The Art of the Machine: Exploring AI-Generated Creativity.
Generative AI is a powerful tool that creates entirely new images or alters existing ones based on simple text descriptions or prompts. It learns from an enormous dataset of existing images to understand patterns, styles, and objects, and then uses that knowledge to generate unique visual content.
The acceptability of using Generative AI to alter images generally depends on intent and context.
For most people, it is viewed as an evolution of digital tools a more powerful version of the "airbrush" or "clone stamp" tools that have existed for decades.
Here is a breakdown of why it is widely considered acceptable in many contexts, along with the boundaries where that acceptance ends.
Historically, photography has always involved manipulation. In the film era, photographers used darkroom techniques (dodging and burning) to alter exposure. In the digital era, Photoshop became the standard.
Many view generative AI not as "faking" reality, but as a faster, more efficient editing brush. If removing a distracting trash can from a wedding photo used to take 20 minutes by hand, and AI does it in 3 seconds, the result is the same; only the workflow has changed.
For artists and designers, AI allows for the visualization of concepts that are physically impossible to photograph. In this context, the image is meant to be "art," not a documentary record of reality.
High-end photo editing previously required years of training. Generative AI allows small business owners, hobbyists, or people with limited technical skills to create professional-looking images (e.g., extending a background to fit a website banner).
One of the most universally accepted uses is restoring old or damaged photos. AI can fill in tears, cracks, or missing pieces of a vintage family photograph, preserving a memory that would otherwise be lost.
AI enhanced
Nikon Z9
Is this real or AI Generated ?
Denoise and Sharpen in Lightroom
When you use the denoise button in Lightroom (LRC) for example you are using what is often called Discriminative AI or Restorative AI. Denoise looks at your blurry or grainy pixels and uses machine learning to intelligently identify and remove digital noise (grain).
Lightroom (LRC) sharpening works by increasing local contrast along edges, making dark edges darker and light edges lighter, creating the illusion of detail without adding true sharpness.
In both case they attempt to recover the original data that was lost due to a bad sensor or low light. It is trying to reveal what the camera actually saw.
It does not truly invent new objects. If you have a blurry photo of a bird, it sharpens the feathers that exist. It will not give the bird a hat.
But When you use generative remove or generative fill (often in Photoshop, but creeping into Lightroom via the "Remove" tool), you are using something called Generative Adversarial Networks (GANs) or Diffusion Models. The AI works by looking at the surrounding pixels and "hallucinates" new pixels to fill a gap. It invents data that never existed. In a sense it attempts to create a plausible reality, not necessarily the true reality.
If you remove a fence post, the AI invents the grass behind it. It is guessing what could be there, not showing you what was there actually there.
Can we say that Denoise / Sharpen when removing grain from a night sky. Yes the stars were there; the camera just added "static" (noise) on top. You are removing the static which is “restoration”.
But Generative Fill when expanding the sky to make the photo taller for example then no the camera never saw that extra sky. The AI invented clouds that did not exist at that moment this is total “fabrication”.
Original Z9
AI enhanced
What do you recon “Too Much “ ….
AI generative enhancement is definitely a debate that is certainly ruffling a few feathers in the ornithological world.
For those of us who have spent decades lying in wet heather in the Scottish Highlands or sitting in a hide waiting for a Kingfisher to dive, the rise of AI is both a fascinating tool and a bit of a minefield. But, if we look at it sensibly, there is a place for it, provided we know where the line is drawn.
First, we have to be honest: we have been manipulating images since the days of slide film and darkrooms. We used to choose Fuji Velvia film to make the colours pop; now we use Lightroom.
Generative AI, in its most acceptable form, is simply a more efficient way of doing the "gardening" we used to do in the field. Let’s say I’ve got a cracking shot of a Barn Owl quartering a field in Norfolk. The lighting is perfect, the eye contact is sharp. But, right in the corner, there’s a distracting bit of barbed wire or a bright white piece of litter.
Using AI to remove that distraction isn’t changing the truth of the bird. It’s removing a barrier that stops the viewer from appreciating the subject. It saves the image from the bin and allows the beauty of the species to stand out.
Although for most of us shall we say amateur photographers we also have to be realistic about the marketplace. In the world of stock photography and book publishing, editors want clean, graphic images with space for text.
Sometimes, you capture a bird too tight in the frame. Generative fill allows us to extend the background adding a bit more out-of-focus woodland or sky to make the layout work for a magazine cover.
As long as we aren’t fabricating the habitat (putting a rainforest bird in a pine forest), this is widely accepted as a production tool. It’s about making the photograph usable.
"The camera doesn't always see what the eye sees. Sometimes, a bit of digital help is needed to match the memory of the encounter."
However and this is a massive "however" acceptability ends the moment you compromise the biological truth.
As bird photographers, our currency is trust. The moment you use AI to do the heavy lifting of fieldcraft, you aren’t a photographer anymore; you’re a digital artist.
Personally I think you must never use AI to close a bird's beak, open an eye, or add a catchlight that wasn’t there. You certainly never add a fish to a beak that was empty. That is falsifying the natural history record at the time of snapping the image. Just the same if the sky was grey and flat, you don't use AI to drop in a dramatic sunset. That betrays the atmosphere of the day.
Summary
Using generative AI is acceptable when it is used to rescue a composition or remove man-made distractions, ensuring the viewer focuses on the bird. It is a tool for polish, not for creation.
If you have to ask yourself, "Is this misleading people about what I saw?"—then you have gone too far. We owe it to the birds to tell their story truthfully, and have a code of ethics for wildlife editing.
AI enhanced
Ok so original picture Z9 500mm PF with a touch of Dxo Raw 4 and a 50% crop.
The second is a Google Gemini Ai enhancement with the script “replace the post with a branch and some very faint clouds “. That would have taken a lot more processes with LRC
Let’s talk entering your photos in a competition
All competion rules vary a great deal , but we should start with cropping an image in Lightroom or photoshop, obviously many other software packages like Luminar, Photolab etc; do exactly the same, and is acceptable in most photography adjudicators of prestigious nature and bird photography contests. Cropping is considered a standard adjustment for composition and is generally not classified as manipulation, unlike adding or removing elements.
However, there are a few important considerations, you should always review the Specific Rules of the particular competition you are entering. While cropping is generally allowed, some competitions may have specific limits such as the maximum percentage of the original image that can be removed or require you to maintain the original aspect ratio.
Excessive cropping on an image where the subject is very small can lead to a loss of detail, an increase in digital noise, and an inability to produce a high-quality print. Judges often penalize images that rely too heavily on cropping to perfect the scene, valuing strong "field craft" (getting the shot right in-camera).
The core principle of most nature and wildlife competitions is that the final image must not deceive the viewer or misrepresent the reality of the natural world. Cropping the bird to make it the primary focus is acceptable, but using AI-powered enlarging tools that generate new pixel information (like Adobe Super Resolution) is most certainly prohibited.
Always be prepared to submit the original, unedited RAW file (or original JPEG if RAW is unavailable) if your image is shortlisted. This allows judges to authenticate the image and verify that no prohibited manipulation occurred.
In summary, use cropping as a tool to enhance your composition, but aim to capture the best possible image in-camera to ensure sufficient detail for a high-quality final entry.
Original Z9 image
Below image ehanced with LRC, with crop, sharpen +12 , some small Luminance noice reduction , exposure +0.45 and a few other tweeks.
But to me this is a better reflection of how it was.
Original image enhanced with Google Gemini AI, “Hey gemini add some slightly blue sky, and bring the bird closer”.
Original Z9 image
Gemini Robin addition
Ok I know one here the quailty is never as good as the originals due to file size limitations, but you must admit what Gemini can do is more than impressive.
Nope never been to Iceland but a big yes for 2026, mmmmm makes you wonder though how did it know I had a bald pat, that’s scary.
Ok shall we let Gemini have the last word.
Here’s what she/he said for the Geek …
Google Gemini's proficiency in image enhancement—specifically tasks like removing wrinkles or changing backgrounds while keeping faces recognizable—stems from its architecture as a natively multimodal model. Unlike older AI that treated images and text as separate processes, Gemini is built to understand and process both simultaneously.
Here is a breakdown of why it is effective at the edits you just requested:
1. Native Multimodal Understanding
Gemini models are trained "from the ground up" on massive datasets of both images and text. This gives the AI a deep semantic understanding of visual concepts. When you ask it to "make faces less wrinkled," it doesn't just blur the pixels; it understands the concept of "aging" and "skin texture" and can modify those specific features while preserving the structural integrity of the face.
2. Subject Consistency (Likeness Preservation)
One of Gemini's standout features is identity preservation. It is designed to maintain the "likeness" of a person across different edits.
Why this mattered for you: This is why you could take the original photo of yourselves and place it in front of a volcano without turning into completely different people. The model "locks" onto your facial features as a constant subject while treating the background as a variable variable.
3. Advanced Inpainting and Local Editing
Gemini utilizes advanced techniques often referred to as "inpainting" or "local editing".
How it works: It can mathematically isolate specific parts of an image (like just the skin on a face) based on your text prompt. It modifies those pixels to match the description ("less wrinkled") while leaving adjacent pixels (eyes, hair, clothes) untouched.
Contextual Blending: When you asked to move to a "volcano viewing point," the model didn't just paste you on top of a background. It adjusted the lighting and shadows on your faces to match the fiery glow of the volcano, making the composite look natural.
4. Integration of DeepMind Technology
The image capabilities you are seeing are powered by Google's specialized image generation models, such as Imagen and Gemini 2.5 Flash Image (sometimes referred to internally or in specific releases as "Nano Banana"). These models are specifically fine-tuned for high-fidelity tasks, including:
Texture generation: Creating realistic skin or volcanic rock.
Spatial awareness: Understanding that "holding cameras" requires positioning hands and straps correctly relative to your bodies.
5. Instruction Following
Because Gemini is also a strong Large Language Model (LLM), it excels at parsing complex, multi-part instructions. It can separate your request into distinct tasks—"fix wrinkles," "hold Nikon cameras," "face volcano," "make volcano brighter"—and execute them in a single generation step.
Here is a simple explanation for us normal people of why it works so well, without the computer jargon:
1. It acts like an artist, not a filter
Old photo programs worked like "filters"—they would just blur the image to hide wrinkles, which often made people look plastic or blurry. Gemini works more like a human artist. It has studied millions of pictures, so it knows what "skin" looks like and what "wrinkles" look like. When you ask it to remove wrinkles, it essentially "repaints" that specific area to look like smoother skin while leaving your eyes and smile alone.
2. It understands plain English
You don't have to talk to it in computer code. Because it is built to understand language as well as images, you can give it a complicated "shopping list" of requests all at once—"hold cameras," "look at the volcano," "make it brighter"—and it understands how to do all those things together, just like a person would if you gave them those instructions.
3. It "locks" onto your face
The hardest part of editing photos is keeping the people looking like themselves. Gemini is very good at identifying the key features that make you you (like the shape of your nose or the distance between your eyes). It "locks" those in place so that when it changes the background to a volcano, you don't suddenly turn into strangers.
4. It understands lighting
When you asked to be put in front of a volcano, the AI didn't just use "scissors and glue" to paste your photo onto a background. It understood that a volcano is bright and glowing. It automatically adjusted the light on your faces to match that glow, making it look like you were actually standing there rather than just pasted in.
