Top Open-Source AI Tools for Watermark Removal, PPT Creation, and TTS

This week’s GitHub Trending highlighted a surge of projects focused on improving the quality of AI‑generated content, covering watermark removal, PPT creation, speech synthesis, text humanization, and writing guidance.

1. Remove AI Watermarks – One‑Click AI Image Watermark Removal

A command‑line tool that eliminates visible and invisible watermarks from major AI image generators such as Gemini, DALL‑E, Stable Diffusion, Midjourney, Doubao, and Jimo.

Visible watermarks – removes Gemini’s “Nana Banana” flash, Doubao’s “Doubao AI generated” text bar, Jimo’s “★ Jimo AI” tag, Samsung Galaxy AI annotations by reverse‑alpha fusion to restore original pixels.

Invisible watermarks – clears SynthID implicit watermarks through regeneration.

Metadata – strips C2PA Content Credentials and EXIF/XMP “Made with AI” tags.

The project offers both a CLI and a Python library; removal of visible watermarks and metadata is free.

GitHub: https://github.com/wiltodelta/remove-ai-watermarks Stars: 3k+ | Language: Python pip install remove-ai-watermarks

2. GordenSuperPPTSkills – GPT‑Powered Premium PPTs

The project brands itself as the “PPT track terminator” and splits PPT creation into two independent skills:

GordenImagePPTGen – leverages GPT’s image generation to produce luxurious image‑based PPT slides.

GordenImage2PPTX – converts the image‑based PPT back into a fully editable PPTX file.

Workflow: the user tells GPT the presentation topic; GPT first generates an image‑only PPT (background, layout, icons, text separated), then decomposes it into four layers (background, framework, decorative icons, text). The layers are placed at specific coordinates in PowerPoint to assemble an editable PPTX.

The output is a dense, complex‑style PPT rather than simple text layout, and it is currently limited to Codex usage.

GitHub: https://github.com/GordenSun/GordenSuperPPTSkills Stars: 370+ | Language: Python (Agent Skill) | Applicable: Codex

3. Dots.TTS – Xiaohongshu’s Open‑Source End‑to‑End Speech Synthesis

A fully continuous, end‑to‑end TTS system from Xiaohongshu AI Lab with 2 B parameters. Unlike traditional TTS, Dots.TTS does not rely on discrete tokens; it uses a continuous signal flow comprising a semantic encoder, LLM, AR flow‑matching acoustic head, and a 48 kHz AudioVAE.

On multilingual test sets it achieves open‑source state‑of‑the‑art results: Chinese WER 0.94 %, English WER 1.30 %, speaker similarity 81.0 / 77.1.

Three pretrained models are provided: base (foundation), soar (self‑aligned, best for voice cloning), and mf (MeanFlow distillation for faster inference).

GitHub: https://github.com/rednote-hilab/dots.tts Stars: 300+ | Language: Python | Online demo: HuggingFace Spaces

4. Text Humanizer – Free Open‑Source AI Text Humanization

A completely free tool that uses DeepSeek LLM’s multi‑turn rewrite pipeline to turn AI‑generated text into more human‑like prose with added personality.

Processing pipeline:

DeepSeek rewrites the original text while preserving semantics, altering sentence structure, and inserting Chinese as an intermediate language.

Google Translate converts the Chinese version to Turkish, introducing additional syntactic perturbation.

Optionally, DeepL translates the Turkish text to Japanese for a second round of disturbance.

DeepSeek finally translates the result back to the original language, cleans translation artifacts, and restores readability.

The tool supports eight languages, is fully open‑source, and can be self‑hosted.

GitHub: https://github.com/khrisat/text-humanizer Stars: 380+ | Language: Python | Features: completely free, open‑source

5. Renwei Writing – “Are You Still Here?” AI Writing Guidance

This is not a tool but a practical checklist created by a writer to retain the author’s presence after AI editing. The core principle is “human‑flavor = presence”: the text should reflect a concrete person with specific circumstances, stance, and cost.

AI can rephrase and polish, but it should not erase the author’s existence. The repository includes a Chinese‑adapted AI‑writing feature checklist based on Wikipedia’s “Signs of AI Writing” and a set of failure‑case analyses with side‑by‑side comparisons of original, flawed, and final versions.

GitHub: https://github.com/orange2ai/renwei-writing Stars: 410+ | Language: Agent Skill | Installation: Cola / Claude Code agent skill

These five projects represent different dimensions of AI output quality: removing AI watermarks, optimizing AI‑generated content, lowering barriers for AI speech synthesis, evading AI detection, and preserving the author’s expressive style after AI revision.