Added Device Selection: Included a widget to select between CPU, GPU, and AUTO.
3. Documentation: Added a README.md and detailed comments explaining the LCM pipeline.

Ready for another look when you have a chance!

[jgespino]

@aleksandr-mokrov Could you please help assign to the team for review?

[aleksandr-mokrov]

@jgespino assigned, thank you for the PR! The team is off next week, so please expect a slight delay. We will be able to review the PR the week after next.

[openvino-dev-samples]

Hi @humbeaniket2006-max thanks for your contribution. 2 questions:

1. Where is the inference part of this notebook?
2. Since we already have a notebook to demonstrate LCM, can you explain more on what is the difference between them ? or any additional stuff you want to highlight ?

[humbeaniket2006-max]

Hi! Thanks for the review.

1. Regarding the Inference Part:
   I have just pushed a major refactor (in the latest commit) that removes the external CLI wrapper. The notebook now explicitly demonstrates the full inference pipeline using Python:

• It loads the model using OVLatentConsistencyModelPipeline.
• It compiles it for the selected device (CPU/GPU).
• It runs the generation loop pipe(prompt, num_inference_steps=4).

2. Difference from Existing LCM Notebooks:
   While there are existing notebooks for general LCM generation (like latent-consistency-models-image-generation), this notebook provides a unique value add:

• Domain Specificity: It is tailored for synthetic biometric data generation (specifically Iris textures). This demonstrates a real-world use case of GenAI for medical/security research, rather than just general "text-to-image."
• Optimum-Intel Native: It explicitly uses the optimum.intel library's OVLatentConsistencyModelPipeline class, serving as a clean, minimal reference for developers who want to use the Hugging Face + OpenVINO integration directly, without the complexity of manual model conversion or ControlNet layers found in other tutorials.

I believe this "UseCase-First" approach helps users understand how to apply OpenVINO to specific industry problems.

[openvino-dev-samples]

the inference part is missing

Reply via ReviewNB