Step 1
Decide the line
Where your business draws it, written down, in language a freelancer can apply without ringing anyone.
AI-assisted workflows · AI image and video assist · Sydney
AI image and video assist
These tools are genuinely useful for the unglamorous half of production and genuinely dangerous for the half that carries your credibility. The line between the two is clearer than the debate suggests.
Grab the snowball. Give it a spin, and watch it grow as you scroll. Just like the craft.
2categories, and only one is safe
1line that does not move
Never peopleno synthetic customers or staff
Disclosedwhen it could mislead
The short answer
Where do generative tools genuinely help?
In production, generative and machine-learning tools are excellent at the mechanical work around real footage and photographs: removing a distraction, extending a background, upscaling, cleaning audio, rough-cutting long recordings, generating variations of a layout for testing.
They are dangerous the moment output could be mistaken for a record of something that happened. A synthetic customer, a fabricated product shot, a person who does not exist in a testimonial: these are not stylistic choices, they are claims about reality, and they carry legal as well as reputational risk.
So we draw the line at fabrication rather than at assistance, and we disclose where output could mislead. It sits inside our wider AI-assisted workflows service, and because our Snowball SEO platform automates the search heavy lifting other agencies bill by the hour, more of your budget goes into filming the real thing.
The difference
Where exactly is the line?
One question, asked of every use: could a reasonable person take this as a record of something that happened? If yes, it has to be real or clearly disclosed.
Where we use these tools
- Removing distractions and blemishes from real photographs
- Extending a background so a real image fits a different crop
- Upscaling, denoising and audio clean-up on real footage
- Rough assembly of long recordings before a person edits
Where we do not, at all
- Any person who does not exist, in any context
- Product images showing a product that was not photographed
- Results, before-and-afters or anything a buyer would rely on
- Faces or voices of real people, altered or generated
For the real thing, which is nearly always the better answer, see our photography and video production services.
The cost
What drives the price of assisted production?
Almost the same things as unassisted production. The tools compress specific tasks; they do not change what a shoot costs or what judgement is worth.
Assist on existing work
Per project
Clean-up, extension and versioning applied to footage and photographs you already have, which is where these tools earn most of their keep.
Assisted post-production
Per shoot
The same, built into the post workflow on work we produce, which is faster than treating it as a separate stage.
A policy and workflow
Setup
Where the line sits for your business, written down, with a disclosure position, which matters more in regulated categories than the tooling does.
We quote per project or per shoot after a free consult. The honest answer is often that filming or photographing the real thing is both better and cheaper than working around not having it.
What moves the number, in order
- How much source material exists to work from
- Whether the work is clean-up or reconstruction, which are very different
- How many versions and crops are needed
- Whether a disclosure position has to be developed for your category
- Regulatory constraints, which are tightening rather than loosening
- How much review the output needs before it can be published
The real thing is usually cheaper than working around it
A great deal of generative work in marketing exists to avoid a shoot. Once the reviewing, correcting, re-prompting and legal checking are counted, it frequently costs more than photographing the thing would have.
It also produces an asset you cannot fully stand behind, in a market where being able to stand behind your images is becoming a differentiator rather than a baseline.
So we will often recommend a half-day shoot instead, and say why. That is a smaller job for us in one respect and a better answer, and the recommendation is genuinely made that way round.
The scope
What you receive
Work you can publish without a caveat, and a written position on where the line is.
The assisted output
Clean-up, extension and versioning applied to real material, delivered at the sizes and formats your channels need.
A written policy
Where the line sits for your business, so the decision is made once rather than argued per asset by whoever is nearest.
A disclosure position
What you tell audiences and clients, decided deliberately rather than when a journalist or a regulator asks.
A record of what was done
Which assets were assisted and how, kept, because you will be asked eventually and reconstructing it later is impossible.
The workflow
Where in post these tools sit, so the assistance is a stage rather than something applied ad hoc by individuals.
A review date
Because the tools, the norms and the law are all moving, and a policy set once will be wrong within a year.
The return
What does this actually save?
Time on the mechanical work, which is real and worth having. Not the cost of having something true to show, which is the expensive part and is not compressible.
Provenance is becoming an asset
As synthetic imagery becomes ordinary, being able to say that a photograph is of a real customer in a real place stops being an assumption and starts being a claim worth making.
Brands that can demonstrate provenance will have something their competitors cannot cheaply copy, which is the same argument our authenticity work makes about written content.
So we keep a record of what was assisted and how, not only for compliance but because in two years it may be a selling point rather than a disclosure.
The process
How assisted production runs
A fortnight to set the policy, then it is simply part of post-production.
Step 2
Set the disclosure position
What you say, and where. Decided before somebody asks rather than during the conversation in which they ask.
Step 3
Build it into post
The assisted stages placed in the existing workflow rather than bolted on, so it is consistent rather than individual.
Step 4
Record what was done
Per asset, kept. Reconstructing this after the fact is effectively impossible and is exactly when it is wanted.
Step 5
Review before publishing
A person checks output against the policy, because the failure mode here is fast, plausible and wrong.
Step 6
Revisit
On a date, because the tools, the norms and the regulations are all moving faster than any policy written once.
The brief
What to ask before you use generative tools
The questions worth asking are about honesty and record-keeping, not about capability.
Where is your line?
It should be written down and specific. A supplier who treats this case by case will eventually make the call you would not have made, on a deadline.
Will any person be synthetic?
The answer should be no, unambiguously. Fabricated customers and staff are a claim about reality, and they are the fastest way to lose trust you cannot buy back.
What is recorded?
Which assets were assisted and how. You will be asked eventually, and reconstructing it afterwards is not possible.
Would the real thing be cheaper?
Frequently, once reviewing, correcting and checking are counted. Ask, and be wary of a supplier who never reaches that conclusion.
We are happy to answer all four, including recommending a real shoot instead, which we do often. See photography and video production.
Start here
Ready to draw the line properly?
Tell us what you are trying to make and why filming it is difficult. The free consult is a working session, not a sales call, and you leave with three things whether or not you book us.
- An honest answer on whether a real shoot would be better and cheaper
- A written line for your business, in language a freelancer can apply
- A fixed price, with the disclosure position included rather than assumed
Good questions
AI image and video FAQs
Where do generative tools genuinely help in production?
The mechanical work around real material: removing distractions, extending backgrounds so an image fits another crop, upscaling and denoising, cleaning audio, and rough-assembling long recordings before a person edits.
Would you generate an image of a person?
No. A person who does not exist, in any context that could read as a record of reality, is not a stylistic choice but a claim about the world. That includes customers, staff and anyone in a testimonial.
Do you disclose when tools were used?
Where output could mislead, yes, and we decide the position before anyone asks rather than during the conversation in which they do. We also keep a record per asset of what was assisted and how.
Does this save money?
On the mechanical stages, genuinely. Not on the expensive part, which is deciding what to make and having something real to show. Anyone promising a transformation in the cost of visual content is describing the small half.
Is it cheaper than a real shoot?
Often it is not, once reviewing, correcting, re-prompting and legal checking are counted. We recommend a real shoot frequently, and it produces an asset you can fully stand behind, which is becoming a differentiator.
What about product images?
If it could be taken as a record of the actual product, it must be a photograph of the actual product. Consumer law does not care how an image was made, only whether it misled somebody.
What are the legal risks?
Misleading representations, rights in training data and likeness rights are all live and all moving. We keep the line conservative and the records complete, which is the only defensible position while the law settles.
How often should the policy be reviewed?
At least annually, and sooner in a regulated category. The tools, the norms and the regulations are all moving faster than any document written once, and a stale policy is worse than none because it is trusted.