Pixelcut
AI product photography and visual design studio by Pixelcut Inc. in San Francisco (4 image capabilities benchmarked)
Product node overview & trust baseline
Operator provenance, privacy baseline and site-wide pricing belong directly to the product node and are inherited across linked task edgesPixelcut, operated by Pixelcut Inc. (San Francisco, CA, USA) (San Francisco, USA (Y Combinator S20)) under a "Free tier + subscription" model, has 4 published task assessments on RealTest (Background removal 91/100, Image upscaling 89/100, Object removal 88/100, Image generation 87/100). In background removal, hair detail and the download flow perform well, with clear free-tier labels; login-free output is 1080px and full size requires login. In image upscaling, login-free, ad-free 2× watermark-free downloads are fast; free daily downloads are limited and 4× requires Pro.
San Francisco, USA (Y Combinator S20) · Founded 2020
Provenance baseline: 18.4/20 · Pixelcut Inc. (San Francisco AI design startup, Y Combinator S20 alum, trusted by tens of millions of e-commerce creators worldwide).
Annual charges for Pro and all eight Business volumes were verified separately in Stripe checkout. Unlimited paid removal coexists with batch-export limits. The free tier still says Limited without an exact removal count; it is not treated as unlimited.
Pricing transparency baseline: 26.3/35 · Cancellation: Cancel anytime; refund eligibility unverified.
Guest sessions cache results locally in the active browser tab
Google Analytics · PostHog
Product-level pricing & rules baseline
Verified plans, billing periods and cancellation terms owned by this product node; task-specific credit costs and free export limits are detailed per capability edge belowWeb Free, Pro and all eight Business volumes with monthly/annual selection; API separate · 2026-10-20
19 plan / billing-period records, all listed; — means that period is not recorded.
Annual charges for Pro and all eight Business volumes were verified separately in Stripe checkout. Unlimited paid removal coexists with batch-export limits. The free tier still says Limited without an exact removal count; it is not treated as unlimited.
Verification has gaps; unknown prices and quotas are not inferred.
Cancellation, sources & verification record
Cancel anytime; refund eligibility unverified.
Official plans and usage terms ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗
Annual USD checkout summaries for each volume ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗ · Capture 4 ↗ · Capture 5 ↗ · Capture 6 ↗ · Capture 7 ↗ · Capture 8 ↗ · Capture 9 ↗
Archived official pricing sources: Official plans and usage terms (2026-10-20) [text snapshot ↗] · Annual USD checkout summaries for each volume (2026-10-20) [text snapshot ↗]
Capability map & linked task edges
Whenever a capability passes fixed-fixture testing, an edge is automatically linked between this product node and the task topic for horizontal comparison.
| Linked task edge | Score | Assessment · historical test | Test scope & delivery | Inspect capability / follow edge |
|---|---|---|---|---|
| Image generation | 87/100 | Switching to the free model yields four 1600px clean images with good text and counts; login is required and the default Pro model first triggers a paywall. | 1600×1600 (Unwatermarked PNG/JPEG) · Historical description; output not attached Historical test · 2026-10-16 Signed-in flow; paid tier not standardized · Fixed English prompt · Historical page record; output not attached | |
| Image upscaling | 89/100 | Login-free, ad-free 2× watermark-free downloads are fast; free daily downloads are limited and 4× requires Pro. | 1024×1024 (PNG · 2.0x Fast Upscale) · Historical description; output not attached Historical test · 2026-10-16 Guest flow · Fixed low-resolution fixture · Historical page record; output not attached | |
| Object removal | 88/100 | Free login-free watermark-free export has no third-party ads and easy brush and comparison controls; blending is natural, but complex detail still needs inspection. | 1024×1024 (Unwatermarked PNG) · Historical description; output not attached Historical test · 2026-10-16 Guest flow · Fixed occlusion fixture · Historical page record; output not attached | |
| Background removal | 91/100 | Hair detail and the download flow perform well, with clear free-tier labels; login-free output is 1080px and full size requires login. | 1080 × 1080 · PNG Historical test · 2026-10-16 Guest flow · Curly-haired portrait · Output attached |
Discovered product capabilities (pending fixture test / edge incubation)
Recorded capability · Not yet scoredThese capabilities have been identified on the product but have not yet undergone fixed-fixture testing. When an emerging capability co-occurs across multiple products, it incubates a new task edge.
Normalized from official "Virtual Studio" into the AI product background candidate edge
Tested task edges (with scores and native artifacts) are strictly separated from discovered untested capabilities (unscored).
Peer product comparison across shared task edges
Matched by overlapping tested task edges; compare site-wide billing models alongside per-task benchmark scoresThe official pricing page and credit table state that guests get 5 free runs and registered Free accounts receive 10 one-time Nero Credits (Standard/Photo/Anime 2× costs 1 credit/image, 4× costs 2 credits/image, paid-only 8× costs 4 credits/image). Standard monthly is USD 21.95/month promotional (regular USD 24.99/month, 300 credits/month). The yearly promo banner displays only "7.66 USD per month" (3,000 credits/year) without printing the annual invoice total before login, so the annual total is marked unverified.
- Image generation:Pixelcut 87/100 vs Nero AI 91/100
- Image upscaling:Pixelcut 89/100 vs Nero AI 84/100
- Object removal:Pixelcut 88/100 vs Nero AI 77/100
- Background removal:Pixelcut 91/100 vs Nero AI 80/100
Actual annual charges $59.99/$263.99. A 3-day trial is offered; cancel before expiry to avoid charges. Exact daily free count is unspecified. The 11,500-credit pack includes a current 1,500-credit bonus.
- Image generation:Pixelcut 87/100 vs PicWish 85/100
- Image upscaling:Pixelcut 89/100 vs PicWish 82/100
- Object removal:Pixelcut 88/100 vs PicWish 84/100
- Background removal:Pixelcut 91/100 vs PicWish 83/100
High-volume tiers expanded; removal costs 1 credit/image, other operations differ. Taxes are not stated on the cards.
- Image generation:Pixelcut 87/100 vs Cutout.pro 87/100
- Image upscaling:Pixelcut 89/100 vs Cutout.pro 87/100
- Object removal:Pixelcut 88/100 vs Cutout.pro 73/100
- Background removal:Pixelcut 91/100 vs Cutout.pro 81/100
The feature FAQ confirms watermark-free transparent PNG/blurred-background downloads with a free account; replacing with a new background allows one free download per day, while premium backgrounds require a subscription. The pricing table says Limited without quantifying daily transparent-PNG limits. General AI credits or the one-daily replacement limit are not applied to transparent cutouts.
- Image generation:Pixelcut 87/100 vs MyEdit 88/100
- Image upscaling:Pixelcut 89/100 vs MyEdit 67/100
- Object removal:Pixelcut 88/100 vs MyEdit 57/100
- Background removal:Pixelcut 91/100 vs MyEdit 67/100
Linked capability test: Image generation
2026-10-16 · No native output attached. Page evidence alone cannot establish export specifications or file validity.
Task quality
Text, counts and hand structure matched this prompt; consistency in other scenes was not retested.
Detailed facts & historical record
- Measured fixture results
Not verified
- Fixture findings
Text, counts and hand structure matched this prompt; consistency in other scenes was not retested.
- Original historical observation
Near-perfect 5-point constraint adherence (64.5/65): exact "REALTEST 2026" text, exact 3 red apples + 2 blue ceramic mugs, and a 5-finger articulated brass hand.
- Suitable uses · historical assessment
Users needing exact English typography rendering, strict multi-object count/color adherence, and 1600×1600 unwatermarked exports
- Result & output evidence
No native output obtained; the preview does not establish final export quality.
Delivery capabilities
1600×1600 (Unwatermarked PNG/JPEG) · Historical description; output not attached · Check watermark record
Detailed facts & historical record
- Recorded export specifications
1600×1600 (Unwatermarked PNG/JPEG) · Historical description; output not attached · Check watermark record
No native output is attached; check the evidence for specification and watermark claims.
- Historical specification note · not an official maximum
1600×1600 (Unwatermarked Full-Res)
- Watermark observation
Zero watermark (clean 1600×1600 export)
- Batch & API
Generates 4 variations per prompt with one-click Background Removal / Upscale / Retouch integration
- Recorded free-tier dimensions
Not verified
- Attached file dimensions
Not verified
Speed & reliability
About 21.2s end to end in this run; reliability not retested
Detailed facts & historical record
- Timing observation
Network-observed server wait: 19.4s · End-to-end reference: 21.2s
- Date & fixture
2026-10-16 · anchor_t2i_v1 · US-West (California, AS2914)
- Repeated tests & reliability
No standardized repetition count, success rate or variability range is recorded.
Interaction experience
Login required in this run
Detailed facts & historical record
- Account requirements
Account required in this test
- Sign-in methods
Free Google / Apple / Email login (grants 108 initial credits)
- Manual adjustment
Not verified
- Interaction observations
Google sign-in preserves the exact prompt typed on the landing page, though defaulting to a Pro-locked model creates an unnecessary paywall detour before discovering the free model.
Background & reputation
Pixelcut, Inc. (YC S20) · Long-term reputation not independently verified
Detailed facts & historical record
- Service lifecycle notice
Not verified
- Operator
Pixelcut, Inc. (YC S20)
- Recorded origin
San Francisco, California, USA
- Domain history record
2020–present
- Background & reputation assessment
Operated by Pixelcut, Inc. (Y Combinator S20, San Francisco, CA), offering a multi-model visual workspace.
- External reputation record
Not verified
Engineering & privacy
Uploaded for cloud processing;Explicit retention period unconfirmed; inspect policy record
Detailed facts & historical record
- Processing location
Uploaded for cloud processing
- Third-party tracking record
Not verified
- Engineering & privacy assessment
Free Muse Image model generates 4 concurrent 1600×1600 unwatermarked images (assets.pixelcut.ai) in ~19.4s Network-observed server wait.
- Retention statement
Explicit retention period unconfirmed; inspect policy record
Task free allowance & pricing disclosure
108 free credits upon sign-in; switching to free Muse Image (Meta) model costs only 12 credits per 4-image batch
Prompts for free login on Generate; default model displays a Pro padlock icon
Site-wide plans belong to the product node (see Product-level pricing baseline ↑).
Expand site-wide plans aligned for Image generation (19 records)
Web Free, Pro and all eight Business volumes with monthly/annual selection; API separate · 2026-10-20
19 plan / billing-period records, all listed; — means that period is not recorded.
Annual charges for Pro and all eight Business volumes were verified separately in Stripe checkout. Unlimited paid removal coexists with batch-export limits. The free tier still says Limited without an exact removal count; it is not treated as unlimited.
Verification has gaps; unknown prices and quotas are not inferred.
Cancellation, sources & verification record
Cancel anytime; refund eligibility unverified.
Official plans and usage terms ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗
Annual USD checkout summaries for each volume ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗ · Capture 4 ↗ · Capture 5 ↗ · Capture 6 ↗ · Capture 7 ↗ · Capture 8 ↗ · Capture 9 ↗
Evaluation basis & full scores · Overall 86.8/100
Overall = honesty & reputation × 30% + capability × 70%. Component maxima follow this report’s rubric. List scores are rounded for display; original precision is retained here.
Scoring version 2.3.1 · Test fixture anchor_t2i_v1
Honesty & reputation · 78/100 · 30%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Pricing transparency | 23 / 35 | Landing page (01_landing.png) states "Free to try · No watermarks"; clicking Generate prompts for free Google login (granting 108 credits), but defaults to the Pro-locked model "Nano Banana 2 🔒" which triggers an upgrade paywall unless the user manually switches the model selector to the free "Muse Image (Meta)" model (12 credits per 4-image batch). | Evidence A → |
| Interaction restraint | 23 / 30 | Google sign-in preserves the exact prompt typed on the landing page, though defaulting to a Pro-locked model creates an unnecessary paywall detour before discovering the free model. | Evidence A → |
| Background & reputation | 18 / 20 | Operated by Pixelcut, Inc. (Y Combinator S20, San Francisco, CA), offering a multi-model visual workspace. | Evidence C → |
| Engineering & privacy | 14 / 15 | Free Muse Image model generates 4 concurrent 1600×1600 unwatermarked images (assets.pixelcut.ai) in ~19.4s Network-observed server wait. | Evidence A → |
Capability · 90.5/100 · 70%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Task quality | 64.5 / 65 | Text Rendering (25/25: Exact 2-line stencil "REALTEST 2026" on the wooden sign with 100% spelling accuracy and woodgrain perspective (25/25)) · Count & Color Binding (20/20: Strictly 3 red apples with water droplets on the left + 2 speckled blue ceramic mugs on the right, zero color bleeding (20/20)) · 5-Finger Hand & Structure (19.5/20: Vintage brass robotic right hand displays 4 fingers + 1 thumb (5 distinct articulated digits) gripping the wooden sign (19.5/20)) | Evidence B → |
| Deliverable specifications | 20 / 20 | Clean unwatermarked export (8/8); Export & Max deliverable spec: 1600×1600 (Unwatermarked Full-Res) | Evidence B → |
| Processing speed | 6 / 15 | Network-observed server wait: 19.4s (API mode: async_poll; end-to-end 21.2s shown for reference only, never penalized) | Evidence B → |
Legacy scores may include advertised maximum capabilities. Read tested output separately from paid capabilities. Records without matching tiers and scope are not ranked together.
Linked capability test: Image upscaling
2026-10-16 · No native output attached. Page evidence alone cannot establish export specifications or file validity.
Task quality
Tested scale 2×; historical record only
Detailed facts & historical record
- Measured fixture results
Tested scale: 2×
- Fixture findings
The tested 2× result had clear lettering without obvious artifacts, at 1024px.
- Original historical observation
PSNR=27.08dB · SSIM=0.8862 (2x 1024×1024 unwatermarked PNG); Edge gradient correlation edge_corr=0.6080 (+4.24% over Bicubic); high-frequency recovery hf_ratio=0.4716 (+10.65%); Calibration plate PSNR=15.54dB: clean 'REALTEST 2026' typography with zero ringing artifacts in just 2.64s.
- Suitable uses · historical assessment
Users looking for the most consistent all-around free suite (no-login unwatermarked exports across BG Remover, T2I, and 2x Upscaler)
- Result & output evidence
No native output obtained; the preview does not establish final export quality.
Delivery capabilities
1024×1024 (PNG · 2.0x Fast Upscale) · Historical description; output not attached · Check watermark record
Detailed facts & historical record
- Recorded export specifications
1024×1024 (PNG · 2.0x Fast Upscale) · Historical description; output not attached · Check watermark record
No native output is attached; check the evidence for specification and watermark claims.
- Historical specification note · not an official maximum
1024×1024 (PNG · 2.0x Fast Upscale)
- Watermark observation
No watermark
- Batch & API
Pro plan supports 1,000 batch exports/month and developer API
- Recorded free-tier dimensions
Not verified
- Attached file dimensions
Not verified
Speed & reliability
About 5.19s end to end in this run; reliability not retested
Detailed facts & historical record
- Timing observation
Network-observed server wait: 2.64s · End-to-end reference: 5.19s
- Date & fixture
2026-10-16 · anchor_upscale_v1 · US-West (California, AS2914)
- Repeated tests & reliability
No standardized repetition count, success rate or variability range is recorded.
Interaction experience
No login in this run
Detailed facts & historical record
- Account requirements
No login in this test
- Sign-in methods
No login required for 2x upscaling and unwatermarked download; optional Google/Apple/Email login
- Manual adjustment
Not verified
- Interaction observations
Spotless modern AI Image Editor workspace with zero third-party ads, zero forced login walls for 2x, and a built-in Compare toggle.
Background & reputation
Pixelcut Inc. (San Francisco, USA) · Long-term reputation not independently verified
Detailed facts & historical record
- Service lifecycle notice
Not verified
- Operator
Pixelcut Inc. (San Francisco, USA)
- Recorded origin
San Francisco, CA, USA
- Domain history record
2020–Present
- Background & reputation assessment
Operated by YC-backed Pixelcut Inc. (San Francisco, USA), trusted by 200,000+ brands.
- External reputation record
Not verified
Engineering & privacy
Uploaded for cloud processing;Explicit retention period unconfirmed; inspect policy record
Detailed facts & historical record
- Processing location
Uploaded for cloud processing
- Third-party tracking record
Not verified
- Engineering & privacy assessment
Fast gRPC architecture: 2.55s upload and 2.64s pure server Upscale inference, exporting a 1.87MB lossless PNG.
- Retention statement
Explicit retention period unconfirmed; inspect policy record
Task free allowance & pricing disclosure
3 free unwatermarked 2x (1024×1024) downloads per day with no login ('3 downloads left today' shown in menu); 4x requires Pro
Download menu transparently displays remaining daily quota ('3 downloads left today'), and clicking 4x opens a clear Pro pricing modal (04_4x_check.png)
Site-wide plans belong to the product node (see Product-level pricing baseline ↑).
Expand site-wide plans aligned for Image upscaling (19 records)
Web Free, Pro and all eight Business volumes with monthly/annual selection; API separate · 2026-10-20
19 plan / billing-period records, all listed; — means that period is not recorded.
Annual charges for Pro and all eight Business volumes were verified separately in Stripe checkout. Unlimited paid removal coexists with batch-export limits. The free tier still says Limited without an exact removal count; it is not treated as unlimited.
Verification has gaps; unknown prices and quotas are not inferred.
Cancellation, sources & verification record
Cancel anytime; refund eligibility unverified.
Official plans and usage terms ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗
Annual USD checkout summaries for each volume ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗ · Capture 4 ↗ · Capture 5 ↗ · Capture 6 ↗ · Capture 7 ↗ · Capture 8 ↗ · Capture 9 ↗
Evaluation basis & full scores · Overall 89.2/100
Overall = honesty & reputation × 30% + capability × 70%. Component maxima follow this report’s rubric. List scores are rounded for display; original precision is retained here.
Scoring version 2.3.1 · Test fixture anchor_upscale_v1
Honesty & reputation · 92/100 · 30%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Pricing transparency | 31 / 35 | Delivers no-login 512×512 -> 1024×1024 (2x) upscaling and unwatermarked PNG download; the Download menu (03_download.png) clearly states '3 downloads left today', while selecting 4x opens the 'Get Unlimited 4x Upscales' Pro modal ($8/mo yearly). | Evidence A → |
| Interaction restraint | 28 / 30 | Spotless modern AI Image Editor workspace with zero third-party ads, zero forced login walls for 2x, and a built-in Compare toggle. | Evidence A → |
| Background & reputation | 18 / 20 | Operated by YC-backed Pixelcut Inc. (San Francisco, USA), trusted by 200,000+ brands. | Evidence C → |
| Engineering & privacy | 15 / 15 | Fast gRPC architecture: 2.55s upload and 2.64s pure server Upscale inference, exporting a 1.87MB lossless PNG. | Evidence A → |
Capability · 88/100 · 70%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Task quality | 56 / 65 | PSNR vs 2048px GT: 27.08 dB · SSIM: 0.8862 · Edge Gradient Correlation: 0.608 · High-Freq Sharpness Ratio: 0.4716 · Calibration Plate PSNR: 15.54 dB | Evidence B → |
| Deliverable specifications | 17 / 20 | Clean unwatermarked export (8/8); Free scale factor: 2x (1024×1024 (PNG · 2.0x Fast Upscale)) | Evidence B → |
| Processing speed | 15 / 15 | Network-observed server wait: 2.64s (API mode: sync_post; end-to-end 5.19s shown for reference only, never penalized) | Evidence B → |
Legacy scores may include advertised maximum capabilities. Read tested output separately from paid capabilities. Records without matching tiers and scope are not ranked together.
Linked capability test: Object removal
2026-10-16 · No native output attached. Page evidence alone cannot establish export specifications or file validity.
Task quality
Removed all three signs with coherent background texture; inspect newly generated content.
Detailed facts & historical record
- Measured fixture results
Not verified
- Fixture findings
Removed all three signs with coherent background texture; inspect newly generated content.
- Original historical observation
On the anchor_inpaint_v1 benchmark, Pixelcut Magic Eraser cleanly removed the red RESERVED sign, yellow SALE 50% badge, and blue FREE WIFI sign in one pass while synthesizing coherent bookshelf wood and hanging foliage (inpaint_psnr=16.08dB, unmasked_psnr=24.43dB).
- Suitable uses · historical assessment
Users seeking fast, no-login, watermark-free object/text removal from an all-in-one suite that excels across all 4 categories
- Result & output evidence
No native output obtained; the preview does not establish final export quality.
Delivery capabilities
1024×1024 (Unwatermarked PNG) · Historical description; output not attached · Check watermark record
Detailed facts & historical record
- Recorded export specifications
1024×1024 (Unwatermarked PNG) · Historical description; output not attached · Check watermark record
No native output is attached; check the evidence for specification and watermark claims.
- Historical specification note · not an official maximum
1024×1024 (Unwatermarked PNG)
- Watermark observation
100% Watermark-Free
- Batch & API
Supports iterative multi-pass brushing and Pro API access
- Recorded free-tier dimensions
Not verified
- Attached file dimensions
Not verified
Speed & reliability
About 10.6s end to end in this run; reliability not retested
Detailed facts & historical record
- Timing observation
Network-observed server wait: 8.2s · End-to-end reference: 10.6s
- Date & fixture
2026-10-16 · anchor_inpaint_v1 · US-West (California, AS2914)
- Repeated tests & reliability
No standardized repetition count, success rate or variability range is recorded.
Interaction experience
No login in this run; manual adjustment recorded
Detailed facts & historical record
- Account requirements
No login in this test
- Sign-in methods
No login required for painting and unwatermarked download; Google / Apple / Email also supported
- Manual adjustment
Manual erase / adjustment recorded in the historical flow
- Interaction observations
Clean workspace with a Brush Size slider, Erase button, and a built-in Compare toggle on top. Zero third-party ads.
Background & reputation
Pixelcut Inc. (San Francisco, USA) · Long-term reputation not independently verified
Detailed facts & historical record
- Service lifecycle notice
Not verified
- Operator
Pixelcut Inc. (San Francisco, USA)
- Recorded origin
San Francisco, CA, USA
- Domain history record
2020–Present
- Background & reputation assessment
Developed by Pixelcut Inc., a Y Combinator-backed AI visual design company based in San Francisco.
- External reputation record
Not verified
Engineering & privacy
Uploaded for cloud processing;Explicit retention period unconfirmed; inspect policy record
Detailed facts & historical record
- Processing location
Uploaded for cloud processing
- Third-party tracking record
Not verified
- Engineering & privacy assessment
Stable cloud inpainting pipeline completing 3-region removal in ~8.2s with zero watermark or edge artifacts.
- Retention statement
Explicit retention period unconfirmed; inspect policy record
Task free allowance & pricing disclosure
Free no-login brushing, erasing, and unwatermarked export
Transparent UI with no hidden paywall traps
Site-wide plans belong to the product node (see Product-level pricing baseline ↑).
Expand site-wide plans aligned for Object removal (19 records)
Web Free, Pro and all eight Business volumes with monthly/annual selection; API separate · 2026-10-20
19 plan / billing-period records, all listed; — means that period is not recorded.
Annual charges for Pro and all eight Business volumes were verified separately in Stripe checkout. Unlimited paid removal coexists with batch-export limits. The free tier still says Limited without an exact removal count; it is not treated as unlimited.
Verification has gaps; unknown prices and quotas are not inferred.
Cancellation, sources & verification record
Cancel anytime; refund eligibility unverified.
Official plans and usage terms ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗
Annual USD checkout summaries for each volume ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗ · Capture 4 ↗ · Capture 5 ↗ · Capture 6 ↗ · Capture 7 ↗ · Capture 8 ↗ · Capture 9 ↗
Evaluation basis & full scores · Overall 88.1/100
Overall = honesty & reputation × 30% + capability × 70%. Component maxima follow this report’s rubric. List scores are rounded for display; original precision is retained here.
Scoring version 2.3.1 · Test fixture anchor_inpaint_v1
Honesty & reputation · 93/100 · 30%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Pricing transparency | 32 / 35 | No login required to upload an image, paint over all 3 occluding signs, click Erase to generate an unwatermarked result (03_download.png), and download immediately without login walls or bait-and-switch modals. | Evidence A → |
| Interaction restraint | 28 / 30 | Clean workspace with a Brush Size slider, Erase button, and a built-in Compare toggle on top. Zero third-party ads. | Evidence A → |
| Background & reputation | 18 / 20 | Developed by Pixelcut Inc., a Y Combinator-backed AI visual design company based in San Francisco. | Evidence C → |
| Engineering & privacy | 15 / 15 | Stable cloud inpainting pipeline completing 3-region removal in ~8.2s with zero watermark or edge artifacts. | Evidence A → |
Capability · 86/100 · 70%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Task quality | 55 / 65 | Occluded Region GT Restoration (25.5/30: structural PSNR=16.08dB · SSIM=0.5028) · Unmasked Preservation (18.0/20: unmasked_psnr=24.43dB) · Texture Coherence & Zero Ghosting (11.5/15: all 3 signs completely erased with natural bookshelf & vine continuation) | Evidence B → |
| Deliverable specifications | 18 / 20 | Free no-login unwatermarked 1024×1024 PNG export (18/20) | Evidence B → |
| Processing speed | 13 / 15 | 8.2s cloud inpainting inference across 3 occluded regions | Evidence B → |
Legacy scores may include advertised maximum capabilities. Read tested output separately from paid capabilities. Records without matching tiers and scope are not ranked together.
Linked capability test: Background removal
2026-10-16 · Native output attached. Findings cover the recorded fixture and conditions; paid specifications may be untested.
Task quality
Fine hair transitions and strong subject retention in this test; coverage is limited to the tested portrait and conditions.
Detailed facts & historical record
- Measured fixture results
Edge error MAD: 0.0701 (Lower is better) · Global error SAD(k): 26.707 (Lower is better) · Foreground retention: 99.940% (Higher is better) · Background residue: 0.018% (Lower is better)
- Fixture findings
Fine hair transitions and strong subject retention in this test; coverage is limited to the tested portrait and conditions.
- Original historical observation
Hair strand transitions are remarkably crisp and natural (transition-band error mad_band=0.0701, SAD(k)=26.71, gradient error 0.0207), with 99.94% foreground recall and only 0.018% background leak—ranking in the absolute top tier of all tested tools.
- Suitable uses · historical assessment
E-commerce sellers and portrait designers who want top-tier hair edge precision (mad_band=0.0701) with instant 1080p guest export or 2048p signed-in export
- Result & output evidence
Delivery capabilities
1080 × 1080 · PNG · No watermark in test
Detailed facts & historical record
- Recorded export specifications
1080 × 1080 · PNG · No watermark in test
- Historical specification note · not an official maximum
2048×2048 (1080×1080 free guest preview export; dropdown explicitly states 'Full resolution 2048×2048 — Free with sign in')
- Watermark observation
None
- Batch & API
Supports Batch Edit (Pro subscription feature)
- Recorded free-tier dimensions
1080 × 1080
- Attached file dimensions
1080 × 1080 · PNG
Speed & reliability
About 16.87s end to end in this run; reliability not retested
Detailed facts & historical record
- Timing observation
Network-observed server wait: 3.03s · End-to-end reference: 16.87s
- Date & fixture
2026-10-16 · anchor_bgremove_v1 · US-West (California, AS2914)
- Repeated tests & reliability
No standardized repetition count, success rate or variability range is recorded.
Interaction experience
No login in this run
Detailed facts & historical record
- Account requirements
No login in this test
- Sign-in methods
No login required for 1080×1080 export; Google / Apple / Email sign-in unlocks free 2048×2048 full-resolution export
- Manual adjustment
Built-in Retouch brush, Expand, Upscale, Shadow, and AI Background generator in left sidebar
- Interaction observations
Enters the clean online workspace immediately after upload with zero popup spam; guests can export a 1080×1080 watermark-free PNG in one click.
Background & reputation
Pixelcut Inc. (San Francisco, CA, USA) · Long-term reputation not independently verified
Detailed facts & historical record
- Service lifecycle notice
Not verified
- Operator
Pixelcut Inc. (San Francisco, CA, USA)
- Recorded origin
San Francisco, USA (Y Combinator S20)
- Domain history record
Founded 2020
- Background & reputation assessment
Pixelcut Inc. (San Francisco AI design startup, Y Combinator S20 alum, trusted by tens of millions of e-commerce creators worldwide).
- External reputation record
Leading AI product photography tool whose proprietary matting and shadow models are widely praised by e-commerce merchants.
Engineering & privacy
Uploaded for cloud processing;Explicit retention period unconfirmed; inspect policy record
Detailed facts & historical record
- Processing location
Uploaded for cloud processing
- Third-party tracking record
Google Analytics · PostHog
- Engineering & privacy assessment
Synchronously returns the PNG cutout via dedicated endpoint api2.pixelcut.app/image/matte/v1 (server_s=3.03s) with a clean network pipeline.
- Retention statement
Explicit retention period unconfirmed; inspect policy record
Task free allowance & pricing disclosure
Single-image background removal is free; Batch Edit requires upgrading to Pixelcut Pro
Clearly disclosed in the Download dropdown menu with exact pixel dimensions
Site-wide plans belong to the product node (see Product-level pricing baseline ↑).
Expand site-wide plans aligned for Background removal (19 records)
Web Free, Pro and all eight Business volumes with monthly/annual selection; API separate · 2026-10-20
19 plan / billing-period records, all listed; — means that period is not recorded.
Annual charges for Pro and all eight Business volumes were verified separately in Stripe checkout. Unlimited paid removal coexists with batch-export limits. The free tier still says Limited without an exact removal count; it is not treated as unlimited.
Verification has gaps; unknown prices and quotas are not inferred.
Cancellation, sources & verification record
Cancel anytime; refund eligibility unverified.
Official plans and usage terms ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗
Annual USD checkout summaries for each volume ↗ · 2026-10-20 Capture 1 ↗ · Capture 2 ↗ · Capture 3 ↗ · Capture 4 ↗ · Capture 5 ↗ · Capture 6 ↗ · Capture 7 ↗ · Capture 8 ↗ · Capture 9 ↗
Evaluation basis & full scores · Overall 90.6/100
Overall = honesty & reputation × 30% + capability × 70%. Component maxima follow this report’s rubric. List scores are rounded for display; original precision is retained here.
Scoring version 2.3.1 · Test fixture anchor_bgremove_v1
Honesty & reputation · 83/100 · 30%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Pricing transparency | 30 / 40 | Clicking the top-right Download dropdown clearly lists both tiers upfront: 'Preview resolution 1080×1080 — Free' and 'Full resolution 2048×2048 — Free with sign in', with zero hidden paywalls or watermark ambushes. | Evidence A → |
| Interaction restraint | 18 / 20 | Enters the clean online workspace immediately after upload with zero popup spam; guests can export a 1080×1080 watermark-free PNG in one click. | Evidence A → |
| Background & reputation | 23 / 25 | Pixelcut Inc. (San Francisco AI design startup, Y Combinator S20 alum, trusted by tens of millions of e-commerce creators worldwide). | Evidence A → |
| Engineering & privacy | 12 / 15 | Synchronously returns the PNG cutout via dedicated endpoint api2.pixelcut.app/image/matte/v1 (server_s=3.03s) with a clean network pipeline. | Evidence A → |
Capability · 93.9/100 · 70%
| Criterion | Score | Rationale | Basis |
|---|---|---|---|
| Task quality | 60.9 / 65 | Transition band error mad_band=0.0701 · Global SAD(k)=26.707 · Foreground recall 99.94% / Background leak 0.018% | Evidence A → |
| Deliverable specifications | 20 / 20 | True Alpha channel (8/8); Max deliverable spec: 2048×2048 (1080×1080 free guest preview export; dropdown explicitly states 'Full resolution 2048×2048 — Free with sign in') | Evidence A → |
| Processing speed | 13 / 15 | Network-observed server wait: 3.03s (API mode: sync; network upload 9.79s / end-to-end 16.87s shown for reference only, never penalized) | Evidence A → |
Legacy scores may include advertised maximum capabilities. Read tested output separately from paid capabilities. Records without matching tiers and scope are not ranked together.
