Upload one still
Drop a PNG, JPG, or WebP up to 8 MB. This I2V job treats that file as the opening frame. Aspect follows the photo. Wrong types never leave the browser.
Hailuo 3 · I2V
MiniMax H3 Image to Video
Upload one still as the first frame, describe the motion, download an MP4. Buy a pack to generate. 768p unlocks after a purchase. No GPU, no ComfyUI node, no last-frame switch. Independent studio, not official MiniMax.
0:00 / 0:00
0%
Generating your video…
Warming up the model…
Something went wrong.
Motion demo — atmosphere plate, not a site render
Prefer words only? Use the MiniMax H3 text to video generator.
The job
What MiniMax H3 image to video does here
I2V on this site is a hosted MiniMax H3 (Hailuo 3) run: pixels from your still, motion from your sentence, a short MP4 back in the tab. It is the same Account stack as the homepage generator, with image mode already open.
MiniMax documents image-to-video as first-frame, last-frame, or both, plus a separate omni-reference mode. Those contracts exist on the official API and on ComfyUI nodes. This URL only implements the first-frame slice we can actually call: one PNG, JPG, or WebP up to 8 MB, optional motion prompt, 5 / 10 / 15 seconds. Buy a pack to generate. A purchase unlocks 768p. That is the product, not a teaser for 2K.
People also search a local workflow and an image-to-video node because the open weights landed in ComfyUI the same week as the model. Those searches want graphs, GGUF files, and VRAM tables. We do not host them. If you want a still animated without installing anything, stay on this page. If you want the node, Comfy-Org’s docs are the right tab.
The generator is independent. Google sign-in and credits go through Account. Unsafe prompts and images are blocked before a debit. Failed runs do not keep the charge when Account reports a failure. Clips you keep after sign-in sit in My works for about a week. This page does not accept YouTube Shorts, live streams, or music.youtube links as input.
Steps
How to animate a still on this page
Three actions: drop the photo, name the motion, press Generate. Wrong file types never leave the browser. How to use the form is the same studio as the homepage, already opened on the image tab.
Describe the motion
Write the camera move or the one action you want in the still. Do not rebuild the whole scene in words — the pixels already did that job. Press Generate when the brief is one beat.
Generate and download
Pick 5, 10, or 15 seconds after you buy a pack. When the MP4 lands, play it here and save the file. Signed-in clips also go to My works. 768p unlocks after a purchase.
On this URL
What this I2V page actually runs
One file, one motion sentence, one MP4. Buy a pack to generate. Modes the stack cannot press stay off the form instead of appearing as grey tiles.
One still, not a node graph
This URL is a browser upload. You do not install a MiniMaxH3ImageToVideo node, pick a GGUF, or rent a 4090. The still stays in this tab until you press Generate. The job is the same Account path the homepage already runs.
First frame only
Official Hailuo 3 can lock a last frame or a first-and-last pair. This page does not. The photo you drop is frame zero. Motion after that comes from the prompt, not from a second still or a nine-image omni stack.
768p after a purchase
A purchase unlocks 768p, and credits unlock 10s and 15s. We do not sell the 2K regenerate pass on this form.
Moderated hosted path
The stack is a Chinese AutoDL MiniMax H3 workflow behind Account. Unsafe prompts and images are blocked before credits move. Reddit NSFW LoRAs are a different product and are not on this form.
Why here
Why use this hosted path
Use this URL when you already like the still and you only need it to move, without renting a GPU or pretending the official 2K regenerate is a toggle.
No GPU install
ComfyUI I2V wants weights, VRAM, and a node named MiniMaxH3ImageToVideo. This page is for people who will not rent a 4090. The still never becomes a local checkpoint; it is sent through the hosted path we already run.
Limits sit next to Generate
Last frame, omni nine-image, and 2K are named as off. Competitor landings often list those because their API has them. We list them so you do not pay for a switch that is not here.
Failed runs do not keep the debit
Account reports a failure, the charge does not stick. That is the billing contract, not a slogan about unlimited free 2K.
Identity stays in the pixels
A face or packshot you already like should stay the first frame. The prompt is for wind, a push-in, or steam — not for replacing the subject with three new characters the still never showed.
Writing
Image to video prompts that stay on one beat
After the still is on the drop zone, the text box is a motion brief, not a second painting. That is the difference between a clean 5-second I2V take and a fight with the pixels.
A reliable prompt names camera, a single physical change, and maybe weather that is already visible. “Slow push-in, steam rising, cool window light” works on a coffee still. “Replace the room with a spaceship and add three new characters” asks the model to ignore the photo. Official omni prompts that label @image1 as a character plate belong on hosts that accept nine files. Here you have one.
Guest clips are 5 seconds. Pack one action into that window. Hair, steam, a head turn, a dolly. If you need 10 or 15 seconds, sign in and spend credits — the still is still the first frame, the extra time is just a longer hold on the same beat, not a second scene.
One beat per duration
Five seconds holds one action. Hair, steam, a head turn, a dolly. Ten and fifteen seconds are a longer hold on the same beat, not a second scene crammed into the same still. Sign in before you spend credits on a longer take of a prompt that already fights the photo.
Weather only if it is already there
Ask for rain on glass when the photo shows rain. Inventing a storm on a dry street fights the first frame and usually looks like a smear, not weather. The same rule applies to extra people, extra products, and new rooms.
Keep the prompt off the scenery
The still already has the room, the product, and the face. Name camera and one physical change. A second painting in the text box is how I2V turns into a different picture. Official omni prompts that label @image1 belong on hosts that accept nine files.
Empty box still runs
If you only upload a still, Generate uses a cinematic-motion default. Add words when you care about the camera; skip them when the photo should simply start moving. Wrong file types and oversize files are rejected before a request is sent.
Honest no
What this I2V page does not run
Competitors’ image-to-video landings often list last-frame, 2K, and omni references because their API does. We list the same ideas so you can see they are off. The form does not hide them behind a grey tile.
Last frame / FL2VA
A second still as a landing frame is real on MiniMax’s app and on some APIs. It is not a control here. If you need a locked ending composition, use a host that exposes first-and-last-frame I2V.
Omni video / audio references
Official omni input is up to nine images, three video clips, and three audio clips. This page takes stills only: one photo is the opening frame; two to nine photos are identity/style references. Extra video and audio files are not a hidden mode.
ComfyUI workflow / node
Searchers look up a local I2V workflow because the open weights run on a GPU box. This studio does not ship nodes, Sage Attention flags, or a 3060 VRAM table. It is the hosted path only: 480p until you pay, then 768p.
YouTube URLs
This page does not accept YouTube Shorts, live streams, or music.youtube links. There is no paste box for youtu.be or an 11-character ID. The input is an image file plus an optional motion sentence.
Recipes
Who this image to video path is for
Use it when you already have a frame you like and you only need it to move. Skip it when the job is a two-still morph, a nine-image character lock, or a local Comfy graph.
Product still that needs a camera move
Need. A packshot with readable label and a simple background.
Do. Upload the packshot. Prompt only the camera: slow orbit, soft wrap light, one petal or drip if the still already shows it.
Skip. Do not ask the model to replace the label, add a second product, or jump to 2K.
Portrait that should breathe
Need. A face or bust with a clear subject and not a collage of six people.
Do. Keep identity in the pixels. Prompt a small head turn, hair in wind, or a slow push-in. One beat per clip.
Skip. Do not request a last-frame expression change unless you have another host that accepts two stills.
Location plate for a social loop
Need. A landscape or interior with a single vanishing point.
Do. Describe parallax or a dolly. Mention weather only if it is already in the photo (rain on glass, steam, dust).
Skip. Do not paste a YouTube link or a Comfy workflow JSON. This page will not parse either.
Answers
MiniMax H3 image to video FAQ
What is MiniMax H3 image to video?
MiniMax H3 image to video on this page means one still becomes the opening frame of a short MP4. You upload a PNG, JPG, or WebP, describe the camera or subject motion, and this independent studio returns a clip. Buy a pack to generate. A purchase unlocks 768p. It is not the official Hailuo omni app and not a ComfyUI graph.
What is MiniMax H3 I2V?
I2V is the same job as this hosted generator: image in, video out. Here I2V is one reference still as frame zero, then a motion prompt. Official first-and-last-frame I2V and nine-image reference-to-video are different contracts and are not switches on this URL.
Is Hailuo 3 image to video the same model?
Yes. Hailuo 3, Hailuo 3.0, and MiniMax H3 are one family. People search Hailuo 3 image to video for the same generator line. This hosted page runs the one-still slice of that family at 480p until you pay, then 768p. It does not run MiniMax’s 2K regenerate or omni reference stack.
Can I set a last frame or first and last frame?
Not on this page. Official Hailuo 3 can take a first frame, a last frame, or both. Here you may upload one still as the opening frame. If you need a locked landing frame, use MiniMax’s Hailuo app or a host that exposes FL2VA.
How do I write a MiniMax H3 image to video prompt?
Upload the still first, then describe only motion: a slow push-in, wind in hair, a lighting change, or a small expression. Do not rewrite the whole scene or the model fights the pixels. If the box is empty, this generator uses a simple cinematic-motion default. Keep one action per 5-to-15-second clip.
Do I need a ComfyUI MiniMax H3 image to video workflow?
No. Searchers look up ComfyUI workflows and nodes because the open weights run locally. This page is a hosted browser studio. You do not install nodes, pick a GGUF, or rent a 4090. The job here is upload, prompt, generate, download MP4.
What is the MiniMax H3 image to video node?
The MiniMaxH3ImageToVideo node is a ComfyUI widget on local graphs. We do not ship that node. The hosted job on minimaxh3.lol is the same product family called through our Account API with one still: 480p until you pay, then 768p. For the node itself, use Comfy-Org docs, not this URL.
What do Reddit threads say about MiniMax H3 I2V?
Reddit mixes three jobs: local ComfyUI I2V, NSFW LoRAs, and hosted APIs. This page only matches the hosted, moderated path. We cannot run uncensored LoRAs or your 3090 workflow. If you want a still animated in the browser without a GPU, that is the slice we actually sell: 480p until you pay, then 768p.
Is MiniMax H3 image to video free?
No. Sign in with Google and buy a credit pack to generate. 768p unlocks after a purchase. Failed runs are not charged when Account reports a failure.
Which image formats can I upload?
PNG, JPG, or WebP, up to 8 MB. The file is used as the first frame. Aspect ratio follows the photo, so the 16:9 / 9:16 picker hides in this mode. Oversized files are rejected in the browser before a generate request is sent.
How long can an image to video clip be?
On this generator you pick 5, 10, or 15 seconds. A purchase unlocks 768p. Official docs also describe a 2K regenerate pass after a 768p take; we do not offer 2K or 1080p here.
Does MiniMax H3 image to video include audio?
MiniMax H3 is built to return native stereo with the picture. This site downloads a plain MP4 as the stack returns it. There is no separate voice upload or lip-sync control on this page. If you need official omni audio references, use MiniMax’s own app.
Do you offer 2K or omni reference image to video?
No 2K regenerate. You can upload one still as the opening frame, or up to nine stills as references. Official omni video clips and audio clips are not on this form. Buy a pack to generate. 768p unlocks after a purchase.
Are you the official MiniMax or Hailuo product?
No. minimaxh3.lol is independent. We are not affiliated with, endorsed by, or operated by MiniMax, Hailuo AI, or Kuaishou. After a purchase, MiniMax H3 clips on this stack can be 768p. That is still not MiniMax’s own Hailuo app.
Can I download the image to video result as an MP4?
Yes. When a clip finishes it plays on the page and you can download the MP4. Signed-in files also save to My works for about a week. Guest clips stay on this browser until you sign in on the same device.
Does MiniMax H3 image to video work on a phone?
Yes. The generator runs in a modern mobile or desktop browser. There is no app store install, no ComfyUI, and no GPU on your device. Upload the still from the camera roll, write the motion, and wait for the MP4. Buy a pack to generate. 768p unlocks after a purchase.
Animate a still with MiniMax H3 image to video
Drop a PNG, JPG, or WebP, describe one motion, and download the MP4. Buy a pack to generate. 768p unlocks after a purchase. Text-only scenes stay on the homepage generator.
Upload a still