NeuralWriter vs Rasa Paraphraser
NeuralWriter
neuralwriter.comRasa Paraphraser
github.comSide by side
| Figure | NeuralWriter | Rasa Paraphraser |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | — | — |
| Open source | — | — |
| Web | Yes | — |
| Linux | — | Yes |
| API | Yes | — |
| Self-hosted | — | Yes |
| Plans published | 1 | 1 |
| Spec score · AI Paraphrasing Tools | 8.2 (#7) | 7.7 (#16) |
| Headquarters | Eixample, Barcelona, Spain | San Francisco, United States, with engineering in Berlin, Germany |
| Languages | The paraphrasing API page lists support for 30 languages. | The model supports 30 languages, but the README says the tool has only been tested in English, German, French, Spanish, and Russian. |
Plans
In their own words
NeuralWriter
NeuralWriter is a set of writing, image, and text-analysis tools available on the web, through a Chrome extension, and by API. Its writing tools cover paraphrasing, prompt generation, translation, summarization, word counting, and headline generation. The paraphraser has Light, Medium, and Hard modes and accepts up to 10,000 characters; its API page lists 30 supported languages. The prompt generator creates prompts for ChatGPT and GPT models with the same character limit and language count. Summarization accepts up to 10,000 characters and allows output from 20% to 90% of the source. The word counter accepts up to 25,000 characters and reports readability measures. An AI detector compares five engines, highlights sentences, and gives a consensus verdict without sign-up. Image tools include generation and upscaling, with listed resolutions through 8K and 4K respectively. The free online tools plan costs nothing.
Rasa Paraphraser
Rasa Paraphraser is a free, MIT-licensed open-source tool for generating sentence alternatives that keep intended meaning while changing wording and structure. Rasa built it mainly for data augmentation in Rasa Open Source, with other paraphrasing uses also possible. The multilingual model supports 30 languages, although the README says testing has covered English, German, French, Spanish and Russian. It uses a Prism-framework neural machine translation model with decoding changes intended to increase diversity. Users can work interactively at the command line or generate batches from files. Bulk input accepts Rasa NLU training data or CSV rows containing a sentence and optional label; output can use either format and includes similarity scores. Controls cover language, lexical diversity, n-gram penalties, beam search and candidate count. Docker is required, with CPU and GPU container instructions. A lightweight mode is described as about 2.5 times faster, with possible quality impact.