Turn One Image Into a Full AI Shot List with MiniMax H3

Quick Answer

You can turn one approved product image into a full AI shot list by separating shot planning from video generation.

In this workflow, I started with one 16:9 hero image of a wireless gaming controller. I used that image to generate a six-panel shot sheet, cropped the sheet into six independent shot references, and then sent each shot into MiniMax H3 as a separate video branch.

The workflow was:

one hero image → six-shot planning sheet → crop each shot → generate each shot separately → select the best sections for the final edit

The most useful rule from my test was simple:

Plan the shots together. Generate them separately.

The original product image worked well as an identity anchor for shots close to its original viewpoint. But for a larger camera-angle change, the original image could pull the result back toward its own hero composition. In that case, the shot-specific reference needed more control.

What You Need Before Turning One Image Into a Shot List

Start With One Clear Product Hero Image

I used a fictional wireless gaming controller so the test could focus on the workflow rather than an existing brand.

The master image was generated with GPT Image 2.5 Sunburst at 1024 × 576, or 16:9, at 1K resolution.

The controller had a matte dark body, blue accent lighting, textured grips, two analog sticks, a D-pad, and four face buttons. I saved this image in Astorie as:

@controller_dark

This image became the visual source of truth for the product.

Before expanding any product image into more shots, check the details that should stay stable:

• overall shape and proportions
• button layout
• stick placement
• materials and textures
• color accents
• visible seams and structural details

A product can look consistent from shot to shot while still changing small design details. That matters more in product work than it does in many character or lifestyle examples.

Decide What the Final Edit Actually Needs

A shot list is not just a collection of different angles.

Each shot should have a reason to exist.

For this test, I planned a short product launch video of about 10 seconds. I used six shot roles:

The shots did not need to form one continuous camera move.

That was intentional.

A short product promo can use jump cuts between a reveal, a hero frame, a macro detail, another structural angle, and a closing shot. Forcing all six ideas into one continuous generation would make the video harder to control.

How to Turn One Image Into a Full AI Shot List With MiniMax H3

Step 1: Generate One Shot Sheet From the Product Image

I used the original controller image to generate one 2 × 3 shot sheet.

The six panels were designed as independent promotional shots:

• silhouette reveal
• three-quarter hero
• front-control close-up
• trigger close-up
• alternate structural view
• closing hero

The important instruction was not just to make six images.

I asked for six different editorial jobs.

That distinction mattered.

If two panels are both described as "a different three-quarter product angle," they can easily become nearly identical. I saw this problem with the original plan for Shot 5. It was supposed to add a different structural view, but its role was too close to the main hero shot.

A stronger shot list defines the purpose of each panel before worrying about camera language.

Step 2: QA the Shot Sheet Before Generating Video

The six-panel sheet was good enough to continue, but it was not perfectly accurate.

The overall controller identity stayed recognizable across the sheet. The dark body, blue lighting, button arrangement, and general form remained stable.

Small structural details still changed between panels.

That gave me an important product-workflow rule:

Consistency is not the same as accuracy.

A sheet can clearly show "the same controller" while changing the exact shoulder geometry, edge structure, or other small parts.

I therefore checked two things separately.

First, I checked shot diversity:

• Does every panel serve a different purpose?
• Are the macro shots actually close-ups?
• Is the alternate view meaningfully different from the hero shot?
• Do the opening and closing shots feel distinct?

Then I checked product accuracy:

• Are the buttons still in the right places?
• Are both sticks preserved?
• Does the D-pad keep the same form?
• Do the grip textures stay recognizable?
• Do the blue accents remain in the correct areas?

I did not regenerate the whole sheet because the result was already useful enough for shot planning.

The goal was to build a workable production system, not to spend credits chasing a perfect contact sheet.

Step 3: Crop the Shot Sheet Into Independent References

Next, I used Astorie's grid-crop workflow to separate the six panels.

The canvas changed from:

one image → one six-panel sheet

to:

one image → one shot sheet → six independent image branches

Each cropped panel could now act as its own shot reference.

This was the point where the shot sheet became more than a planning image.

Each panel could be:

• sent into video generation
• regenerated independently
• replaced without affecting the other shots
• compared with its resulting video
• kept or rejected on its own

That is more useful than treating the entire storyboard as one final video input.

Step 4: Separate Product Identity From Shot Composition

My first H3 setup used two visual references for each shot.

The original product image, @controller_dark, was intended to preserve product identity.

The cropped shot image was intended to preserve framing and composition.

Conceptually, the roles were:

master image → product identity
shot crop → shot composition

For most of the shots, this worked well.

Shot 1 stayed close to the planned silhouette framing.

Shot 2 stayed close to the hero composition.

The macro shots also worked because they mainly moved closer to parts already visible in the original image.

The problem appeared when I asked for a larger full-product viewpoint change.

Step 5: Watch for Reference Conflict on Large Angle Changes

Shot 5 exposed the main limitation in my test.

Its cropped reference showed the controller from a more unusual structural angle. The full product orientation was clearly different from the original hero image.

I first generated it with both:

@controller_dark + Shot 5 crop

The video moved back toward the original hero-style angle.

I tightened the prompt and explicitly said that the Shot 5 image should control the camera angle while @controller_dark should only preserve product details.

The result still moved toward the original product orientation.

That told me the two references were not behaving like perfectly isolated control layers.

The master image was not only influencing product identity. It was also carrying viewpoint information.

This produced one of the most useful conclusions from the whole test:

When the master image and shot reference imply very different full-product viewpoints, the master image can pull the generated shot back toward its original composition.

For that kind of shot, I stopped trying to solve the conflict with longer instructions.

Instead, I used the cropped Shot 5 image as the main visual anchor.

This kept the workflow efficient and avoided repeated retries.

Step 6: Generate Each Shot as a Separate H3 Clip

I generated all six video shots separately.

Each H3 clip was 4 seconds, because that was the shortest generation length available in my workflow.

The planned final promo was still only about 10 seconds.

That means the generation timeline and the editing timeline were not the same thing.

I treated every 4-second output as source footage.

The planned motion stayed simple:

This was enough.

A short product shot does not need several camera moves inside the same four seconds. More motion would only make it harder to find a clean section for editing.

Step 7: Treat the H3 Outputs as Source Footage

The six outputs do not need to be used at full length.

A 10-second edit could take only the strongest part from each 4-second clip.

For example:

reveal → hero → detail → detail → form → closing

The final edit can use hard cuts, short transitions, or a simple music bed.

I did not add voiceover during generation because voice was not part of the problem this workflow was testing.

The main question was whether one product image could become a usable visual shot system.

It could.

What I Learned From Building the Six-Shot Product Workflow

One Image Was Enough to Start the Whole Shot System

I did not create a front-side-back product sheet before building the shot list.

I started with one hero image.

That image was enough to create:

• the full six-panel shot plan
• six cropped composition references
• six independent H3 generation branches

This kept the article faithful to the actual problem: starting with one product image.

It also kept the workflow fast.

A new product-view sheet should only be added if a specific shot truly needs information that the original image cannot support.

Similar Views Were Easier Than Large Viewpoint Changes

Not every shot had the same difficulty.

The reveal, hero, macros, and closing shot stayed fairly close to the original product view. Those were easier to control.

The difficult case was the alternate full-product orientation.

That shot asked the model to preserve the same object while also changing how the whole object was seen in space.

The original hero image then became both helpful and restrictive.

It helped preserve the product.

It also pulled the camera back toward the original composition.

That means I would not use the same reference strategy for every shot.

A master image is useful when the new shot remains visually close to it.

A shot-specific reference may need more authority when the new viewpoint changes substantially.

The Shot Sheet Was More Useful as a Planning Asset Than as One Final Input

The most useful part of this workflow was not the six-panel image itself.

It was what happened after the sheet was cropped.

Each shot became its own editable branch.

That made the workflow easier to manage because I could:

• inspect one shot at a time
• rerun only one weak shot
• keep successful shots untouched
• change reference strength by shot
• compare each planned frame with its video result

This is especially useful for short product promos, where the shots do not need continuous motion between them.

The edit can create the rhythm later.

Four-Second Generation Does Not Mean Four-Second Editing

I originally planned the promo around much shorter shot lengths.

Then I found that my H3 workflow used a minimum 4-second generation.

That did not require changing the shot list.

It only changed how I treated the outputs.

Instead of trying to make every generated clip equal its final timeline length, I generated clean four-second source clips and planned to extract the strongest section later.

The practical rule is:

Generate for usable footage. Edit for final timing.

Common Problems When Building a Shot List From One Image

Two Shots Look Too Similar

This usually starts in the shot plan.

If two shots both ask for a "three-quarter product angle," a model may produce two versions of the same hero composition.

Give each shot a different editorial purpose.

For example:

hero view and top-form view

is clearer than:

hero angle and alternate hero angle

H3 Moves Away From the Planned Camera Angle

Check whether the master image and the shot reference disagree.

In my Shot 5 test, the master hero image pulled the video back toward its original orientation even when the cropped shot showed another viewpoint.

I stopped adding more prompt instructions and changed the reference strategy instead.

For a large viewpoint change, the shot-specific image may need to become the main visual anchor.

Product Details Change Across the Shot Sheet

Do not judge only by overall resemblance.

Check the actual product structure.

AI-generated panels may preserve color and general form while changing smaller physical details.

For a real product, compare every approved shot against the original asset before moving into final production.

The Generated Clips Are Longer Than the Final Promo

That is normal in this workflow.

The generated clips are footage options, not fixed timeline blocks.

Keep the motion simple enough that several parts of the clip remain usable.

Then choose the best section during editing.

FAQ

Can I build a full MiniMax H3 shot list from only one image?

Yes, one approved image can be enough to build the planning workflow.

In my test, one controller hero image became a six-panel shot sheet, six cropped shot references, and six independent video branches.

Large unseen viewpoint changes were less predictable than shots close to the original angle, so I would not assume that one image can accurately define every hidden part of a real product.

Should I send the whole shot sheet into MiniMax H3?

You can keep the sheet as a planning asset and crop it into individual shot references.

That was the approach I used because each shot could then be generated, reviewed, and rerun independently.

Should every H3 shot also use the original product image?

Not automatically.

The master image worked well as an identity anchor when the shot stayed close to its original viewpoint.

For a major viewpoint change, I found that the master image could compete with the cropped shot composition. In that case, the shot-specific image needed more control.

Do I need to edit all six clips into a finished commercial?

No.

If your goal is to build the shot system, the six independent clips are already useful production outputs.

A simple edit can confirm that the coverage works, but the value of the workflow comes from having separate, controllable shots rather than one finished timeline.