OneTranslate 2024 Year-End Progress Update

The closed beta launch in March went fairly well, and since then, there have been significant research and development efforts into milestone 4. These developments are shaping up to deliver some amazing advancements in 2025.

Beta Plans and Rationale
Originally, I planned to release the app into open beta later this year. However, I’ve decided to extend the Patreon closed beta into early next year to focus on deploying new changes over the coming months.

Here’s why:
While I haven’t received many specific feature requests beyond quality-of-life updates and expanded language support, the current feature set feels sufficient for an open beta. That said, with the rapid AI advancements and the big tech push toward AGI, it’s clear that automating manga translation through AI remains the priority. Reducing human effort through new AI systems will always take precedence.

So here's what happened and what's going to happen:

Miscellaneous Updates

• Hardware upgrades: 4x VRAM now, with 8x VRAM coming soon, enabling much larger models.

• Patreon automation: New patrons now receive invite keys immediately.

• Backend stability: While random network hangs haven’t been fully resolved, their frequency has dropped significantly, from once every few weeks to once every few months. Work on this is ongoing.

Features Targeted for End of 2025

• Per-line translation with editable Japanese text.

• Multi-language support, starting with popular Latin-based languages.

Milestone 4 - Next Level AI Research and Development Progress

• Developing a custom fine-tuned LLM specifically for AI translation assistance, moving away from reliance on OpenAI. This approach will enable human-grade translation for NSFW content.

• Building a custom LLM backend to allow seamless switching between popular LLM providers (e.g., OpenAI, Anthropic, x.AI).

• Experimenting with vision models achieving >90% accuracy in scene description and speaker identification. Integrating this will bring long-form context and speaker consideration into translations, potentially surpassing human-level accuracy.

These R&D efforts will be part of the experimental features included in the open beta slated for early next year.

I will have another update around New Years time. Cheers and Happy Holidays!