Test log
Pokecut test data
Raw results for the Pokecut 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 | – | n/a |
| T02 | Character & Subject Consistency | – | n/a |
| T03 | Editing, Inpainting & Image-to-Image | – | n/a |
| T04 | Prompt Adherence | Elements correct: 16 of 18 | 4/5 |
| T05 | Photorealism & Anatomy | Images that pass: 2 of 2 | 4/5 |
| T06 | Text in Images | Words right: 11 of 11 | 4/5 |
| T07 | Resolution & Upscaling | Native size: 1,536 and 2,656 px square | 3/5 |
| T08 | Speed per Image | Median per image: 30 s | 3/5 |
| T09 | Usage Rights, Watermarks & Privacy | Points met: 5 of 7 checked | 3/5 |
| T10 | Pricing & Value | Free images: 4 at sign-in | 3/5 |
| Overall (in the review) | 3.8/5 | ||
Style Range
n/aNot Tested- Method
- One subject, a lighthouse, in ten styles named by technique.
- Remark
- Not graded: no style image was made.
Raw notes
One subject, a lighthouse, in ten styles named by technique.
Not run. The ten style images would have cost 50 credits and the account had 20.
- Run the ten style prompts on a later pass with credits bought for the test.
Character & Subject Consistency
n/aNot Tested- Method
- A described woman and a described thermos, each in six scenes.
- Remark
- Not graded: no consistency image was made.
Raw notes
A described woman and a described thermos, each in six scenes.
Not run. Twelve images would have cost 60 credits. The canvas also attaches the last image as a reference by itself, which would have changed a clean description-only test.
- Run the twelve scenes on a later pass.
Editing, Inpainting & Image-to-Image
n/aNot Tested- Method
- Twelve fixed edits: four masked fills, four written edits, two restyles of my reference photos and two frame expansions.
- Remark
- Not graded: no edit was run.
Raw notes
Twelve fixed edits: four masked fills, four written edits, two restyles of my reference photos and two frame expansions.
Not run. The canvas offers chat editing, erase, replace, background and expand, and each edit uses credits. The allowance was spent on the three generations.
- Run the twelve edits with the four reference photos on a later pass.
Prompt Adherence
4/5Tested- Method
- Three of the 40 fixed prompts (P4 the chef, T1 the coffee shop, S1 the kitchen scene), typed in by script, first image kept, every written element counted at full size in the downloaded file.
- Elements correct
- 16 of 18
- Busy scene
- 6 of 8
- Remark
- Scale 4: 80 to 89 percent of elements correct. 88.9 percent sits at the top of that band; 90 percent would be a 5.
Raw notes
Three of the 40 fixed prompts (P4 the chef, T1 the coffee shop, S1 the kitchen scene), typed in by script, first image kept, every written element counted at full size in the downloaded file.
16 of 18 elements were right (88.9 percent). Chef 5 of 5, coffee shop 5 of 5, kitchen scene 6 of 8.
Kitchen scene: four apples instead of three, and the dog lies beside the table, not under it. The jug is blue, the apples are red, the cat sits on the chair, the table is a kitchen table and the curtains are yellow.
Added content that changed the scene: the coffee shop image carries the chef's hands, because the canvas had attached the first image as a reference.
- 3 of 40 prompts, 18 elements instead of about 240. The count is mine, made by eye at full size; the first image of each prompt counts, with no rerolls.
Photorealism & Anatomy
4/5Tested- Method
- The chef (P4) and the kitchen scene (S1), the two images made from a prompt alone, inspected at 100 percent for faults that give a generated image away.
- Images that pass
- 2 of 2
- Hand and face faults
- 0
- Remark
- Scale 5 needs 22 to 24 of 24 to pass. Two images cannot show that, so the grade is held at 4: no fault in two of two.
Raw notes
The chef (P4) and the kitchen scene (S1), the two images made from a prompt alone, inspected at 100 percent for faults that give a generated image away. The coffee shop image is left out because it was made from an attached reference.
2 of 2 images pass as photographs. The chef's hands have five fingers each, the knife grip is correct and the skin has pores and hair. The apples carry water drops, and the cat and the dog have correct faces and paws.
Hand faults 0. Face faults 0. Faults in text-like marks 0.
- Two images, not 24. The two portrait and product runs the test asks for were not made.
Text in Images
4/5Tested- Method
- The coffee shop prompt (T1): a sign above the door that reads "Harbour Bean Coffee Roasters" and a board beside it that reads "Open Daily 7 am to 5 pm", 11 words, every word checked at full size.
- Words right
- 11 of 11
- Images fully correct
- 1 of 1
- Remark
- Scale 4: 85 to 94 percent of words, or four images fully correct. Scale 5 needs five of six images fully correct, and one image cannot show that, so the grade is held at 4.
Raw notes
The coffee shop prompt (T1): a sign above the door that reads "Harbour Bean Coffee Roasters" and a board beside it that reads "Open Daily 7 am to 5 pm", 11 words, every word checked at full size.
11 of 11 words were spelled right and in the place asked for (100 percent). The framed menu in the shop window carries small lettering of its own that cannot be read; the set words are all right.
The sign reads "Harbour Bean Coffee Roasters" in carved white letters on two boards. The chalkboard reads "Open Daily" and "7am to 5pm"; the board writes the times without a space.
- One image of the six set-wording prompts, 11 of the 60 words. Counting words is my call: "7am" and "5pm" written without a space count as right.
Resolution & Upscaling
3/5Tested- Method
- Every downloaded file was measured, the frame shapes were read from the output panel, and the Download button was used once for each image.
- Native size
- 1,536 and 2,656 px square
- Frame shapes
- 8
- Remark
- Scale 3: 1 megapixel or more native, with no upscaler run. The native size alone is above 2 megapixels, but the top two bands need a clean upscale to 4,096 pixels.
Raw notes
Every downloaded file was measured, the frame shapes were read from the output panel, and the Download button was used once for each image.
The chef and the kitchen scene are 1,536 by 1,536 pixels (2.36 megapixels); the coffee shop image is 2,656 by 2,656 pixels (7.05 megapixels). All three are JPEG files of 0.3 to 1.1 megabytes.
The output panel under Auto offers eight shapes: 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, 16:9 and 9:21. With a reference image attached the panel shows no shapes and says the output size will match the first reference image.
The downloads had no watermark and the files carry no content credentials label. No upscale was run.
- Three sizes measured. A retake with a wider window captured all eight frame shapes in one picture.
Speed per Image
3/5Tested- Method
- Each of the three generations was timed from pressing the send arrow to the full-size image on screen, read from the screen recording and the clock on the page.
- Median per image
- 30 s
- Slowest image
- 32 s
- Remark
- Scale 3: a median of 20 to 39.9 seconds.
Raw notes
Each of the three generations was timed from pressing the send arrow to the full-size image on screen, read from the screen recording and the clock on the page.
Median about 30 seconds: the chef about 32, the coffee shop about 30 and the kitchen scene about 26.
The page shows a progress ring and "Less than 1 min" while it works. Nothing failed and no queue message appeared.
- Three images, timed to within two seconds from a recording, not 40 timed by script.
Usage Rights, Watermarks & Privacy
3/5Tested- Method
- The Help Center, Terms of Service and Privacy Policy (read 6 November 2026) and the three downloaded files were checked against the ten-point list.
- Points met
- 5 of 7 checked
- Remark
- Scale 3: 6 or 7 of 10. 5 of 7 checked is 71 percent, so about 7 of 10.
Raw notes
The Help Center, Terms of Service and Privacy Policy (read 6 November 2026) and the three downloaded files were checked against the ten-point list. Seven points could be checked on the free plan.
Met (5): the Help Center says "Yes, you can" use images from the AI Image Generator for commercial purposes, with no plan named, which covers both commercial points; the downloads have no visible watermark; the Privacy Policy says uploaded pictures are not used for any purpose other than the user functions; and it names public data sets or purchased data sets as the training source.
Not met (2): the terms do not say who owns the images you make (they say Pokecut owns the app and its content, and give Pokecut a licence to use your content to run the app); the three files carry no content credentials label.
Not tested (3): a watermark on paid downloads (the Help Center says website downloads add no watermark), whether images are private by default (the Privacy Policy says generated images are deleted after 15 days in one section and 30 days in another), and whether images of real people and brands are refused.
- Terms read by me, not a lawyer. The Terms of Service say users must not use the app's content for revenue-generating work, which I read as the app's own designs and templates, not your pictures; I did not ask the company.
Pricing & Value
3/5Tested- Method
- I generated until the allowance was at 75 percent, read the credit cost of every model in the picker, and read the plans window in both billing views and the credit packs.
- Free images
- 4 at sign-in
- Auto image on SE yearly
- about $0.056
- Remark
- Scale 3: $0.05 to $0.099 an image on the cheapest plan that allows commercial use.
Raw notes
I generated until the allowance was at 75 percent, read the credit cost of every model in the picker, and read the plans window in both billing views and the credit packs.
Free: the sign-in offer promised 10 free credits to new users and the balance read 20; no daily allowance is stated and I saw no refill. The Auto model costs 5 credits, so that is four images. Other models cost 10 credits (ChatGPT Image 2.5, Nano Banana 2 Lite), 15 (Seedream 5.0 Lite and 4.0), 20 (ChatGPT Image 2.0), 30 (Nano Banana 2, Seedream 5.0 Pro) and 35 (Nano Banana Pro).
The cheapest paid plan is SE at $3.33 a month paid yearly ($39.99 a year) with 300 credits a month; the monthly view shows $1 for the first month and no later price. 300 credits are 60 Auto images, about $0.056 an image on the yearly price. Pro is $12.99 a month ($5.83 yearly) with 800 credits, Ultra $18.99 ($7.50 yearly) with 1,500.
Credit packs run from 300 to 5,000 credits at $0.033 down to $0.016 a credit, so an Auto image costs about $0.17 on the smallest pack and $0.08 on the largest. The Help Center offers a refund within 1 day of payment.
- No plan bought. The per-image cost assumes a paid Auto image costs the same 5 credits as a free one.
Coverage
Tested
- Three of the 40 fixed prompts, first image kept: a chef slicing a tomato, a coffee shop with a sign and a board, and the kitchen scene with three apples
- Time and credit cost of every image, read from the screen recording and the balance
- The Download button, three files saved and checked for size, watermark and a content credentials label
- The credit cost of every model in the picker, and the frame shapes under Auto
- The Help Center, Terms of Service and Privacy Policy read, and the plans window in both billing views
Not tested or limited
- The 20 credits paid for three images, so only 3 of the 40 prompts were run: one portrait, one set-wording image and one busy scene
- Style range, consistency and editing were not graded: each needs 10 or 12 more images or edits than the allowance paid for
- No upscale was run, so resolution is graded on native size only; the enhancer costs credits and a fourth image would have used the last of them
- The second image was made on the canvas with the first image attached as a reference by the app, so it contains the chef's hands; its wording is graded, its scene is not counted as a miss
- Only the Auto model was run; the paid models in the picker were not tried, and Auto does not say which model it used
- No plan was bought: a watermark on paid downloads, private-by-default images and whether prompts naming real people or brands are refused were not tested
Still to test
- The ten style prompts and the consistency scenes on a paid plan
- The twelve edits with my reference photos
- An upscale to the largest size
- The paid models, a watermark on paid downloads and whether a request for a real person is refused