Create Video with AI
Create video with AI from a sentence, a first frame or reference images, with Seedance, Veo, MiniMax and Wan
Create video with AI: what it is and who it's for
blubiu lets you create video with AI from a shot description, optionally with a first frame, a first and last frame, or up to 9 reference images: choose Seedance 2.0, MiniMax H3, Veo 3.1, Wan 3.0 or another model, get 1 to 15 seconds per clip depending on the model, and pay per second or per clip.
Checked 2026-11-03
Who it's for
- +Single-beat clips: a product reveal, a character action, an atmospheric shot.
- +You already have an image and want it to move: add it and it becomes the first frame, exactly as in image to video.
- +You want to test motion cheaply at 480P, then rerun the keeper in high definition.
- +You haven't settled on an input mode: the same box does text to video, first and last frames and reference images, switching by what you add.
- +Transitions, transformations and before-and-after shots: add a first and a last frame, and the model fills in only the motion between them.
- +An IP character has to look the same in every clip: add up to 9 reference images (character sheets, outfits, props, settings) and say in the prompt what each one is.
- +You want to try one prompt on several models: Seedance 2.0, Veo 3.1 Fast, MiniMax H3 and Wan 3.0 sit in the same model list.
When not to use it
- −You want a whole multi-shot short film — hand it to the Agent, which writes the script, plans the shots and generates each clip. Go →
- −You need 2K — see MiniMax H3. MiniMax H3 (Hailuo) AI Video Generator →
- −You need one continuous shot longer than 15 seconds — split it into clips and join them.
How to create video with AI
- 01
Add a frame, two frames or references
Add one image and the clip starts from it, so the subject stays far more stable than with words alone; add a second and the clip ends on it; add three or more and they become references for a brand-new shot. Add nothing and the prompt does all the work.
- 02
Describe just these seconds
One shot, one action: who does what, how the camera moves, what mood. Do not cram a plot into 5s.
- 03
Billed per second
Resolution and duration drive the cost, and the Generate button shows it before you confirm (Veo 3.1 Fast is one price per clip). Test motion at 480P for 5 seconds, then rerun the keeper.
Tips to create video with AI that looks right
Let what you add pick the mode
The box reads your media: nothing is text to video, one image is a first frame, two images are a first and a last frame, three or more are references, and any video or audio clip makes it an omni reference. With one or two images, click an image's tag to switch between frame and reference; the line under the box shows what was detected.
Choose the model by what the shot needs
Sound: Seedance 2.0, Veo 3.1 Fast or Wan 3.0; MiniMax H3 and H3 Lite are silent. 2K: only MiniMax H3. Under 4 seconds: H3 Lite (from 1 second) or Wan 3.0 (from 2). Exactly 8 seconds at one price: Veo 3.1 Fast. If the selected model can't take what you added, the box switches to one that can and tells you why.
Draft at 480P, finish once
480P is the cheapest resolution. Get the motion right with Seedance 2.0 Mini (the default) or MiniMax H3 Lite, then rerun the same prompt on Seedance 2.0 at 1080P or MiniMax H3 at 2K. A rerun is a new take, not the same clip made sharper; to keep the exact draft, run it through the AI video upscaler.
Write one shot and name the sound
Each generation is one continuous shot: who does what, how the camera moves, what mood. Several unrelated actions in a few seconds come out rushed or skipped. Seedance, Veo and Wan generate sound by default, so say what you want to hear, such as street ambience, no music; otherwise you may get music you didn't ask for.
Keep a character the same across clips
Words alone redraw the character on every run; in our first full test, clips made from text alone drifted away from the character design. Make the character once as a character turnaround sheet or a consistent character sheet, add it here as a reference image, and describe the character with the same sentence in every clip.
Plan longer films as several clips
No model here makes one shot longer than 15 seconds. Split the scene into shots and describe the end of each clip as the start of the next. When a cut has to line up exactly, end one clip on an image with a last frame and use that same image as the next clip's first frame.
First and last frames: make the last from the first
Everything that differs between the two frames has to be invented in between, so the closer they match, the cleaner the motion. Give the first frame to the image generator AI as a reference and describe only the change, such as sitting instead of standing, so the character, setting, camera height and light stay put. Then describe the path, not the ends: what moves, which way the camera travels and how fast. Seedance 2.0, Veo 3.1 Fast, MiniMax H3 Lite and Wan 3.0 take a last frame; MiniMax H3 doesn't.
References: number them and say what each one is
References are numbered in the order you add them. Start the prompt with what each one is, such as image 1 is the character, image 2 is the café, image 3 is the cup, and only then write the action; unlabelled references get blended together. Seedance 2.0, MiniMax H3 and H3 Lite, and Wan 3.0 take up to 9; Veo 3.1 Fast takes none.
Control the framing with a storyboard sketch
A first frame and references can't be combined in one generation. To fix the layout and keep the character, sketch the shot and add the sketch as one more reference after the character images, labeled in the prompt as layout only, rather than using it as a first frame.
Create video with AI: examples
All of these were generated on blubiu.
Ways to create video with AI, compared
| Tool | You give | You get | Best for |
|---|---|---|---|
| Create video with AI (this page) | A prompt, plus any images, video or audio you add | A 1–15 second clip; the input mode is picked from what you add | Starting anywhere: one box for every input mode and every video model |
| Text to video | Words only | A clip where the model invents the whole picture | Ideas with no images yet: establishing shots, mood shots, quick tests of a direction |
| Image to video | One image as the first frame, plus the motion | A clip that opens on your exact image | Animating a picture whose framing and subject are already right |
| First and last frame video (this page) | Two images, opening and ending, plus what happens between | A clip that starts on one image and lands on the other | Transitions, transformations and before-and-after shots where both ends are fixed |
| Reference to video (this page) | Up to 9 reference images (characters, props, settings), each labeled in the prompt | A new shot with those characters and objects in it | Keeping a character on-model across many clips |
Models to create video with AI
Make 4–15 second videos with sound from text, a first frame or up to 9 reference images with ByteDance's Seedance 2.0 — Mini, Fast and 2.0 compared.
Generate an 8-second 720p clip with optional sound using Google's Veo 3.1 Fast, from text or from a first and last frame. One fixed price per clip.
Use MiniMax H3, the Hailuo video model, for 2K clips with reference images and videos, or H3 Lite for the cheapest silent drafts from 1 second.
Use Alibaba's Wan 3.0 for 2–15 second videos with sound from text, a first frame or reference images, or to replace the person in a video.
Create video with AI: FAQ
How long does it take to create video with AI?
It depends on the model, the length and the resolution; one measured example is Wan 3.0, which takes about 6 minutes on Standard and about 1 on Prime. You can close the page: the job finishes server-side and shows up on your Creations page.
Why is video so much more expensive than images?
Most video models are billed per second, and a few seconds of video costs the upstream provider many times what one image does. The cheapest ways in are a few seconds of MiniMax H3 Lite at 480P, or one 8-second clip from Veo 3.1 Fast.
How long can a clip be?
It depends on the model: Seedance 2.0 runs 4–15 seconds, Wan 3.0 2–15, MiniMax H3 5–15 and H3 Lite 1–15, and Veo 3.1 Fast is fixed at 8. Longer films are made by generating shots separately and joining them.
Are credits refunded on failure?
Yes — credits are refunded automatically when the upstream job fails, the submission is rejected, or you cancel.
Which video model should I start with?
Start with Seedance 2.0 Mini, the default: it takes a prompt, a first frame or references and can add sound. Switch to MiniMax H3 for 2K, Veo 3.1 Fast for a fixed 8-second clip with sound, Wan 3.0 for clips from 2 seconds or a fast Prime run, and H3 Lite for cheap silent drafts. The AI video models page lists them side by side.
Can I create video with AI for free?
Yes. The Free plan includes monthly credits, and the Generate button shows the exact cost before you confirm; a short 480P draft is the cheapest way to start. Failed or rejected generations are refunded, and paid subscribers get 40% off every video model; see pricing.
Can it make video with sound?
Yes, with Seedance 2.0, Veo 3.1 Fast and Wan 3.0: sound is on by default and is generated together with the picture, and the Sound on / Muted switch turns it off. MiniMax H3 and H3 Lite always produce silent video.
What aspect ratios and resolutions can I pick?
Ratios are 16:9 (the default), 9:16, 1:1, 4:3 and 3:4. Resolution depends on the model: Seedance 2.0 and Wan 3.0 go up to 1080P (Seedance Mini and Fast up to 720P), Veo 3.1 Fast is 720P, MiniMax H3 renders 768P or 2K, and H3 Lite 480P or 768P.
Can I add my own images, video or audio?
Yes: up to 9 images, 3 video clips and 3 audio clips per generation. One image becomes the first frame, two become a first and a last frame, three or more become references, and any video or audio turns on omni reference (Seedance 2.0, MiniMax H3 and Wan 3.0; H3 Lite takes audio but not video). Seconds of reference video are billed on top of the output. To make a person in a photo copy the moves from a clip instead, use the motion control AI video generator.
Which models support first and last frames?
Seedance 2.0 (Mini, Fast and the full model), Veo 3.1 Fast, MiniMax H3 Lite and Wan 3.0 (Standard and Prime). MiniMax H3 doesn't take frames at all, only reference images. Veo 3.1 Fast is a fixed 8 seconds; the others let you pick the length.
Can the first and last frames be very different?
Yes, but the bigger the gap, the more likely the middle turns into an abrupt morph. Make an in-between image and generate two clips, first to middle and middle to last; the shared image lines up the cut. If the two images share nothing, such as different characters in unrelated places, generate two separate clips and cut between them instead.
Reference images vs a first frame: what's the difference?
A first frame is the exact picture the clip opens on. References put the characters or objects from your images into a brand-new shot, so the opening frame doesn't have to match any of them. One generation uses either frames or references, not both; with one or two images, click an image's tag to switch it from frame to reference.
Can I show a transformation or an outfit change?
Yes, that is what a first and a last frame are for. Keep the pose, camera and setting the same in both images and change only the look, for example with two outfits from the AI outfit changer. The fewer things change at once, the cleaner the middle.
What can I do with a clip once it's made?
Every clip appears on your Creations page. Sharpen a draft you like with the AI video upscaler, or open the canvas at the bottom of this page to wire script, character, shot and voice nodes into a multi-shot film generated clip by clip.
Before and after you create video with AI
Image Generator AI →
Make the opening frame as a still first: fix the framing in an image, then add it here so the clip starts from it.
Consistent Character Sheet Generator →
Make the reference first: three face angles plus front, side and back full body on one sheet, ready to add here as a reference image.
AI Character Turnaround Sheet Generator →
Design a recurring character once as a front, side and back sheet, then add it as a reference so every clip shows the same character.
AI Video Upscaler →
Keep a 480P or 720P take exactly as it is and lift it to 1080P, 2K or 4K; the motion and length don't change.
Recast Video with AI →
Already have footage? Put your character into an existing clip instead of generating the motion from scratch.
AI Video Models →
See every video model side by side: inputs, clip length, resolution, sound and price per second.
Other creation tools
Wire script, character, shot and voice nodes into one pipeline and generate it clip by clip, with the same character in every clip.