LLI-MT

English (source)

Spanish (target)

Thinking...

How it works

Model. Answers are generated by Gemma 4 (E4B) running on our own server, without fine-tuning: the behaviour is driven by the system prompt. Quality may vary between languages and text types, so please review important translations. The first request after a period of inactivity can take a minute while the model loads.

System prompts. Audiovisual, Financial and Post-editing are ready-made translation prompts. Their text is shown above and you can edit it.

Languages. The Source language and Target language selectors only work with the ready-made translation prompts: they fill in the variables {source_lang}, {source_code}, {target_lang} and {target_code} of the prompt. With Image interaction and Write your own they are greyed out and ignored (your choice is remembered).

Editing a prompt. As soon as you change a ready-made prompt, the selector switches to Write your own. From then on the variables are not replaced: if you keep a name such as {target_lang} in your text, the model will receive it literally, so write the languages directly in your prompt. Choosing a ready-made prompt again restores its original text. Choosing Write your own empties the box.

Post-editing. This prompt works differently: the left box takes the original text and the middle box takes the machine translation draft you want to improve (type or paste it). After Ask, the post-edited text appears in a third box on the right, so you can compare the three versions side by side; the draft is never overwritten. You can edit the original or the draft and press Ask again: the result is marked as outdated until you do. Use Hide to close the result box and Show post-edited result to bring it back. The result is cleared when you switch to another system prompt. Tip: you can translate first with a translation prompt and then switch to Post-editing; the translation stays in the middle box as the draft.

Images. You can attach one image (any image format, no PDFs, up to 20 MB) by dragging it onto the input box, pasting it or using Add image. This is only available with Image interaction and Write your own. Text is always required: ask your question or describe your request about the image.

Thinking. With Yes the model reasons before answering, which can improve difficult texts but is slower. No gives faster answers.

Temperature. Controls how much the model varies its wording, from 0.0 (the most consistent and literal; usually best for translation) to 1.0 (more varied and creative). Use the − and + buttons to change it in steps of 0.1.

Privacy. This service does not store your texts or images.