Kling 4.0 multilingual text in video: a test plan before you trust it
Kling says Kling 4.0 renders text, logos and emoji in 10+ languages. A 12-sample test plan to run on any video model before a localized ad depends on it.
Kling's announcement says Kling 4.0 renders multilingual text, logos and emoji in more than 10 languages. That is a vendor claim about a model in early access, and Sume does not list a kling-4 model yet, so you cannot run this test on Sume today. The test below works on any video model, including the ones Sume does run.
Run it before a localized ad depends on in-video text, because a single wrong glyph in a headline is a rework.
What should the test contain?
Twelve prompts are enough to see a pattern. Use short, real headlines.
| Variable | Values to cover |
|---|---|
| Script | Latin with accents, Cyrillic, Arabic, Devanagari, Hangul, kana, Chinese |
| Length | One word, a short line, a two-line headline |
| Surface | Flat sign, curved packaging, moving screen |
| Extras | One logo, one emoji |
How do I score it?
For each clip, read the text at full size and again at phone size. Mark exact, one error, or unreadable, and score every frame where the text is held, not only the first. Text that drifts mid-clip counts as a fail for an ad.
What do I do on Sume meanwhile?
The safe pattern is to avoid depending on in-video text at all: generate a clip without text and add the headline as an overlay in the timeline or an editor. Sume's video models are the same call; run the same test prompts on kling-3 and seedance-2.5 and compare the results yourself.
Sources
Related posts
More in Models
- Kling motion control keep_original_sound: get a silent clip
keep_original_sound on Sume's Kling 3.0 Motion Control defaults to true, so the driving video's audio rides along. Send false for a silent clip.
- Which LLM for each step of a video agent: script to assembly
A hypothesis for matching LLMs to video agent steps (script, shot list, frame check, assembly), using documented tool prices, to test with a Format A/B run.
- LTX-2.3 LoRAs on LTX-2.5: most run unchanged; Sume has no LoRA field
Lightricks says the large majority of LTX-2.3 LoRAs and IC-LoRAs run on LTX-2.5 unchanged, with a few exceptions to test. Sume has no LoRA field.
- LTX-2.5 text encoder: use the bundled Gemma 4 12B, not stock
LTX-2.5 needs its own Gemma 4 12B text encoder; Google's stock Gemma 4 release is not a substitute. What to download and what the 66 GB package holds.
Written by Sume