Correction, 2026-10-01: an earlier version listed Kling v3 as a text-to-video option on 8frame, with its shapes "from text" and a text-only example prompt. On 8frame, Kling v3 Standard and Pro are image-to-video only and need a start image. The Kling text-to-video route is Kling 2.6 Pro, which the shapes, the example and the prices now use. A second pass the same day fixed three more things: the page said Seedance generates sound with every clip (on 8frame its sound is on by default, costs the same either way and can be switched off), gave the shortest single generation as 2 seconds (Grok Imagine goes down to 1), and lumped the shapes of Seedance and Hailuo H3 together; each model's shapes are now listed, along with which models need a start image.
Correction, 2026-09-30: the first version of this page said Kling v3 generates native 4K and Veo 3.1 runs at up to 4K and 60fps on 8frame (video on 8frame tops out at 1080p, or 2K on Hailuo H3 and Creatify Boreal), put clip lengths at up to 60 seconds, described example clips with generation times we cannot reproduce, and said Sora 2 workflows had largely migrated, which we can't support. This version uses registry settings.
Text-to-video AI is a class of generative model that takes a written prompt and outputs a video clip, handling camera motion, lighting and timing without filming or editing.
The process is the same across the major models: you describe a scene, the model generates frames conditioned on that description, and you get a rendered clip. On 8frame, single generations run from 1 second (Grok Imagine) to 30 (Wan 3.0 and Seedance 2.5). Models differ in how they handle motion physics, prompt adherence and consistency across frames: a model that looks sharp on a static shot can still fall apart on a walking figure or a panning camera, so test your own prompt on the model before committing to it.
How text-to-video AI works
Many current video models are diffusion models trained on large video datasets. The model learns how text relates to visual patterns, then at generation time it denoises a random signal into a sequence of frames that match your description. Nothing is recorded or composited; every pixel is synthesized.
What you control:
- Prompt. "Slow dolly toward a coffee cup on a marble counter, soft morning light" produces a very different result from "coffee cup, table."
- Shape. Veo 3.1 makes 16:9 and 9:16 only. From text, Kling 2.6 Pro and Wan 3.0 offer 16:9, 9:16 and 1:1; Seedance offers 16:9, 9:16, 1:1, 4:3, 3:4 and "Auto"; Hailuo H3 offers 21:9, 16:9, 4:3, 1:1, 3:4 and 9:16. Every model, shape by shape: AI video aspect ratios by model.
- Resolution. On 8frame: Veo 3.1 at 1080p (Lite 720p), Wan 3.0 at 480p, 720p or 1080p, Seedance at 480p or 720p, Hailuo H3 at 2K. Google's own API offers Veo 3.1 at 4K; 8frame doesn't.
- Length. Veo 3.1 4, 6 or 8 seconds; Kling 2.6 Pro 5 or 10; Seedance 2.5 4 to 30; Wan 3.0 2 to 30; Grok Imagine 1 to 15. Longer pieces are cut from several generations.
- Sound. Some models generate audio with the picture. Veo 3.1 and Kling 2.6 Pro have it as a setting that changes the price (Veo 3.1: 448 credits for 8 seconds with audio, 224 silent; Kling 2.6 Pro: 95 for 5 seconds with audio, 47 silent). Seedance has sound on by default and costs the same with or without it. Grok Imagine and Gemini Omni Flash always add sound.
- Reference inputs. Most models also accept a start image, which makes it image-to-video, a related but distinct mode.
Which models run from text alone
Not every video model on 8frame takes a prompt on its own. Kling 2.6 Pro, Wan 3.0, Veo 3.1, Seedance, Hailuo H3 and Gemini Omni Flash run from text alone. Kling v3 (Standard and Pro), Kling O3 and Wan 2.2 need a start image. The Kling tool opens on Kling v3 Standard, so for a text-only Kling prompt, switch the node to Kling 2.6 Pro. The full list, with what each model needs: which AI video models work from text only.
When you use text-to-video AI
Use it when you need moving visuals without a shoot. Common jobs:
- Ad creative. A product in motion with a subtle camera push, in several variants for testing.
- Brand films and mood pieces. Wide landscapes, atmosphere and light; Veo 3.1 renders them at 1080p in 16:9.
- Storyboard prototypes. Rough motion comps before committing a crew to a concept; Veo 3.1 Lite (17 credits for 4 silent seconds) and Wan 3.0 at 480p (24 for 5 seconds) are cheap enough to try many.
- UGC-style content. Kling 2.6 Pro in 9:16 with a lifestyle prompt.
Text alone is the wrong tool when you need a specific face, a real location or exact product placement. Those need a reference image or a hybrid with real footage.
Examples to try
Veo 3.1:
Aerial view of a coastal city at golden hour, slow push forward, shallow depth of field
Set 16:9 and 8 seconds on the node. The Veo tool opens on Veo 3.1 Fast, 168 credits for 8 seconds with sound; Veo 3.1 itself is 448. What to check: light scatter on the water, and whether the push stays smooth without warping buildings.
Kling 2.6 Pro:
Young woman unpacking a skincare product, natural window light, lifestyle feel
The Kling tool opens on Kling v3, which on 8frame animates a start image, so switch the model to Kling 2.6 Pro for a text-only prompt. Set 9:16. Kling 2.6 Pro costs 47 credits for 5 silent seconds, 95 with audio. What to check: hands on the packaging and a product that keeps its shape.
Sora 2 (retired): OpenAI shut down the Sora 2 models and its Videos API on 2026-09-24, and 8frame no longer offers Sora 2 for new generations. See Sora 2 vs Veo 3.1 for where Sora-style work can go.
FAQ
What's the longest clip text-to-video AI can make?
On 8frame, 30 seconds in one generation (Wan 3.0 and Seedance 2.5). Longer videos are assembled from several clips.
Can every AI video model work from text alone?
No. On 8frame, Kling v3 Standard and Pro, Kling O3 and Wan 2.2 need a start image. Kling 2.6 Pro, Wan 3.0, Veo 3.1, Seedance, Hailuo H3 and Gemini Omni Flash run from a prompt alone.
Can text-to-video AI make 4K?
Not on 8frame, where video tops out at 1080p (2K on Hailuo H3 and Creatify Boreal). Google's Gemini API offers Veo 3.1 at 4K for $0.60 a second.
Does text-to-video AI make sound?
Some models do. Veo 3.1 and Kling 2.6 Pro have an audio setting that changes the price; Seedance has sound on by default at the same price either way; Grok Imagine and Gemini Omni Flash always add it.
Related concepts
- Image-to-video: a still image is the starting frame instead of a text prompt. Most models support both, and some (Kling v3, Kling O3, Wan 2.2) support only this.
- Camera motion: dolly, pan, orbit, set in the prompt or through model-specific controls.
- Prices across models: AI video price per second by model.
- Which models put sound in the output: AI video models with audio.
- Veo 3.1, Kling v3 and Sora 2 compared: Veo 3.1 vs Sora 2 vs Kling v3.
Sources
- OpenAI deprecations (Sora 2 and Videos API shutdown 2026-09-24): developers.openai.com/api/docs/deprecations, checked 2026-10-01
- Gemini API pricing for Veo 3.1 (Standard: $0.40 a second at 720p and 1080p, $0.60 at 4K): ai.google.dev/gemini-api/docs/pricing, checked 2026-10-01
- 8frame video prices, lengths, shapes and resolutions: the 8frame tool registry (duration, aspect-ratio and resolution options per tool) and aspect-ratio configuration, read 2026-10-01
- Which models need a start image, Kling v3 start-image requirement, the Kling and Veo tool defaults: 8frame video generation code and tool registry, read 2026-10-01
- Prices: 8frame tool registry credit functions (calculateVeo31Cost, calculateKling26Cost, Wan 3.0 rates), read 2026-10-01
- Seedance audio setting and price: 8frame tool registry (the Seedance node's audio toggle; calculateSeedance2Cost takes no audio input), read 2026-10-01
Run text-to-video on Veo 3.1, Kling 2.6 Pro, Seedance, Wan and more from one board.
Try it without signing up: the 8frame canvas opens on a real board. The canvas is free and unlimited; generation is paid from $19/month.