Text to video gets the headlines, but the three questions we get most often as the builders of ReelMint are about starting from something you already have: a photo, a clip, or a portrait. Here is how each one works, which model to use, and what it costs, with free examples you can open to see the result before trying your own.
1. Image to video: animate a photo
Upload a photo or a generated still, describe the motion, and pick an image-to-video model. The photo locks the face, the outfit and the place; the prompt sets what moves. "Slow push in, she turns and smiles, wind in the trees" is a good prompt. "Make it cinematic" is not.
Models: Seedance 2.5 for the best 1080p result up to 30 seconds; Grok Imagine when you want it cheap; Wan 3.0 for native audio. Free examples: a Lisbon street photo brought to life, a poster that starts moving, and an astronaut scene, all on the examples page with their prompts. The guide has the steps.
2. Video to video: restyle a clip, or film from a reference
There are two different things people mean by this, and the studio does both.
Restyle keeps the motion of your clip and changes the look: the street chef becomes claymation, the puppy becomes a storybook cartoon. Wan 2.7 Video Edit and Kling O3 Transformation do this. The output is as long as the input, up to the model's cap.
Reference films a new clip "like this one". On Wan 3.0, Seedance 2.5 and 2.0, MiniMax H3, Kling O3 and Gemini Omni you attach a reference clip plus up to ten reference stills and write a prompt such as "replace the person in Video 1 with the person in Image 1, same motion". The surf clip refilmed by moonlight in the examples is this. Billing counts the reference clip's seconds plus the output seconds, and the button shows the total. The video-to-video guide covers both.
3. Talking photo: make a portrait speak
Give the studio a portrait and either an audio clip or a line of text with a voice, and you get a video of that face speaking with matching lips. Kling AI Avatar does it at 720p and 1080p; OmniHuman 1.5 handles more expressive delivery. Keep the input face front-on and well lit, and keep the first version under 15 seconds while you learn what the model does with your photo. The free examples include three, one of them our assistant Minty saying hello. The talking-photo guide explains voices, including cloning your own from a short sample.
What they cost
All three are priced on the Generate button before you click, from the provider's real cost. Image-to-video on the budget models is a few dozen credits for 10 seconds; restyle and reference jobs cost the input plus the output; talking photos are priced per second of speech. Plans start at Lite, $9.99 a month for 800 credits, every model on every plan.
Then post it
Whatever you make lands in the library with a Share button that publishes or schedules it to TikTok, YouTube, Instagram, Facebook and X, each with its own caption. That is the half of the job the other tools leave to you.
Top comments (0)