How to Generate Product Shots in Scenario With Your Own Model
How to Generate Product Shots in Scenario With Your Own Model
Short answer: Upload 10–20 photos of your product, train a custom model in Scenario, then generate new scenes by combining that model with a style and an environment prompt. Training takes 15–30 minutes; each generation costs credits from the $29/month Starter plan upward.
The alternative — prompt-and-pray with a general image model — produces a convincing photo of something that is almost your product. Almost is useless for ecommerce.
What is Scenario?
Scenario is an AI image generation platform aimed at game studios and brands. Its two defining features are custom-trained models (your product, your art style) and game engine export — Unity, Unreal and Godot plugins that push generated assets straight into a project. It also runs a Gen-Blend workflow where models can be chained.
Pricing runs from a limited free tier through Starter at $29/month, with credit consumption per generation. Enterprise plans add private models and team seats.
How do you train a model on your product?
Photo quality beats photo count. Fifteen clean shots will outperform fifty sloppy ones.
- Collect 10–20 images of the product from multiple angles: front, 3/4, side, back, top. Include one shot showing scale next to a familiar object.
- Remove backgrounds or keep them consistent. Mixed white-background studio shots plus lifestyle shots usually confuse the training step.
- In Scenario, go to Train Model → Object. Name it with a unique trigger word, like
acmebottle. - Upload the set and set the object type. Bottles, apparel and boxes each take different defaults.
- Start training. Queue wait plus training is typically 15–30 minutes for an object model.
- Test with a plain prompt:
acmebottle on a white surface, studio light. If the label or logo warps, add more straight-on shots and retrain.
Expect the first model to be mediocre. The second one, trained after you see what failed, is usually the keeper.
How do you generate actual product shots?
Generation is a two-part prompt: the trigger word plus the scene.
- Studio pack shot:
acmebottle, on matte concrete, soft key light from left, shallow depth of field, 50mm - Lifestyle scene:
acmebottle, on a cafe table next to a laptop, morning window light, no text - Seasonal variant:
acmebottle, in a winter gift box arrangement, pine branches, warm interior
The failure mode to watch is label distortion. Any text on your packaging will smear unless the model was trained on legible shots. For products where the label carries the brand, plan to composite the real label back in Photoshop — that is faster than chasing a perfect generation.
Another practical one: shading. Generated shadows often do not match the environment, and shoppers notice unconsciously. Fix contact shadows manually or generate three variants and pick the one with a plausible cast shadow.
How do credits actually work?
Each generation consumes credits, and higher resolution or an upscale pass costs more. The Starter plan’s allowance covers a decent test batch — a few dozen images per month at standard resolution — but a 300-SKU catalog needs a bigger plan or a workflow where you generate one hero scene and reuse the background.
The trick that saves credits: generate the scene empty, then paste the product. Or generate at lower resolution in batches, pick winners, upscale only those.
How does it compare with a general image model?
| Requirement | Scenario | General image model |
|---|---|---|
| Consistent product identity | trained model | prompt-dependent |
| Brand style lock | custom style model | reference images per run |
| Engine export for games | native plugins | manual asset import |
| Price floor | free tier, then $29/month | often per-image API pricing |
| Learning curve | moderate | low |
For a game studio shipping 2D assets, Scenario’s engine export is the reason to use it. For a brand doing five product images a quarter, a photographer is still cheaper and better.
What are the honest drawbacks?
- Training is fiddly. Bad input sets produce models that hallucinate detail, and there is no automatic quality warning.
- Label text remains the weak point across the category.
- Credits run out faster than expected once you start iterating on scenes.
- UI density. Scenarios, models, and assets are three separate concepts, and the first hour is spent figuring out which tab does what.
Who is this for?
Ecommerce teams shooting more than about 30 product images a year, and game studios needing branded asset variation. Not for a one-off photo of a single SKU — a local photographer at $150 a shot wins on both quality and risk.
Quick Summary
- Training input: 10–20 multi-angle photos, consistent background, unique trigger word.
- Training time: 15–30 minutes plus queue.
- Price: free tier to test, Starter at $29/month, credits per generation.
- Weak point: text on labels distorts; plan to composite it.
- Best fit: repeatable product catalogs and game asset pipelines.
Last updated October 2026.
