How to make product and promo visuals without booking a shoot

This page may contain affiliate links from which Blur Studio earns commissions.

A product photo shoot is slow and expensive for a reason: it's the only way to guarantee the visual matches the actual product, lit well, from the right angle, consistently across a whole catalog. AI-generated visuals trade some of that guarantee for speed. That's a fair trade for a lot of marketing work (promo graphics, social variations, ad creative testing) and a bad trade when the image's job is to show exactly what a customer will receive.

What does ElevenLabs' Image & Video tool actually do?

ElevenLabs built Image & Video into its broader creative platform, ElevenCreative, generating images and short video from a text prompt, with an Avatars feature added in June 2026 that lets you assemble a consistent "character" (built from source images) and put them in different scenes and outfits with a default, swappable voice attached. On the Free plan, image generation is included; Starter and above add video generation on top of that.

The useful technical detail most people don't know before they start: speech generation for avatar video is billed at 1 credit per character, with hard caps of 5,000 characters and 3 minutes per single generation. If you're scripting a longer promo, you'll be generating it in chunks rather than one continuous take, which matters for planning your workflow and your credit budget before you start, not after.

A more recent addition worth knowing about: per ElevenLabs' own changelog, image generation added support for GPT Image 2.5 models in September 2026, which brought reference-image input and 4K output into the platform. Reference-image support matters specifically for the consistency problem that trips most people up with generated product visuals: instead of describing your product from scratch in every prompt and hoping the result matches the last one, you can feed in an actual photo of your product as a reference and ask the model to place it in a new scene, which keeps the product itself far more consistent across a set of variants than a pure text prompt ever will.

A worked example: a seasonal banner set across several products

Say you're running a seasonal promo across 8 products and need a consistent-looking banner for each one. Rather than writing 8 separate prompts from nothing, start with one real product photo as your reference image, write a single prompt describing the seasonal scene and mood you want (a specific lighting, a specific background, a specific color palette tied to the season), and generate each product's banner using that same prompt with that product's reference photo swapped in. The seasonal styling stays consistent across all 8 because the prompt doesn't change, while each product still looks like itself because the reference image anchors it. Check all 8 side by side before publishing, not just the first one, since even with a reference image the model can occasionally drift on color or proportion on a given run; regenerate anything that's off rather than letting a slightly-wrong banner go live just because 7 of 8 looked right. To turn the same set into video, see seasonal promo videos built in one day.

What I'd actually tell you about using this for product visuals

For promo graphics, seasonal variations of an existing hero image, or ad creative you're going to A/B test and throw half of away anyway, generated visuals are genuinely good value: you get volume and speed a shoot can't match, and the stakes of any one version being slightly off are low. Using images to test ideas first is covered in testing ad ideas with images before you pay for a photoshoot. For your actual product listing photos, the ones a buyer uses to decide whether what arrives matches what they saw, I'd be much more careful. Customers notice when a generated image shows a product with a texture, color or proportion that doesn't match reality, and that mismatch costs you in returns and bad reviews more than it saves you in shoot fees.

The honest rule I use: generated visuals for anything that's selling an idea or a feeling (a lifestyle promo, an ad concept, a seasonal banner) and real photography for anything that's selling a specific physical object a stranger is about to pay for sight unseen. Client-facing versions of this work are covered in AI product photos and videos for small ecommerce clients.

Step by step: making a promo visual without a shoot

1. Start with a specific, detailed prompt describing the scene, lighting and mood you want, not just the product. Vague prompts produce generic-looking results. "A candle on a wooden table" produces something generic; "a white ceramic candle on a weathered oak table, soft morning light from the left, shallow depth of field, cozy autumn mood" gives the model enough to work with to produce something you'd actually want to publish.
2. If you have an actual photo of the product, upload it as a reference image rather than describing the product in the prompt. This keeps the product itself accurate while still letting you change the scene around it.
3. Generate a few variations before committing to one. The first result is rarely the best one.
4. If you're using Avatars for a talking promo, keep each script segment under the 5,000-character, 3-minute generation cap, and plan your script in those chunks from the start.
5. Check every generated image against your actual product for color, texture and proportion before publishing anywhere customers will compare it against the real item.
6. Where your prompt doesn't produce a reliable look across a whole set (say, 20 product variants), test on a few before committing to the full batch, and keep a note of exactly which prompt and reference image produced the version you liked, so you can reproduce that same look for the next batch instead of starting from scratch.

Image generation starts on ElevenLabs' free tier; video generation needs a Starter plan or above to unlock.

For a talking promo specifically (an avatar introducing a product or walking through a feature), script it in short, clearly separated beats rather than one long pitch: "here's the problem," then "here's what this does," then "here's how to get started" works better as three separate generations stitched together than one continuous 3-minute take, since it's easier to regenerate just the one beat that comes out wrong instead of redoing the whole clip. It also gives you reusable pieces: the same "here's what this does" beat can get swapped into different promos for different products without rebuilding the whole script from scratch each time.

Canva, Runway and Midjourney as alternatives

Canva is the lower-effort option if you mainly need templated graphics with your own real photos dropped in, not fully generated scenes; it's faster to learn and cheaper for straightforward layout work. Runway and Midjourney are the more flexible, more purpose-built options for fully generated imagery and video, with deeper creative control over style and composition than a tool built primarily around voice and speech, which is ElevenLabs' actual core strength. If your real need is photorealistic generated product scenes at scale, it's worth comparing results from a dedicated image/video generator directly against ElevenLabs' output before committing, since this isn't the part of ElevenLabs' platform with the longest track record. Wider tool picks are in best AI tools for ecommerce store owners.

The practical difference worth knowing before you pick one: Canva's generated-image features sit inside a design tool built around templates and real assets, so it's the better fit if most of a piece is your own photography and brand elements with AI filling in a gap, not generating the whole scene. Runway and Midjourney are built specifically for generation quality and give you more control over composition, style and consistency across a series, at the cost of a steeper learning curve and a tool that doesn't also handle your voice, narration or avatar needs the way ElevenLabs does in one place. If you're already paying for ElevenLabs for its voice tools, testing its Image & Video output first before adding a second subscription is the cheaper first move; switch only if the quality gap for your specific use case turns out to matter.

When should you skip this entirely?

Never let a generated image show your product as something it isn't; that's not a style choice, it's a trust problem that shows up later as returns and bad reviews. If your margin on returns can't absorb a few customers feeling misled by a visual that oversold the product, book the real shoot. A concrete test before you decide: pull up a generated image next to the real product side by side and ask whether you'd be comfortable if a customer did exactly that comparison after their order arrived. If the honest answer is no, that's your answer regardless of how good the image looks in isolation.

How I'd approach this

For a one-off promo push, generating visuals yourself is the right call and doesn't need my involvement at all. Where this becomes a real project is at catalog scale: hundreds of SKUs needing consistent promo visuals generated and checked automatically, or a pipeline that pulls product data straight from your store and produces on-brand variations without someone manually prompting each one. That's the kind of build I do, connected to your actual product data rather than a one-off tool someone has to run by hand every time.