SITEWIDE NOTICE · UPDATED 30 SEPTEMBER 2026

Terms, responsibility
& original-author credits

This notice applies to the entire website and all its interactive tools, including StyleGAN2, AnimeGANv3, Real-ESRGAN and Stable Diffusion / SD-Turbo. It also applies to selected images, generated images, transformed images and enlarged images.

Research and demonstration purpose

This is Aravind Sivakumaran’s personal portfolio and research demonstration. The image lab illustrates methods discussed in my final-year research, alongside separately identified browser experiments. These experimental demonstrations are offered for educational and research exploration. Third-party models and research are credited below; their inclusion does not mean that I authored them or that their creators endorse this website.

Your inputs and use of outputs

By using this website and its tools, you accept responsibility for your inputs, every image you generate or process, and your use or sharing of those images. You must comply with applicable laws, third-party rights and model licences. You are responsible for addressing complaints or claims arising from your unlawful use or infringement of third-party rights, to the extent required by applicable law. Do not use the tools to infringe copyright or privacy, impersonate people deceptively, or create or distribute unlawful content. Review outputs before using them.

Generated or transformed images may contain errors, bias, offensive material or similarities to existing works. Availability of a tool or an output does not establish ownership or grant permission for commercial use. The fact that this site is a research demonstration does not remove applicable legal or licensing obligations.

Experimental tools and responsibility

To the extent permitted by applicable law, the website and demonstrations are provided “as is”, without guarantees of accuracy, quality, fitness for a purpose, availability or non-infringement. Aravind Sivakumaran does not authorize improper use and is not responsible for a user’s unlawful or improper use of these tools or their outputs. Nothing in this notice excludes or limits any liability that cannot lawfully be excluded or limited.

This notice does not automatically transfer all legal claims to users, bind third-party claimants, or exclude responsibilities that the law places on the site operator. Any limitation of liability remains subject to applicable law.

Rights, licences and attribution

Rights in third-party models, weights, software, research and other materials remain with their respective authors and rights holders. Attribution does not replace permission or a licence, and this notice does not grant rights in selected images or clear the rights in generated outputs. The original licence terms prevail. Some components carry non-commercial restrictions; do not assume a commercial-use licence from this website.

• StyleGAN2. Original research and implementation by NVIDIA and the StyleGAN authors. The trained FYP model demonstrated here is my recovered research model: an inference export of a StyleGAN2-family (Diffusion-StyleGAN2 with DiffAugment) network. I no longer have its original checkpoint hash or conversion record, so I cannot say it is identical to the model behind the reported results. It is offered for non-commercial demonstration only. Original project · Original licence.
• Diffusion-GAN. Research by Zhendong Wang and collaborators, referenced in my FYP and the architecture explanations. Research, authors and licence.
• AnimeGANv3. By TachibanaYoshino and the project’s authors. The supplied ONNX model exports are used for the image styling demonstrations. Project, authors and licence terms. Follow the author’s restrictions, including applicable non-commercial conditions.
• Real-ESRGAN. By Xintao Wang and collaborators. This site uses the supplied RealESRGAN_x4plus_anime_6B ONNX export. Original project · Bundled BSD-3-Clause notice.
• SD-Turbo. By Stability AI and the model’s authors, using schmuell’s FP16 ONNX export. Original model and research · ONNX export · Export’s supplied licence. The export is identified as subject to Stability AI non-commercial terms.
• Browser implementation. SD-Turbo pipeline adapted from the Microsoft ONNX Runtime example under its MIT licence. Cyril Diagne’s SD-Turbo + TAESD demo was a browser reference.
• Runtime and tokenizer. ONNX Runtime by Microsoft and contributors (bundled licence); Transformers.js by Hugging Face, Xenova and contributors; CLIP by OpenAI and its authors, with Xenova’s tokenizer files.
• dmakit. My IOMMU and DMA emulator is MIT-licensed, except the files derived from the Linux kernel’s DMA code (SWIOTLB, dma-direct, dma-iommu), which are under the GNU GPL version 2; see its README. Linux is by Linus Torvalds and the kernel’s contributors.
• Motion. Scroll and interaction animation uses Anime.js 4.5.0 by Julian Garnier (MIT licence, bundled).
• Typography. Manrope, JetBrains Mono and Cinzel by their original authors and contributors. Bundled font licences: Manrope, JetBrains Mono, Cinzel.

Training data

My StyleGAN2 model was trained, for a non-commercial university research project, on anime illustrations I collected and filtered myself. The dataset is not published or redistributed, and the model is offered here only for non-commercial demonstration. The illustrations belong to their artists. If you believe your work was included and want it considered for removal from future training, write to the address below.

Employer

This is a personal site. The views and content are my own and are not AMD’s. The IOMMU notes are written from publicly available documents, such as AMD’s published specifications (amd.com, docs.amd.com), the Linux kernel and Arm’s public documentation, and are intended to contain no employer code, confidential information or unreleased product details. Some chapters also cite my own study notes on confidential computing, written from AMD’s public documents and not published; those citations are marked. The notes are not reviewed or endorsed by AMD. AMD, EPYC and related names are trademarks of Advanced Micro Devices, Inc., used here only to name the documents and products discussed. Intel, Arm, PCIe and other names are trademarks of their respective owners.

Fan references

A few pages carry small references to films, books and anime I enjoy. They are fan references only. All names and trademarks belong to their owners, and no affiliation or endorsement is implied. The site uses no logos, stills or character art from those works. The AnimeGANv3 style models are named by their original author after the artists and studio whose look they imitate (for example “Hayao”, “Shinkai”); this site has no affiliation with Studio Ghibli, Hayao Miyazaki or Makoto Shinkai.

IOMMU notes and external compilation

The IOMMU notes contain independent educational drafts using synthetic C++ models. They are not employer publications, hardware specifications or evidence of silicon correctness. No employer endorsement is implied. The inline runner sends the source you choose to run to Compiler Explorer for compilation and execution. Do not submit confidential or proprietary material. Availability and processing are subject to that service’s policies.

Compilation provided by Compiler Explorer, created by Matt Godbolt and contributors. This site uses its external compilation service; it does not host a C++ compiler or execute submitted native programs on the portfolio server.

Server logs

The site is served from a small home server. Its web server may keep ordinary access logs (IP address, time, page requested, browser) to keep the site running; they are kept only as long as needed to run the site, and are not shared or used for tracking. The site sets no cookies and runs no analytics.

Photo handling

The image tools process selected photos, prompts and generated images in your browser. Their current implementations do not upload that content to an inference server. Model and library downloads still contact their hosting services, which may receive ordinary connection information such as your IP address.

Settings and progress kept in your browser

The site sets no cookies and runs no analytics. It does keep a few small items in your browser's local storage, only on your device and never sent to the server: your theme, palette and motion choices; the reading mode, the chapter you last read, which guides you have opened and your reading path; (for this tab only, cleared when it closes) command-palette history; the code-block theme, your place when you return to a note and the home-page cosmos setting; your playground look and achievements; and the lab's last settings. They stay until you clear this site's data in your browser settings (usually under privacy or site data). Downloaded model files may also be kept by your browser's ordinary cache, which the same setting clears. Nothing is uploaded: the image tools read the photos you select in your browser.

Questions or rights concerns

For questions about attribution or material on this site, email aravindnova598@gmail.com with the relevant page, material and rights concern. I will look into it promptly.