Which AI video model can render text on screen
On-screen text is where AI video still breaks. We asked 30 models for the same animated title card, three words and a date over festival footage, and kept every result including the three refusals. Here is who can spell. The grid includes the September 18, October 19, October 22 and October 23 additions; each model page dates its own run.
Anvisha Pai, Co-founder & CEO, Voyager
What the side-by-side shows
Five models set the card cleanly: Veo 3.1 and Veo 3.1 Fast typeset both lines and held them, Seedance 2.0 animated the words in and kept them intact, MiniMax H3 and Hailuo 2.3 were crisp from the cheapest end of the roster, and Gemini Omni Flash resolved to clean type after a bokeh opening. Kling is the clear failure in both tiers, producing 'MDDA SUMMER SESSIIONS' and 'MODA SUMER SOSSSIONS'; Sora 2 set thin type that wobbled and lost letters by the last frame; Veo 3.1 Lite overlapped its two lines and cropped one; Grok opened with a stray 'MME' before settling. Both Wan endpoints refused the prompt outright on fal's content checker, the only models to do so. Luma Ray 3.2, added later, shows the title faintly for about a second mid-clip and then loses it. FLUX 3 Max, added on October 19, spelled both lines correctly over a navy wash. Of the 22 October additions, Seedance 2.0 Mini set the card exactly as asked, tangerine on a navy bar with the date beneath; Grok Imagine Video 1.5 Lite, Gemini Omni Flash 1.1 and HeyGen Video also spelled both lines. Both LTX 2.5 tiers garbled it, and Wan 3.0 Prime refused it like the other Wan endpoints. Kandinsky 6.0 Pro garbled it, and Kandinsky 6.0 Lite returned a fox reading a newspaper instead. Kling 4.0, added on 23 October, is the first Kling to spell it: both lines and the date, tangerine on navy bars, held after a cut from the crowd to the stage.
- 1.Veo 3.1 typeset both lines on a card and held them still over the crowd.
- 2.Seedance 2.0 animated the words in and kept every letter, with the date on its own line.
- 3.Veo 3.1 Fast the same clean type as the full model at a quarter of the price.
- 4.MiniMax H3 crisp lettering from a $0.30 clip.
- 5.Hailuo 2.3 correct text from the cheapest silent model on the roster.
- 6.Seedance 2.0 Mini tangerine on navy as asked, both lines spelled, the date set smaller beneath.
- 7.Kling 4.0 both lines and the date spelled on tangerine-on-navy bars, where Kling 3.0 garbled the title.
Every video model on this prompt, 27 clips
One try each, nothing re-rolled or retouched, generated by us. Ranked models come first; the rest follow in family order. Open a model for the rest of its set.
›The prompt every model got
A title card reading 'MODA SUMMER SESSIONS' animates in over slow-motion footage of a festival crowd at dusk, then a smaller line 'Friday 12 June' fades in beneath it. Bold condensed type, tangerine on navy.
Set text that stays spelled correctly while it moves.
typeset both lines on a card and held them still over the crowd.
animated the words in and kept every letter, with the date on its own line.
the same clean type as the full model at a quarter of the price.
crisp lettering from a $0.30 clip.
correct text from the cheapest silent model on the roster.
tangerine on navy as asked, both lines spelled, the date set smaller beneath.
both lines and the date spelled on tangerine-on-navy bars, where Kling 3.0 garbled the title.
The other prompts
- Which AI video model holds an art style
- Which AI video model handles a busy scene best
- Which AI video model handles a camera move best
- Which AI video model handles hands and objects best
- Which AI image-to-video model keeps the product intact
- Which AI video model gets physics right
- Which AI video model makes the best product turntable
- Which AI video model makes the best talking head
Frequently asked questions
Which AI video generator is best for text?
On our title card, Veo 3.1 and Veo 3.1 Fast, Seedance 2.0, MiniMax H3 and Hailuo 2.3 all rendered the words correctly. Kling 3.0, in both tiers, could not spell the title. Kling 4.0, added on 23 October, spelled it too.
Can Kling render text?
Kling 3.0 cannot: both Standard and Pro garbled the three-word title on this prompt. Kling 4.0, added on 23 October, spelled both lines and the date correctly in our one run.
Why did Wan refuse the prompt?
fal's content checker returned a policy error on the festival title card for Wan 2.7, Wan 3.0 and Wan 3.0 Prime. No other vendor objected to the same prompt.
Voyager
Run supported models inside Voyager
Voyager is an agent for creative work that runs supported models. Brief it, and it picks the model for the job, makes the thing and hands back something you can edit. Create an account to get started.