From image to video: start frames, end frames and how to control them

The most reliable video does not come from text — it comes from an image: you lock the composition first, and the model only adds motion. In Visnea this works through frames — a start frame and an end frame. Here is the full process.

How it works

1. Open the Studio and switch to "Video" mode.
2. Pick a model. If it supports frames, a "Video frames" block appears under the prompt with uploaders: "Start frame" and — on some models — "End frame".
3. Upload an image. The start frame literally becomes the first frame of the clip: the model animates your exact composition, light and character.
4. In the prompt, describe the motion, not the scene — the scene is already in the picture.
5. Check the "Cost" line and submit. Video renders asynchronously: usually 30 seconds to a couple of minutes; the result lands in your history with a player.

Which model supports what

At the time of publication:

• Start + end: Kling 2.5 Turbo, WAN 2.7, Seedance 2.0 Mini. This is transition mode: the model builds smooth motion from the first frame to the second.
• Start only: Kling 3.0, Kling 3.0 Turbo, Kling 2.6, WAN 2.6, WAN 2.5, Grok Imagine, HappyHorse 1.1, Gemini Omni Video.
• Start required: Hailuo 2.3 and Hailuo 2.3 Pro are pure image animators — the form will not submit without an uploaded frame.
• References instead of frames: Seedance 2.0 and Seedance 2.0 Fast accept up to 4 reference images — not as a first frame, but as a character and style guide. That is how you keep one hero consistent across a series of clips.

Text-to-video still works too: on models where the frame is optional, you can generate from the prompt alone.

Prompting for motion: three ingredients

The formula: what moves + how the camera moves + tempo.

the woman in the portrait slowly turns her head toward the window and smiles,
hair sways gently, camera pushes in smoothly, soft daylight,
calm pace, lifelike micro-movements

steam rises from the coffee cup in a thin stream, blurred cars pass
outside the window, static camera, subtle flicker of sun glare

Common mistakes: re-describing what is already visible in the frame (the model starts repainting the scene), and cramming too many events into 5 seconds — one or two motions look cinematic, five look like chaos.

Transitions between two frames

Start + end mode is a ready-made tool for morphs and story cuts:

1. Generate two stills in "Images" mode — same format (say, both Landscape 16:9) and same style. A convenient trick: build the second frame from the first via References (img2img): "the same scene, but at night".
2. In "Video" mode, upload the first as the "Start frame" and the second as the "End frame".
3. Describe the transition logic in the prompt: "day gradually turns to night, windows light up, the camera pulls back slightly".

Day-to-night, empty-to-crowded, product closed-to-open — all without an editing app.

What it costs

A video's price is the model's base rate times duration, quality and audio — the form recalculates the total before you submit. Reference points at the time of publication:

• Grok Imagine — the cheapest entry: 6 seconds at 480p for 6 cherries;
• Seedance 1.5 Pro — 4 seconds at 480p from 4 cherries;
• Hailuo 2.3 — 6 seconds at 768P for 15 cherries (image-to-video only);
• Kling 3.0 — 5 seconds standard for 34 cherries; 10 seconds doubles it, Pro mode adds ×1.3, AI audio ×1.45.

If a generation fails on the model's side, the cherries are refunded automatically.

A practical route to your first clip

1. Generate a start frame in Seedream 4.5 or Nano Banana, Landscape 16:9 — 5 cherries or less.
2. Animate it in Grok Imagine at 480p, 6 seconds — 6 cherries. Iterate on the motion prompt.
3. Final take — same image and prompt in Kling 3.0 Turbo or Hailuo 2.3 at high quality.

A working prototype costs around ten cherries, and the expensive model runs once — when you already know exactly what you want to see.