Start from an image when the subject already exists or has to match another shot; start from text when the shot is a first look at a scene. In Yeşilçam Studios both paths cost the same — a five-second draft clip is 490,000 tokens at the base rate — so the choice is about control more than price.
What are the three sources?
The Source row of the Create panel has three buttons. From text, Pick from library, and Upload image.
From text. The engine builds the first frame from the sentence in the Scene field. Nothing but words constrains the composition.
Pick from library. Any image already generated in the module. The Library also shows an Animate button on every image, and pressing it opens the Create panel in video mode with that image attached as the source.
Upload image. A file from outside the module. A product photo, a location, a storyboard frame, a frame from an earlier shot.
When an image is added, a control appears that text doesn't offer. A Draw Camera Path button shows up under the thirteen ready-made moves, and when the source is text that button isn't there.
When does a still image win?
Giving motion to a still image is older than any model. The Ken Burns effect — a slow pan and zoom over photographs — is the technique used when there's no film footage. Image-to-video is the same decision, made by the engine instead of a pan, and it wins in the same three situations.
There's a subject. A packaged product, a car, a room, a plate of food. A product described in words is a slightly different product in every clip; a product photo is the same product in all of them.
The shot has to match another shot. Two clips opened from the same still share framing, light, and color temperature. This is the cheapest form of consistency the module offers, and the path the character-consistency guide recommends for one or two shots.
The look was already approved in image mode. A draft image is 29,750 tokens, a draft clip 490,000 — so sixteen stills are cheaper than one clip. When one works, feeding it as the source turns the approved frame into the opening frame instead of a fresh guess.
When does text win?
Text wins when the thing you're after is the framing itself. An FPV drone move through a space, a crowd, weather — anything where the engine has to find the composition rather than animate one you chose. A still fixes the frame before the camera move is chosen, and on these shots that's a constraint more than a help.
It also wins on the first attempt at a scene. The first three versions of any prompt belong in image mode anyway, and what goes into that sentence is covered in the how to write an AI video prompt article.
What does Video to Video do?
In the Library there's a third path for motion that already exists. The Video to Video button opens a restyle window that takes a short mp4 of up to 15 MB and a description of the new look — the example in the field is a watercolor-painting aesthetic — and produces a new clip that keeps the old one's motion and composition. Quality is only Draft or Standard, duration five or ten seconds, resolution 480p, 720p, or 1080p.
The window carries one warning worth reading before you upload: videos containing real people can be rejected by the provider.
Where the sources stop
Here's the limit: a still fixes the first frame, not the motion. The engine still decides what moves and how, and a product photo shot from a low angle is animated from that low angle. Restyle caps out at Standard, so the Cinema tiers can't be used for a clip that starts from an existing image.
There's no discount either. With an image attached, the Create panel showed the same four-tier prices as text — 392,000 to 1,498,000 tokens for five seconds on a Pro account, twenty percent under the base rate. Bringing your own image saves re-shoots rather than per-shot tokens.
FAQ
Does starting from an image spend fewer tokens? No. The Create panel showed the same tier prices with an image attached and without one. Price is set by quality tier and duration.
Can I animate a photo of a real person? Upload image accepts it, and the Video to Video window warns that videos containing real people can be rejected by the provider. A person's image rights are something you have to control in either case.
Which files does Video to Video accept? A short mp4 of up to 15 MB. Output is Draft or Standard, five or ten seconds, 480p, 720p, or 1080p.
