Forum
Consistency and Credentials: The Two Halves of Image Trust

In the space of a week, Google released two things that look unrelated. On 20 August it launched the Pixel 10 phones, which attach Content Credentials to photos taken with the Pixel Camera app. On 26 August it introduced Gemini 2.5 Flash Image, the image model that had circulated in public testing under the nickname “nano-banana”.
One is about making images more flexible. The other is about making them more accountable. Seen together, they describe the two halves of what visual work now needs.
The consistency problem, mostly solved
Anyone who has tried to build a campaign, a comic or a product catalog with image generators knows the frustration. The first image is lovely. The second shows a slightly different face, a different logo placement, a jacket with an extra pocket. Generators have been good at single pictures and poor at series.
Gemini 2.5 Flash Image is aimed squarely at that problem. Google says it can keep a character or object consistent across prompts and edits, place the same subject in new settings, show a product from different angles and make targeted changes described in plain language.
For designers, consistency is not a nice-to-have. It is the basis of identity. A brand is a set of things that stay the same so that other things can change. A tool that holds a character, a product or a visual language steady across many outputs makes generative AI useful for systems, not only for single images.
The provenance problem, being built in
The Pixel 10 announcement addresses the opposite concern. Photos taken with the Pixel Camera app carry Content Credentials, the C2PA standard that records how an image was created and edited. Google applies them to unedited photos as well as images modified with AI tools, and says its camera app reached the highest assurance level currently defined by the C2PA’s conformance program. The credentials can be viewed in apps that support the standard, including Google Photos.
That is an important design choice. Labeling only AI-made images creates a world where the absence of a label means nothing. Signing ordinary photographs as well gives authentic images something positive to show.
Why these belong together
The same week that made it easier to place a consistent character in any scene also made it easier to show where a photograph came from. That pairing is not a contradiction. It is the shape of a healthy visual culture: powerful tools for making, and reliable records of what was made.
For practitioners, it suggests a simple discipline.
- Use consistency for systems you own. Characters, mascots and product shots that belong to you or your client are the natural home for these tools.
- Be wary of consistency with people who did not consent. A model that keeps a face stable across edits is also a model that can put a real person into places they never were.
- Preserve credentials when you can. Export settings, platforms and editing chains can strip metadata. Check where yours are lost.
- Write down the chain. Even where tools do not record it, a short note of what was captured, generated and edited protects the work.
The view from the jury table
Award juries now see entries that blend photography, generation and editing in a single image. Questions on the International AI Award Rules page, such as which model and version were used and what ethical considerations guided the work, are the human-readable counterpart of Content Credentials. One records the chain technically; the other explains the intent behind it.
What to watch next
Consistency will spread quickly across image tools, because it solves a real problem. Credentials will spread more slowly, because they depend on cameras, software and platforms all cooperating. The gap between the two is where most confusion will happen in the coming months. Designers who close that gap in their own workflows, making with confidence and recording with care, will be the ones whose images are easiest to trust.



