A single attractive image is easy to admire and difficult to build a video around. The character may face the wrong way, the important object may sit beneath a caption, or the next image may look as though it belongs to another story. Start by deciding what the images must explain together.
This guide takes a small creator project from a written brief to a usable image handoff. It covers the Nano Banana names you will encounter, how to preserve a subject across shots, and where image generation stops being enough. The three-shot packing brief gives you a concrete reference, two focused edits and a handoff to the motion stage.
What Nano Banana means, and which version you are using
Nano Banana is Google's name for its Gemini image-generation and editing family. Google's current model page lists Nano Banana 2 as Gemini 3.1 Flash Image, Nano Banana Pro as Gemini 3 Pro Image, and Nano Banana 2 Lite as Gemini 3.1 Flash-Lite Image. Older tutorials may refer to the original Gemini 2.5 Flash Image. These names describe different models, rather than spelling variations of one fixed product. Google DeepMind's model-family page, checked 30 October 2026.
Write the model name into your project notes. “Made with Nano Banana” is too vague when you need to revisit an image, compare an edit or explain why another creator sees different controls. Search phrases such as “Nano Banana 2.0” are also not a substitute for the model label displayed in your chosen tool.
The family supports image creation and conversational image editing. That makes it useful for concept frames, recurring objects and visual variations. It does not make the image model a complete video editor. Google's site presents Veo separately as a video model; a still-image capability should not be read as a promise of animated output.
Choose the image's job before choosing a prompt
For a short educational video, each image should either establish context, show an action or make a difference visible. If all three frames only establish the topic, the narration has to carry the explanation alone.
| Image job | What to request | What to check |
|---|---|---|
| Reference frame | Clear subject, costume, object and palette | Distinctive details can be named and reused |
| Explanatory frame | Visible relationship or change | A viewer can identify the important difference |
| Animation starting frame | Room for the intended movement | Hands, props and edges can survive movement |
| Cover | One recognizable subject with uncluttered space | It introduces the episode without requiring its explanation |
A cover and an explanatory frame can share a character while doing different work. Do not force a dense comparison into a cover and then crop away the detail. Equally, a dramatic portrait is not a substitute for the diagram your narration promises.
Build a three-shot brief: the suitcase that will not close
Consider a short story about packing too much for a weekend trip. The viewer should understand one problem and one useful response: remove duplicate items before looking for a larger bag. The production challenge is keeping the suitcase, room and character recognizable while the contents change.
Use a visual anchor: a mustard-yellow hard-shell suitcase, a green luggage tag, a pale wood floor and a character wearing a navy sleeve. The tag is a continuity cue, not a logo. Keep the camera on the suitcase and hands so no presenter needs to appear on screen.
| Shot | Visible information | Later motion |
|---|---|---|
| Problem | Open suitcase with two bulky sweaters; lid cannot meet base | Hands lower the lid until it catches |
| Decision | One sweater lifted out; other packed items remain recognizable | Single removal action |
| Payoff | Same suitcase closed; removed sweater visible beside it | Hand pulls the zipper a short distance |
This brief also tells you what not to generate. A new bedroom, a different suitcase colour or a smiling traveller at an airport would add production work without explaining the decision. Continuity is easier to judge when the intended change is small.
Create the reference frame
Vertical illustration for a short packing story. Overhead three-quarter view of an open mustard-yellow hard-shell suitcase on a pale wood floor. A plain green luggage tag hangs from the right handle. Inside are two thick grey sweaters, one folded white shirt and a small blue wash bag. A navy-sleeved hand holds the lid slightly open. Warm natural light, clean shapes, restrained texture. Keep all important objects in the central area and leave uncluttered floor above the case for a short caption. No lettering or brand marks.
The useful specificity is in the inventory and viewpoint. “Beautiful travel image” would leave those decisions to the model. Here, a reader can compare the output with a list of objects and decide whether it is suitable.
Edit one important variable
Use this image as the visual reference. Keep the suitcase, green tag, floor, camera position, lighting and navy sleeve consistent. Show the hand lifting one grey sweater out of the suitcase. Keep the white shirt and blue wash bag in their original positions. The suitcase remains open. Preserve the clean illustrated style.
Ask for a local change before requesting a new angle. If you ask for removal, a camera move, a lighting change and a new style together, you lose a clear basis for finding the cause of a mismatch. Save the accepted reference before trying variations.
Prepare a frame that can finish the story
Using the same suitcase reference, show the mustard-yellow case closed on the same pale floor. Keep the green luggage tag on the right handle. Put the removed grey sweater on the floor to the left. A navy-sleeved hand rests near the zipper. Preserve the original camera angle and warm light. Leave the zipper unobstructed for a later animation.
The removed sweater matters. Without it, the change from open to closed could look like a jump with no cause. The visual explanation should survive with the sound off, even though the final narration adds context.

Review the sequence before you animate it
Place the three images next to one another. First compare object identity, then composition, then the action. This order avoids polishing colour while a critical prop changes shape.
- Identity: the suitcase keeps its colour, handles, tag and proportions.
- Inventory: the sweater count changes for a reason; the wash bag does not disappear between shots.
- Spatial logic: left and right stay intelligible; the hand can reach the object it is about to move.
- Reading area: the caption can fit without hiding the reason the lid will not close.
- Animation readiness: fingers and zipper are clear, and the composition leaves space for the intended action.
Review at the size the video will actually occupy. Detail visible on a large editing monitor may vanish in a phone feed. A small green tag is useful as a secondary continuity cue; it should not be the only way to tell the cases are the same.
When an edit changes too much, return to the accepted reference and narrow the request. If the important action depends on exact contact between a hand and a zipper, a static illustration with a cut may be more reliable to finish than repeated attempts at intricate animation. That is an editorial choice about this scene, not a measured ranking of models.
Hand images to the video stage with instructions attached
An image filename should help the next production step. Use names such as packing-01-problem.png, packing-02-remove-sweater.png and packing-03-closed-case.png. Keep the model name, accepted prompt and source-reference filename in the same project folder.
Then attach a motion sentence to each frame. For the opening: “Locked camera; the hand gently lowers the lid until it rests against the packed sweaters.” For the middle: “The hand lifts one grey sweater straight upward.” For the ending: “The hand moves the zipper a short distance while the suitcase stays still.” One clear action per shot is easier to judge than a full paragraph of simultaneous instructions.
If you use Runway for the next stage, its image-to-video guide distinguishes the visual reference from instructions describing motion. The same planning distinction is useful even if you choose another video tool. Check that tool's supported input sizes and crop behaviour before sending every frame through it.
What a polished frame cannot prove
A generated image is not evidence that the packing method saves a particular amount of space, that a product has a specific capacity or that a customer achieved a result. If your episode makes one of those claims, obtain the underlying measurement separately. Use generated art to show a concept, rather than to fabricate a product test.
Text and precise labels deserve their own check. When an episode depends on a quotation, price or measurement, set it as selectable text in your editor and compare it with the source. Do not rely on a plausible-looking number inside the image.
Reference-image use also needs a practical boundary. Use images you can supply for this purpose and avoid building a series around a real person's likeness without permission. A model's ability to accept an image does not settle your right to use that image commercially.
Turn the method into a repeatable series
A useful series rule could be “one packing decision per episode, shown with the same suitcase and overhead viewpoint.” New episodes can cover shoes, toiletries or cables while the reader promise stays stable. Repeating the exact sweater story with a different background would not add another useful answer.
Keep a short visual specification: palette, viewpoint, recurring object, space for captions and what may change. The separate Nano Banana prompt examples go deeper into adapting individual instructions. For motion planning, continue with AI camera movements.
If your main need is recurring narrated faceless episodes, rather than manually directing each frame, ShortsFast's faceless video guide explains the series approach. ShortsFast lets you choose a production direction and see a first personalised preview before deciding to buy. It is not a Nano Banana frame editor, and finished ShortsFast videos do not have an in-product content editor.

