Participant

Question

New contributor, AI still lifes rejected for "Quality issues" — what are reviewers seeing that I'm not?

14 replies
176 views
Translate

Hi everyone,

I'm a new Adobe Stock contributor. I submit AI-generated, photorealistic still-life and lifestyle images (no people). So far: 21 submitted, 3 accepted, 13 rejected — and almost all of the rejections say "Quality issues." Every file is submitted with the "Created using generative AI tools" option ticked.

I check every image at 100% before submitting and discard anything with obvious AI mistakes. Even so, a whole batch that looked clean to me was rejected, so I clearly have a blind spot. I've done my own analysis since then, but a few things I still can't explain, and I'd really appreciate experienced eyes on them.

My pipeline, to be transparent: the images are generated with Meta AI, the image-generation feature built into Meta's AI assistant. Its native output is about 2.4 megapixels (2048x1152). I upscale to 3072x1728 with Lanczos resampling — a standard resizer, not an AI upscaler — save as JPEG, and submit. Nothing is composited or retouched by hand.

The examples (I'm attaching 100% crops and reduced-size full views; I'm not attaching the full-resolution originals):

- Rejected A — "Sustainable kitchen" (glass jars of penne, rice and beans): at 100% I cannot find any broken or melted objects, yet it was rejected for quality. At 400%, the textures do look thin and smooth. Is that, by itself, enough to fail?

- Rejected B — "Winter home office desk": the knit blanket's texture falls apart into worm-like blobs at 100% (crop attached). I missed this completely when I submitted — I had only checked the fit-to-screen view. Lesson learned.

- Accepted, for contrast: a study-desk flat lay, and a calm morning still life with a cup of tea (small, localised wisp of steam). Both went through exactly the same generation + upscale pipeline as the rejected ones.

- Also attached: a 100% crop of a rejected hot-pot image whose steam spreads across the upper frame, almost like smoke over the background (relevant to question 3).

My questions:

1. For an image like Rejected A — shapes intact at 100%, textures thin at 300-400% — what is the decisive "quality issue" a reviewer sees? Is fine-texture thinness alone a fail?

2. Is upscaling from ~2.4 MP to 3072x1728 itself a disadvantage in review, or does only the visible result matter? In other words, would generating natively at 4 MP+ actually change outcomes, assuming the same subjects?

3. Are steam and large out-of-focus areas inherently risky in AI images? The accepted morning still life has a small steam wisp over the cup, while the rejected hot-pot image had steam spreading across the upper part of the frame. Is it the amount and placement that matters, or is steam simply something reviewers distrust in AI content?

4. What should a pre-submission check look like from a reviewer's perspective? I now inspect at 100% across the whole frame — at what zoom do you check your own files, and what do you look for beyond "no broken shapes"?

I'm asking because I want to fix my process, not to argue with the rejections. Thank you in advance for any specific pointers.

E

epic_ease85e2Author

Participant

Thank you both. I can see the dark spots on the paper now — I had missed them at 100-200%. I will check light paper areas for dark speckles before submitting watercolor-style work. The print-reproduction point is also noted.

Translate

Nancy OShea

Community Expert

This might work for digital projects but it won’t reproduce well enough for use on printed merchandise.

Translate

Nancy O'Shea— Product User & Community Expert

E

epic_ease85e2Author

Participant

Thank you again for all the advice on my earlier post. An update and a new question.

Since then I have adopted pre-submission inspection at 100% across the full frame, plus 200–300% checks on structural details (bottle threads, lids, cut surfaces) and on the edges of small repeated elements. Five images have been accepted so far, and the structural-defect refusals have clearly decreased. Several later refusals had causes I could identify myself at 200%, so those needed no questions.

One recent case I cannot explain: a watercolor-style illustration (plum blossom branch with blank New Year greeting cards), submitted as an illustration with the AI disclosure. It was refused for "quality issues". I re-inspected the submitted file at 100–200% and found no structural errors, no fused shapes and no garbled details. The image is, however, uniformly soft and low in contrast, as watercolor tends to be.

My questions:

1. For an intentionally painterly illustration, can overall softness / thin texture by itself be a refusal reason, even without any local defect?

2. My earlier question is still open: if texture looks thin only at 300–400%, can that alone fail a review? And is an image upscaled from roughly 2.4MP judged more strictly than one starting at a higher native resolution?

I have attached the full image and a 100% crop. Any pointers on what a reviewer may have seen here would be very welcome.

Translate

daniellei4510

Community Expert

You might try cleaning up the dark spots while still trying to maintain a subtle dibbled look of watercolor paper. Which would probably best be done by hand.

Translate

Adobe Community Expert | I have miles to go before I sleep, but I'm all done on this end.

E

epic_ease85e2Author

Participant

Thank you all — this is extremely helpful, and very specific.

Nancy and daniellei4510: point taken on the knit. I treated it as a background texture, but in that image it sits front and center, and the stitches simply do not hold up. I will stop using knits and other hand-crafted textures as main subjects, and I am adding a 200-300% pass to my pre-submission check, as daniellei4510 described.

oleschwander: yes — the study-desk flat lay was accepted, as was the morning still life. That is honestly part of what confused me: the accepted desk has the pencil and book-cover issues you spotted, while other images failed. It tells me acceptance has a borderline zone, and I should aim well above it rather than at it.

If anyone is willing to add a word on the two questions still open — whether upscaling from about 2.4 MP itself counts against an image, and how steam is judged — I would be grateful. Either way, thank you for the time you took annotating these.

Translate

daniellei4510

Community Expert

I upscale from approximately 38x21 @ 72ppi to 20x11.5 @ 300ppi. I used to do 30x and 24x on the long side but doing so only increased the time I spent editing. I’ll still do an occasion 24x if the subject warrants and the original image was near perfect to begin with. I use Gigapixel AI (the version prior to when they started charging for a monthly subscription). As for steam, I usually just get rid of it entirely. A “steaming hot cup of coffee” is more a marketing gimmick than what I usually come across in real life, unless you set your thermostat to below freezing. :)

Translate

Adobe Community Expert | I have miles to go before I sleep, but I'm all done on this end.

Nancy OShea

Community Expert

The focus is really poor and the knit stitches are not uniformly rendered.

Most AI models struggle with hand-crafted items. You can sometimes get away with it when it’s not the main subject. But when it’s front & center, as this knit is, it should be nearly perfect. Discard & try again.

Translate

Nancy O'Shea— Product User & Community Expert

daniellei4510

Community Expert

Another example.

Translate

Adobe Community Expert | I have miles to go before I sleep, but I'm all done on this end.

daniellei4510

Community Expert

You need to squeeze out more detail from many of the subjects in each image. I review my AI at 200-300% before submitting. My workflow is available here: https://gotosleepdancing.wordpress.com/2026/07/14/romancing-gemini-3/

Translate

Adobe Community Expert | I have miles to go before I sleep, but I'm all done on this end.

O

oleschwander

Legend

There is something odd behind the rice jar above the rice. 2. There’s something wrong with the window in the upper left corner and something odd behind the fabric’s right side. 3. The are far too many pencils on the table. 4. The book cover is not even in both sides - was this accepted?

Translate

yamato713108855

Community Expert

Images may become distorted when enlarged.
Please zoom in on the images to check them.

Sharpening both ends of a pencil is dangerous.
You could injure your eyes if you happen to doze off.

Translate

E

epic_ease85e2Author

Participant

Thank you — the circled penne tip is exactly the kind of thing I missed. At fit-to-screen, and even on a quick 100% pass, it looked fine to me; the distortion at the edges only shows when you zoom right into the details. I will add a zoomed check of the edges of small repeated objects (pasta, grains, jar contents) to my pre-submission routine.

If anyone has thoughts on my other questions — whether the upscaling itself matters, and how steam is judged — I would still be grateful to hear them.

Translate

daniellei4510

Community Expert

Here’s another example of why I review my images at 200-300%, followed up with retouching and re-generating as needed. A review at 100% when it comes to photography is generally adequate, but AI makes too many mistakes or lacks sufficient detail to review at its finished size alone. The top image below is an image that looks fine at 100% or smaller. Below that is a section of the image viewed at 300%, with the original on the left and the edited and re-generated version on the right.

Translate

Adobe Community Expert | I have miles to go before I sleep, but I'm all done on this end.

Sign up

To post, reply, or follow discussions, please sign in with your Adobe ID.

Sign in to Adobe Community

To post, reply, or follow discussions, please sign in with your Adobe ID.