How to Evaluate an AI Commercial Studio’s Portfolio Before Hiring

Every studio’s reel looks good, because the reel is the one asset every vendor optimises. That is exactly why it is a weak buying signal on its own: three completely different kinds of business now show you similar frames under one label, and the pricing pages look similar enough that buyers routinely sign the wrong one. This post is the framework you run against the portfolio in front of you — what to ask for, what a reel cannot prove, and which answers should end the conversation. If you want ranked names instead, we keep two lists: the top 10 AI video production companies for 2026 and a narrower honest comparison of the best AI video production companies. Read those for the names. Read this for how to judge what they show you.

What actually counts as an “AI video production company”?

Three distinct business models compete for the same search term, and they are not substitutes for each other. Getting this taxonomy right eliminates most bad hires before you take a single call.

Self-serve platforms sell you software. You write the script, pick a template or an avatar, and operate the pipeline yourself. The unit of work is a seat or a credit. Template and volume shops sell you throughput — a low per-asset price for output assembled from a fixed set of motion templates, stock beds and a generic voiceover. The unit of work is a deliverable count. Production studios sell you a directed film. The unit of work is a brief, and a named creative person is accountable for whether the finished spot works.

All three now use generative models. That is the trap: the toolchain is no longer the differentiator, so a tool screenshot proves nothing. What differs is who makes the decisions, and what happens to a shot that comes back wrong.

What separates a real AI commercial studio from a template shop?

Four things, and all four are observable in a portfolio before you sign. Everything else — model names, render counts, turnaround boasts — is noise layered on top of these. The broader studio-selection question is covered separately in our guide to choosing the best AI video studio; what follows narrows it to what a reel and a portfolio can actually prove about the studio behind them.

1. Who makes the creative decisions?

A commercial is a sequence of choices: what the first frame withholds, where the cut lands relative to the line, whether the product appears at second three or second eleven. A model makes none of those choices. It executes the choice it was given, plausibly and fast.

Ask a prospective vendor who holds those decisions and what happens when you disagree with one. A studio names a person — at ArcaneWiz, that is Creative Director Daniel Atzil — and can defend a decision in the language of intent. A template shop will redirect you to a revision policy, because there is nobody in the loop whose job is the decision itself.

2. Is the shot designed, or is it prompted?

This is the single most reliable tell. Prompt-led production writes a description, generates, and picks the best of a batch. Designed production settles the frame first — lens feel, blocking, light direction, what the camera does and why — and only then generates against that target. The difference is invisible in a single hero frame and glaring across a 30-second cut, because prompt-led shots do not match each other.

Shot inconsistency across a sequence is the known hard problem in generative video, and it is exactly what practitioners flag about the current model generation. Google DeepMind’s own Veo documentation describes the model in terms of prompt fidelity and cinematic control per generation — per generation being the operative phrase. Continuity across shots is production work, not a model feature.

3. Where do the assets come from, and are they licensing-safe?

Commercials run in paid media, on broadcast, and in front of legal teams. That raises questions a creator-tier workflow never has to answer: what went into the frame, what is disclosed, and what a platform will do when it detects synthetic content.

Three things a serious provider should already have a position on. First, provenance — the C2PA standard defines how content credentials travel with a file, and it is increasingly how the supply chain proves what an asset is. Second, platform disclosure — YouTube requires creators to disclose realistic altered or synthetic content at upload, and that obligation lands on whoever publishes the spot, which is usually you. Third, regulatory disclosure — Article 50 of the EU AI Act sets transparency obligations for AI-generated or manipulated content, so if your media plan touches the EU, a vendor shrug is a real liability.

There is a fourth, older rule that catches people out: the FTC’s endorsement guidance does not care whether a testimonial was rendered or filmed. A synthetic person praising your product is still an endorsement claim. This is one reason ArcaneWiz does not offer likeness cloning of real people — the compliance surface is larger than the creative upside, and it is not a service we sell at any price.

4. Is it finished, or is it exported?

Generated footage is raw material. What makes a spot look bought rather than made is the layer after generation: editorial rhythm, a real grade, sound design and mix, titles built to brand spec, and masters cut to the technical requirements of every placement — broadcast, 16:9 pre-roll, 9:16 social, 1:1 feed, each with its own safe areas and loudness target.

Template shops skip most of this and hand you one master with auto-crops. That is why the output reads as generic even when individual frames are good. Ask to see the deliverables list from a previous project. It is a more honest document than the showreel. What that finishing layer actually involves — and how a spot reaches broadcast standard without a crew on a stage — is the subject of broadcast-grade AI video ads without a film crew.

How do you read an AI commercial demo reel?

Read the reel against the four axes below rather than against your own taste, and score every shortlisted studio before you compare prices. A portfolio is direct evidence for only one of these axes; the other three you have to ask for. A provider that scores low on two or more is not cheaper than the alternative — it is a different product.

How does a real AI commercial actually get made?

Here is the sequence a studio runs, so you can compare it against whatever a vendor describes on the call. If steps are missing, the price difference has an explanation.

1. Brief and strategy lockWhat the spot has to do, for whom, where it runs, and what success looks like. Nothing generates until this is signed off, because every later decision is measured against it.
2. Concept and scriptTwo or three distinct creative routes, not variations on one. You pick a direction before anyone spends a render credit.
3. Storyboard and shot designEvery shot drawn or framed: composition, lens feel, light direction, camera move, cut point. This is the document the whole production is generated against.
4. Art direction and frame developmentMidjourney work to establish the visual world — palette, materials, the look of the product in the frame — so generated shots share a single visual language instead of drifting.
5. Motion generationVeo, Kling and Seedance used shot by shot against the board, with selects reviewed against the storyboard rather than against each other. Shots that miss get re-designed, not just re-rolled.
6. EditorialThe cut assembled to rhythm and intent, with music and pacing decisions made in the timeline, not assumed in the prompt.
7. Grade, sound design and mixA real colour pass for consistency and mood, sound design built to the cut, and a mix that survives a phone speaker and a broadcast chain.
8. Delivery and spec mastersMasters cut per placement with correct aspect ratios, safe areas and loudness, plus the disclosure position agreed for each platform.

What should you ask before you sign?

Ten questions. Use them on every vendor including us. The answers separate categories faster than any showreel.

• Who is the named creative director on my project, and will they be on the calls?
• Can I see a storyboard from a delivered project alongside the final cut?
• What happens when I reject a shot — do you re-design it or re-generate it?
• What is in your deliverables list beyond the hero master?
• Which generative tools are in the pipeline, and which steps are human?
• How do you handle platform disclosure for synthetic content, and who files it?
• What is your position on provenance and content credentials?
• How are usage rights defined for my project, and where is that written?
• What is the realistic timeline from approved brief to delivered master?
• Show me work that ran as an actual campaign, not a spec reel.

What are the red flags in an AI commercial demo reel?

What do AI commercial production services cost, and how long do they take?

Pricing across this market spans two orders of magnitude, which is only confusing until you map it to the three categories above. Self-serve platforms charge a monthly software fee and your team supplies the labour. Template shops price per deliverable, low, and the low price is the product. Studios scope per project, because a 20-second product spot and a 90-second brand film are not the same amount of work.

At ArcaneWiz, commercial projects are scoped to the brief rather than to a deliverable count, and we do not discount by volume. Our entry point for a commercial is From $1,500, and what sits at it is a scope conversation rather than a fixed package. Standard delivery is 7–10 working days from an approved brief to a delivered master — the clock starts at approval, not at first contact, because a brief that is still moving is the most common cause of a late spot. Israeli clients can be invoiced in ILS with a חשבונית מס.

The comparison that matters is not price per video. It is price per usable asset. A cheap spot you cannot run in paid media, cannot clear with legal, or cannot cut to four placements has a much worse effective rate than it appears to. Our AI commercial production guide breaks the scope questions down further, the AI video ads page covers the placement-specific deliverables side, and the pricing page sets out package scope.

What are people actually saying about AI commercial production?

The public discussion is more useful than the vendor marketing, and it converges on the same conclusion this framework does: the models are good, and the craft gap is where projects fail.

On the Hacker News thread for Google’s Veo 3, Imagen 4 and the Flow filmmaking tool, commenters who work in production were direct about the limitation. Individual clips were rated strong in isolation, but stitching them into a coherent piece was described as the unsolved part — one commenter noting that the clips are very good “in themselves” but putting them together “will not be possible for a long time,” and another observing that consistency of quality across scenes is still the problem. Several framed the tooling as most useful as a live storyboard, a communication device rather than an autonomous production replacement. That is close to how it is used in a studio pipeline.

A more recent thread, “AI-generated posters don’t have to be horrible,” argued about the adjacent craft problem and landed somewhere sharper. The recurring criticism of AI-generated design there was not fidelity but clutter and lack of editorial restraint — the observation that generative output tries to visualise every element in the brief instead of deciding what to leave out, and that professionals are paid largely because they are “extremely good editors.” Swap posters for commercials and the diagnosis is identical: the missing ingredient is subtraction, and subtraction is a direction job.

On the effectiveness side, the MIT Initiative on the Digital Economy published research on AI-generated personalised video ads covering more than 21,000 consumers across three organisations. AI-generated personalised video outperformed personalised image ads and generic video on click-through. The authors’ own caveat is the interesting part: they flag that the lift may partly reflect novelty, warn about quality deterioration at scale, and conclude that human judgment should guide AI ad creation rather than full automation. That is the buying argument for a studio, made by researchers with no stake in it.

Where does ArcaneWiz fit — and where doesn’t it?

ArcaneWiz is a boutique AI commercial studio. We take a brief, direct it, generate it through a named toolchain, and finish it to broadcast standard. Clients include Samsung Israel, Anipet, Fun Forest and Homey Panda. Work is on the portfolio, and the service scope sits on the AI commercial page. How the studio model reads from the brand side is covered in our piece on AI filmmaking studios for premium brands.

We are a poor fit for some real, legitimate jobs. If you need 200 localised variants of one creative by Friday, a platform will serve you better and we will say so. If you have an in-house creative director and just want render capacity, you are buying a tool, not a studio. If the deliverable is internal training or documentation video, an avatar platform is genuinely the right product. And if the concept depends on a real person’s face and voice, that is a traditional shoot — we do not clone likeness. If you are an agency evaluating production capacity to run under your own brand rather than buying a single spot, that is a white-label AI video production partner conversation, and it evaluates differently.

Where we are the right call: a brand-facing commercial that has to carry weight, where somebody needs to be accountable for whether it works. If you want a structured way to compare us against the field, the five-axis buyer’s guide to choosing an AI video production agency applies the same discipline across vendors.