Release Notes

Security improvements and introduction of the update module.

• Product Studio — Added separate Standard and Ultra engines. Ultra is experimental and disabled by default.
• Custom scenes — Removed preset restrictions on people and product categories in scene descriptions.
• Scene descriptions — Presets can be edited. Drafts are retained when switching scenes or engines.
• Generation and saving — Simplified the Custom workflow and direct file saving.
• Product appearance — An option in Custom mode adds instructions to preserve reference details. Exact geometry, materials, logos and text are not guaranteed.
• Ultra hardware check — Accepts 10 GB of dedicated VRAM; 16 GB is recommended. CPU offloading also uses RAM. Full generation with 10 GB has not been verified.
• Interface — Improved Studio text readability and translated validation messages.

Watch the full 2.8.5 presentation →

• Language detection — Individual records are checked against the catalog’s main language.
• On-model scenes — Added female and male presentation variants in Product Studio.
• Progress display — Progress and time estimates for large spreadsheets now reflect the amount of text.
• Column names — Simplified labels and numbered duplicate headings.
• Language recognition and diagnostics — Improved Dutch and Flemish detection. Cell errors are reported separately.

• Multilingual catalogs — Automatically detects language columns and protects existing translations from overwriting.
• Missing translations — Enterprise fills empty language fields while retaining existing content.
• Image analysis and OCR — Updated Basic and Advanced Vision and text recognition for scanned PDFs.
• Optional components — Moved image generation and selected AI libraries into separate installation packages.
• Hardware settings — Automatically chooses GPU, combined GPU/CPU or CPU processing based on available memory.
• PDF handling — Separated processing for text-based PDFs and scanned documents.

• Translation instructions — Revised instructions to improve consistency in the target language.
• Marketing Style — Made advanced copywriting available with the more capable translation options.

• Smart Cache — Reuses saved translations to reduce repeated processing.
• Vision AI — Added product descriptions based on photos.
• CPU usage — Improved task distribution across processor threads.
• Interface — Redesigned the layout to make translation tasks easier to manage.

• XML — Improved analysis of document structure.
• XML — Skips technical fields containing images, links, product IDs and Base64 data.
• PDF to Word — Added conversion before translation to help retain document layout.
• Long texts — Improved splitting into smaller sections with C++ tokenizers.

• XML — Added support for XML translation.
• PDF — Added PDF support with text recognition (beta).

• GPU acceleration — Replaced the CUDA backend with Vulkan.
• Installation — Added Vulkan Runtime to the installer.

• Marketing Style — Added an option for adapting translated text for marketing use.
• Task scheduling — Improved processing on multi-core systems.

• Multithreading — Improved use of processor cores.
• Keywords — Added keyword generation.
• Memory — Reduced RAM usage during larger tasks.
• Long texts — Added splitting into smaller sections to limit model load.

• Installation — Included Windows libraries to reduce missing-file errors.
• GPU acceleration — Added optional graphics card processing to speed up translation.
• Memory usage — Added CPU processing support when video memory is limited.
• Multithreading — Added support for parallel processing.
• Text structure — Improved formatting of processed content.

• Interface — Minor improvements and fixes.

• First stable release with a redesigned interface.