Test log
Deep Dream Generator test data
Raw results for the Deep Dream Generator review. The verdict is in the review.
- Tested
- 6 November 2026
- Tester
- Miles Foster
- Account
- Free demo account, signed up by RankDude, nothing paid to the brand
- Tests
- 10
- Screenshots
- 6
- Method
- Published methodology
- Environment
- Google Chrome on desktop
Results
| ID | Test | Key measurement | Score |
|---|---|---|---|
| T01 | Style Range | Styles run: 0 of 10 | n/a |
| T02 | Character & Subject Consistency | Images run: 0 of 12 | n/a |
| T03 | Editing, Inpainting & Image-to-Image | Edits run: 0 of 12 | n/a |
| T04 | Resolution & Upscaling | Native size: 1,024 px (1.05 MP) | n/a |
| T05 | Prompt Adherence | Elements correct: 135 of 139 | 5/5 |
| T06 | Photorealism & Anatomy | Images that pass: 10 of 12 | 4/5 |
| T07 | Text in Images | Words correct: 58 of 61 | 4/5 |
| T08 | Speed per Image | Median per image: 21.1 s | 3/5 |
| T09 | Usage Rights, Watermarks & Privacy | Points met: 4 of 7 checked | 3/5 |
| T10 | Pricing & Value | Free images on my account: 29 | 3/5 |
| Overall (in the review) | 3.9/5 | ||
Style Range
n/aNot Tested- Method
- Ten style prompts for one lighthouse.
- Styles run
- 0 of 10
- Remark
- Not graded: no style prompt was run.
Raw notes
Ten style prompts for one lighthouse.
Not run: ten images would have cost 20 energy and the account had 12 left after the other tests, below my 85 percent stop.
- Run on the next review of this tool.
Character & Subject Consistency
n/aNot Tested- Method
- A described woman and a described thermos in six scenes each.
- Images run
- 0 of 12
- Remark
- Not graded: no consistency prompt was run.
Raw notes
A described woman and a described thermos in six scenes each.
Not run: twelve images would have cost 24 energy, more than was left, and Krea 2 Turbo has no reference image feature.
- Run on a plan or a model with a reference feature.
Editing, Inpainting & Image-to-Image
n/aNot Tested- Method
- Twelve fixed edits.
- Edits run
- 0 of 12
- Remark
- Not graded: no edit was run.
Raw notes
Twelve fixed edits.
Not run: with Krea 2 Turbo selected the app says "Switch to a model that supports image editing to use this option." and the model comparison shows Edit Image as N/A.
- Run on a model that edits, with energy to spare.
Resolution & Upscaling
n/aNot Tested- Method
- Every saved image was measured, and the price of the 4K switch and of the upscaler was read in the app.
- Native size
- 1,024 px (1.05 MP)
- Upscale price
- 10 energy
- Remark
- Not graded: the scale needs an upscale checked at 100 percent, and none was run. Native size alone (1 megapixel or more) would not separate a 3 from a 4 or 5.
Raw notes
Every saved image was measured, and the price of the 4K switch and of the upscaler was read in the app.
All 24 graded images came out at 1,024 by 1,024 pixels (1.05 megapixels) as JPEG files of 102 to 275 KB. The five images made at the 4:3 frame were 1,280 by 960 pixels. The app offers five frames: 1:1, 4:3, 3:4, 16:9 and 9:16.
The 4K image switch changed the price from 2 to 9 energy. The upscaler page showed "Remaining upscales this month: 30" and a price of 10 energy for a 2K upscale, so it is open on the free plan. I did not run one: 10 energy would have taken use to 97 percent of the allowance.
- Wide and tall sizes were not read; no upscale was run.
Prompt Adherence
5/5Tested- Method
- 24 of the 40 fixed prompts (6 portraits, 6 product shots, 6 set-wording images, 6 busy scenes), typed in by script on the Generate Image page with Krea 2 Turbo on a 1:1 frame, first image kept, every written element counted at full size.
- Elements correct
- 135 of 139
- Busy scenes
- 42 of 42
- Remark
- Scale 5: 90 percent or more of elements correct. The sample was 24 prompts and 139 elements, not 40 prompts and about 240; the ten style prompts and six character scenes were not run.
Raw notes
24 of the 40 fixed prompts (6 portraits, 6 product shots, 6 set-wording images, 6 busy scenes), typed in by script on the Generate Image page with Krea 2 Turbo on a 1:1 frame, first image kept, every written element counted at full size.
135 of 139 elements were right (97.1 percent). Portraits 34 of 35, products 32 of 32, set-wording images 27 of 30, busy scenes 42 of 42.
Misses: in the friends at the bus stop only one wore the headphones; the honey label, the fridge note and the street sign each lost one word of the set wording ("Unfiltered", "refrigerator", "No").
Added content that changed the scene: invented lettering on the watch dial and the bus stop poster, small invented notices in the coffee shop window and on the bus stop sign.
- 24 of 40 prompts, one image each. The five images made at the 4:3 frame by mistake were not graded.
Photorealism & Anatomy
4/5Tested- Method
- The six portrait and six product prompts, one image each (12 images), inspected at 100 percent for faults that give a generated image away.
- Images that pass
- 10 of 12
- Hand faults
- 0
- Remark
- Scale 4: 18 to 21 of 24 pass. The sample is 12 images, so 10 of 12 (83 percent) stands for 20 of 24.
Raw notes
The six portrait and six product prompts, one image each (12 images), inspected at 100 percent for faults that give a generated image away.
10 of 12 passed. Passed: the man on the bench, the laughing woman, the girl with the balloon, the chef, the grey-haired woman, the trainers, the pancakes, the olive oil, the pizza and the perfume bottle, with no fault at full size.
Failed (2): the watch dial carries lettering that reads nothing, and the bus stop poster behind the two friends carries a made-up headline.
Hand faults 0. Face faults 0. Faults in text-like marks on objects 2.
- Twelve images in one run, not 24 images in two runs.
Text in Images
4/5Tested- Method
- The six set-wording prompts: the coffee shop sign and board (11 words), the honey jar label (8), the night market poster (13), the book cover (8), the fridge note (10) and the street signs (11), 61 words in all.
- Words correct
- 58 of 61
- Images fully correct
- 2 of 6
- Remark
- Scale 4: 85 to 94 percent of words correct, or four images fully correct. The word share (95.1 percent) reaches the top mark, but only two images were fully correct, so the top mark's second condition is not met.
Raw notes
The six set-wording prompts: the coffee shop sign and board (11 words), the honey jar label (8), the night market poster (13), the book cover (8), the fridge note (10) and the street signs (11), 61 words in all.
58 of 61 words were spelled right (95.1 percent). Two of six images were fully correct: the night market poster and the book cover.
The honey label reads "Unfllltered" and the fridge note "refrigtrater". The white parking sign shows a no-parking symbol and the words "Parking" and "8 am to 6 pm", with no word No.
The coffee shop window and the bus stop sign carried small invented notices that read nothing, so those two images are not fully correct either.
- All six set-wording prompts, one image each.
Speed per Image
3/5Tested- Method
- Every generation in the prompt adherence test was timed from pressing Generate to the finished full-size image on the page.
- Median per image
- 21.1 s
- Slowest image
- 38.1 s
- Remark
- Scale 3: a median of 20 to 39.9 seconds.
Raw notes
Every generation in the prompt adherence test was timed from pressing Generate to the finished full-size image on the page.
Median 21.1 seconds, fastest 19.6, slowest 38.1 across 24 prompt runs. Five images took more than 24 seconds and three took more than 35. My timer checks every half second, so each figure is within that.
No generation failed. The five images made at the wrong frame took a similar time and are not counted.
- 24 runs, not 40. One model, Krea 2 Turbo; slower models were not timed.
Usage Rights, Watermarks & Privacy
3/5Tested- Method
- The terms of service and privacy page (read 6 November 2026), the plans page and the pricing questions were checked against the ten-point list, plus the product's own Download button.
- Points met
- 4 of 7 checked
- Remark
- Scale 3: 6 or 7 of 10. 4 of 7 checked is 57 percent, so about 6 of 10. The terms allow selling without a plan condition, and the pricing questions list commercial use rights as a paid feature; I counted the terms.
Raw notes
The terms of service and privacy page (read 6 November 2026), the plans page and the pricing questions were checked against the ten-point list, plus the product's own Download button. Seven points could be checked.
Met (4): the terms say who owns the output ("you retain any rights you have in content you create"); the terms say you may "use, reproduce, modify, distribute, and sell" it, with no free or paid condition; the pricing questions list commercial use rights under paid plans; the downloaded file was identical to the one on the page, 1,024 pixels square, with no watermark.
Not met (3): the terms license your content to Deep Dream Generator for "operating, improving, promoting, and maintaining the Service" and offer no opt-out from training; no source of training data is named; none of the 25 files (the 24 saved from the page and the download) holds a content credentials label.
Not tested (3): a watermark on paid downloads, whether generations are private by default (the privacy page says only that uploaded pictures are not visible by default), and whether real people and brands are refused.
- Terms read by me, not a lawyer. No paid plan was bought.
Pricing & Value
3/5Tested- Method
- The free energy was used until I reached 83 percent, and the plans page, the energy packs and the pricing questions were read in full.
- Free images on my account
- 29
- Cost per image on Basic
- about $0.024 to $0.14
- Remark
- Scale 3: $0.05 to $0.099 an image. The price depends on the model and the day's discount, from $0.024 on Basic at the discounted price to $0.14 at the listed price, and the cheapest pay-as-you-go price is $0.066; with free energy for about 10 pictures at once, I scored the middle.
Raw notes
The free energy was used until I reached 83 percent, and the plans page, the energy packs and the pricing questions were read in full.
Free: the account started with 70 energy (20 that refill at 1 an hour up to 20, and 50 that never expire). Krea 2 Turbo cost 2 energy on the day, listed at 12 with 80 percent off, so 70 energy paid for 29 images.
The cheapest paid plan is Basic at $9 a month for 750 energy, which is about $0.012 an energy. A Krea 2 Turbo image is about $0.024 at 2 energy and about $0.14 at the listed 12. A one-off pack of 120 energy is $3.99, which is about $0.066 at 2 energy.
Basic and above add unlimited Z-Image Turbo, the PRO models, the upscaler without limits and commercial use rights. Monthly plan energy resets on each billing date; one-off packs never expire.
- No plan bought; per-image costs come from the plans page and the energy table. The 80 percent discount may end.
Coverage
Tested
- 24 of the 40 fixed prompts, first image kept: 6 portraits, 6 product shots, 6 set-wording images, 6 busy scenes
- Time from pressing Generate to the finished image, and the energy cost, for every prompt
- The size of every saved image, and the price of the 4K switch and the upscaler
- The product's own Download button, compared with the file saved from the page
- Terms of service, privacy page, plans page, energy packs and the pricing questions read in full
Not tested or limited
- The account's 70 energy paid for 29 images at 2 energy each; I stopped at 12 energy left (83 percent used), so the 10 style prompts and the 12 character and product scenes were not run
- Krea 2 Turbo has no image editing (the app says to switch to a model that supports it), so the editing test was not run
- The upscaler was not run: 10 energy would have taken use to 97 percent of the allowance, so the size test has no grade
- Photorealism used 12 images (one run of each portrait and product prompt), not 24 images in two runs
- I first set the page's default 4:3 frame by mistake: the five images made at that size (1,280 by 960 pixels) are kept apart, four of them saved, and the 1:1 reruns are the ones graded; the first runs therefore cost 10 extra energy
- No plan was bought: paid watermarks, the PRO models and the paid upscale limits were read from the plans page only, and prompts naming real people or brands were not run
Still to test
- Styles, consistency and editing, on a model that edits or on a paid plan
- The 2K upscale, opened and compared with the original
- A paid download, its watermark and its file
- Whether real people and brands are refused
- A second run of each portrait and product prompt