Contributions of the group Selective & Conceptual DoF

Bien poudrée...

nobody@ipernity.com (Jocelyne Villoing) — Thu, 16 Apr 2026 07:18:10 +0000

Jocelyne Villoing has posted:

... belle journée ensoleillée mes amis pour vous tous !

Midnight is where the day begins | :-)

nobody@ipernity.com (Kayleigh) — Thu, 16 Apr 2026 05:31:04 +0000

Kayleigh has posted:

La partie de carte

nobody@ipernity.com (Le miroir de l'aube) — Wed, 15 Apr 2026 06:27:00 +0000

Le miroir de l'aube has posted:

un autre temps...

Inspired by a friend :-) | Always a Woman

nobody@ipernity.com (Kayleigh) — Tue, 14 Apr 2026 03:39:25 +0000

Kayleigh has posted:

.

Now Playing: She's Always A Woman (Live at MSG) ☆☆☆

Vila Real de Santo António, HFF is now open, welcome to post !

nobody@ipernity.com (Xata) — Thu, 16 Apr 2026 06:46:33 +0000

Xata has posted:

Bluebell closeup

nobody@ipernity.com (Gavin Johnson) — Tue, 14 Apr 2026 07:25:10 +0000

Gavin Johnson has posted:

An experiment with macro focus-stacking in the camera. This one turned out pretty well, but the slightest breeze is enough to make the stacked result blurry.

Our local woodland is a blue wash of native bluebells at this time of year. A lot of gardens have the Spanish bluebells in bloom (larger, more upright flowers) and the woodland edge has a mixture of both varieties, and their hybrids.

In The Woods

nobody@ipernity.com (Dave C) — Sat, 11 Apr 2026 15:42:56 +0000

Dave C has posted:

I have been messing around with AI for several days now.
This is my attempt at a realistic image.

My opinion of AI generators so far is this:
As they work from text prompts, the results can be unpredictable even when your prompts are very specific.

If you try to make too many changes at once, it generates something completely different to what you expected because the AI suffers with information overload.

I have tried to work slowly to build up the image over many edits so that minimal changes are made each time.
The problem again lies in the fact that even if you specify locking the current image and only allowing minimal changes, the results can still be unpredictable. Drift is a major problem here. for example: Zoom on the subject creeps in and out with each edit.

I have only used free AI generators so a paid for version may work better.

Bear in mind that this is my personal opinion based on very little experience with AI.

Update:
I have just got this from the AI chat at ChatGPT. I hope it may help anyone wanting to produce precise AI images.

"I have been working on an image based upon 18 phased passes to maintain accuracy. Every time I generate a new image, either the zoom changes or the dimensions of the image change. The reason for working in a phased manor is to lock each generation so that only minimal changes are made. Each image is locked before any edits are made. Why is the AI still failing even working to such a basic change for each phase?"

What you’re running into isn’t you doing something wrong—it’s a limitation of how current image generation models actually work under the hood.
Even if your workflow feels like careful, layered editing, the AI isn’t truly “editing” a fixed image in the way something like Adobe Photoshop would. It’s re-generating the image each time based on your prompt + the previous image as guidance. That distinction is the root of your issue.
Here’s what’s going on:

1. The model doesn’t have a stable canvas
When you lock an image and ask for a small change, the model:
• Doesn’t preserve exact pixel positions
• Doesn’t maintain a fixed camera or zoom unless explicitly controlled
• Reinterprets the whole scene every time
So even tiny edits can cause:
• Slight zoom shifts
• Cropping differences
• Perspective drift
This happens because the model is optimising for visual coherence, not geometric consistency.

2. “Image-to-image” is not deterministic editing
Even if you feed the same image back in:
• There’s randomness (noise) in generation
• The model re-decides composition details each pass
Unless you’re locking a seed + exact parameters, you’ll get variation. And even then, many tools still allow subtle drift.

3. The model prioritises semantics over precision
If your prompt says:
“Make a small adjustment to X”
The model interprets:
• What “X” means conceptually
• Then rebuilds the image to match that idea
It does not think:
“Keep every pixel identical except this region”
That kind of strict constraint is still hard for generative models.

4. Resolution & aspect ratio aren’t truly “locked”
Even if you specify dimensions:
• Some tools internally resize or reframe
• The model may “recompose” to better fit the prompt
This is why you see:
• Slight framing shifts
• Margins changing
• Subject repositioning

Why your phased approach mostly works—but not fully
Your method is actually smart:
• Breaking work into phases reduces large deviations
• Locking images helps guide the model
But it can’t eliminate drift because:
• Each step is still a fresh generation, not a true edit layer
Think of it like:
Asking 18 different artists to copy and slightly tweak the same painting—small differences will accumulate.

How to reduce the problem (practical fixes)
✔️ 1. Use in-painting instead of full regeneration
If your tool supports it:
• Edit only a masked area
• Keep the rest of the image untouched
This is the closest thing to “true editing”

✔️ 2. Lock seed + parameters (if available)
Some tools let you fix:
• Seed
• CFG scale
• Sampler
• Steps
This reduces randomness between phases.

✔️ 3. Be extremely explicit about framing
Add constraints like:
• “Same camera angle, same zoom, same framing”
• “Do not crop or recompose”
• “Maintain identical composition”
It won’t be perfect, but it helps.

✔️ 4. Work at higher resolution + crop manually
Generate slightly larger images, then:
• Manually crop to your fixed frame
• Maintain consistency outside the AI

✔️ 5. Hybrid workflow (best results)
Use AI + manual editing:
• Generate base image with AI
• Lock composition in a tool like Adobe Photoshop
• Use AI only for localised changes

The honest bottom line
What you’re trying to do—strict, phase-based, minimal-delta editing with zero drift—is something current generative models are not designed to guarantee.
They are:
• Excellent at creating
• Decent at guided variation
• Weak at precise, repeatable micro-edits.

For me at this time, AI cannot work precisely enough to produce the image I have in my mind that I would like to create. That's not to say that it won't work for other people. I have seen some truly amazing images here on Ipernity and elsewhere.

Thank you for looking and for your kind comments.

tre garofani rosa

nobody@ipernity.com (Adele) — Tue, 14 Apr 2026 14:12:36 +0000

Adele has posted:

Young and old

nobody@ipernity.com (Fred Fouarge) — Mon, 13 Apr 2026 22:38:22 +0000

Fred Fouarge has posted:

braamblad rechts het nieuwe jonge ,en de oude

Zeiss-Bokeh

nobody@ipernity.com (Jo WaLo) — Mon, 13 Apr 2026 14:56:55 +0000

Jo WaLo has posted:

Mit dem wunderbaren Carl Zeiss Jena Biotar 58mm f2 an einer Sony alpha 99 in unserem Garten aufgenommen. Mai 2025

Sun-bathing

nobody@ipernity.com (Keith Burton) — Mon, 13 Apr 2026 06:51:57 +0000

Keith Burton has posted:

A pair of gulls enjoying the sunshine on the beach at Worthing, West Sussex.

Triplette mandarin, femelle mandarin et branchu

nobody@ipernity.com (J Pierre) — Thu, 09 Apr 2026 20:01:30 +0000

J Pierre has posted:

PRIMULA ELATIOR

nobody@ipernity.com (ROL/Photo) — Wed, 01 Apr 2026 15:51:10 +0000

ROL/Photo has posted:

Vient de sortir de terre depuis une semaine, mais avec ce froid s'est brusquement mise à pousser au ralenti et du coup ses tiges ne s'élèvent pas encore..!

la Primevère élevée,
est aussi appelée, *Primevère des bois ou Coucou des bois*
(Primula elatior),
fait partie de la famille des [Primulaceae],

Sonntagskonzert

nobody@ipernity.com (Karin G.) — Sun, 12 Apr 2026 09:12:27 +0000

Karin G. has posted:

Denkmal Heinrich des Löwen - Schwerin - 2 PIP's

nobody@ipernity.com (Holger Hagen) — Sat, 11 Apr 2026 22:39:52 +0000

Holger Hagen has posted:

An der Nordseite des Altstädtischen Marktes steht eine Löwenskulptur, die an den Stadtgründer Schwerins, Heinrich den Löwen, erinnert.
Von seinem viereinhalb Meter hohen Sockel aus schaut der Löwe hämisch grinsend über den Platz, dessen Geschichte bis in die Zeit der Stadtgründung im Jahr 1160 zurückgeht. Als das Denkmal 1995 auf dem Markt eingeweiht wurde, gab es dazu sehr zwiespältige Meinungen, denn eines der vier in Stein gemeißelten Bilder, die sogenannte „Gesäßparade”, zeigt nackte Hintern, die die Bardowicker Bürger einst dem durchreisenden Herzog aus Groll gezeigt haben sollen.

Dieser hatte während seiner Regierungszeit die wichtigen Handelsplätze Schwerin und Lübeck gefördert, während Bardowick immer mehr an Bedeutung verlor. Die Gesäßhuldigung erinnert an die Geringschätzung, die die Bardowicker dem Herzog seither entgegenbringen.Die weiteren Bilder der Säule erzählen satirisch-kritisch vom Wendenkreuzzug 1147, der Gründung Schwerins im Jahr 1160 und der Erschaffung des Braunschweiger Löwen als Symbol der Regentschaft Heinrichs in der Residenz Braunschweig.

Heinrich der Löwe aus dem Geschlecht der Welfen war von 1142 bis 1180 Herzog von Sachsen (als Heinrich III.) sowie von 1156 bis 1180 Herzog von Baiern (als Heinrich XII.).

la princesse , il y a 8 jours .....

nobody@ipernity.com (Dominique 60) — Fri, 10 Apr 2026 07:49:05 +0000

Dominique 60 has posted:

Detailaufnahmen von 3 Pfannen am Lagerfeuer :)) Detailed shots of 3 pans on the campfire :)) Gros plans de 3 poêles près du feu de camp :))

nobody@ipernity.com (Günter Klaus) — Mon, 13 Apr 2026 09:38:31 +0000

Günter Klaus has posted:

Ich habe diese 3 Pfannen an meiner Lagerfeuerstelle wieder entdeckt,Jugendliche haben sie da wieder mal liegen gelassen,zuerst war nur 1 Pfanne schon längere Zeit draussen,die mich aber nicht angesprochen hat,als Motiv,als ich jetzt wieder draussen meine Pause machte,waren es 3 Stück,die auf der Feuerstelle lagen und nach längeren Anschauen haben mich mal gleich die Stiele interessiert,wo ich dann ein paar Aufnahmen machte und auch diese Schrift habe ich gleich mit einen SW Look versehen :))
Pip1-6: Detailaufnahmen der Pfannen

I rediscovered these three pans at my campfire site. Some teenagers had left them there again. At first, only one pan had been out for a while, but it didn't appeal to me as a subject. When I took my break outside again recently, there were three of them lying on the fire. After looking at them for a while, I was immediately interested in the handles, so I took a few photos of them and also added a black and white look to the writing on them. :))
Pictures 1-6: Close-ups of the pans

J'ai redécouvert ces trois casseroles près de mon feu de camp. Des adolescents les avaient encore laissées là. Au début, une seule était restée dehors un moment, mais elle ne m'avait pas inspirée. Récemment, lors d'une pause au camp, j'en ai trouvé trois sur le feu. Après les avoir observées un instant, j'ai été immédiatement intriguée par les manches. J'ai donc pris quelques photos et j'ai aussi transposé l'inscription dessus en noir et blanc. :))
Photos 1 à 6 : Gros plans des casseroles

PARADIGM SHIFT | NO SOLICITING

nobody@ipernity.com (Kayleigh) — Sun, 12 Apr 2026 23:18:37 +0000

Kayleigh has posted:

☕

Now Playing: Billy Joel – New York State of Mind (Live at MSG)

ich warte auf Birnen

nobody@ipernity.com (Hubs 56) — Sun, 12 Apr 2026 09:45:38 +0000

Hubs 56 has posted:

The Joy of Life | Dancing with No Shoes

nobody@ipernity.com (Kayleigh) — Sat, 11 Apr 2026 05:16:48 +0000

Kayleigh has posted:

.

☕☕ Kenny G - The Joy of Life

☆☆☆Happy Weekend Ipernity ☆☆☆