More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Image Refinement Studio Pro

    0

    Refine approved product images and posters with source-aware layers, planar patches and protected-pixel checks. Keep exact copy, count and identity in scope.

    $2.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Image Refinement Studio Pro

    Image Refinement Studio Pro

    Example session with this skill installed

    Controlled fictional example: update the approved FORM STUDIO campaign headline from ORIGINAL EDITION to PRECISION EDITION and the supporting line to “Refine what matters. Keep the rest.” Keep the product image, top brand and footer unchanged. Use the supplied explicit allowed/protection masks. Deliver a 1536 x 1024 PNG, same-run baseline/effective mask and verification report. Do not claim a real product, neural restoration or a layered PSD.

    • Read your context and instructions
    • Compiled the image refinement studio
    • Generated the image

    Controlled campaign revision

    Output: campaign-refined.png, actual 1536 x 1024, flattened 8-bit sRGB PNG.

    The requested headline and supporting line were updated within the permitted campaign area. The product still life, brand and footer were protected against the same-run normalized baseline. The photographic source is an original generated fictional test fixture; it is not a real SKU or customer identity test.

    Saved-file checkResult
    Decoding and expected dimensionsPassed
    Protected baseline pixels1,210,664
    Protected pixels changed at tolerance 00
    Independent readerPillow/numpy reopened saved PNGs and compared arrays
    Negative controlOne deliberately changed protected pixel was rejected
    Input filesRetained with SHA-256 checks
    Editing scopeFlattened PNG plus task-specific layer config/masks; no PSD-layer claim

    The actual work products include campaign-refined.png, campaign-baseline.png, campaign-effective.png, campaign-report.json and campaign-verification.json. The rerenderable configuration references authorized task assets, not protected Skill source. A numeric pass establishes scoped pixel behavior, not label authenticity, visual perfection or user acceptance.

    Next step: view the full image and original-scale critical regions, check the exact approved copy, then retrieve the actual files through the host artifact mechanism. Authenticated Agensi buyer transfer and retrieval are still unverified.

    campaign-refined.png

    PNG · 1536×1024

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Keep the approved design. Refine the part that actually needs work.

    Image Refinement Studio Pro is for the moment when the layout is already right, but a product detail, local patch, campaign line or output requirement needs careful attention. It separates design authority, genuine detail sources, style references and defect annotations, then plans a scoped change and checks the actual saved result.

    Practical work, beyond a larger canvas

    The English workflow covers source-matched product detail, packaging and accessory counts, exact campaign text, transparent-edge review, foreground/background ownership, manual planar face mapping and explicit large-output delivery. Twenty-four playbooks span products, materials, ecommerce collages, retail/event copy, marks, formats, editorial work, mockups, masks, EXIF and sample-depth decisions. They are adaptable planning routes, not 24 tested customer projects.

    Four focused utilities

    A single static ESM helper offers inspect, compose, warp and verify. It has no relative Skill-module imports, runtime code extraction or network calls. The agent handles task configs and mask/correspondence preparation within the host's permitted tool rules; the buyer need not write command-line JSON.

    • Inspect actual decoded dimensions, SHA-256, orientation, colour space and alpha rather than trusting a file label.
    • Compose explicitly ordered layers with source references, visible masks and final global protection.
    • Map a genuine planar patch from four supplied corners; report its actual cropped offset and transformation limits.
    • Reopen PNGs and verify dimensions/hashes plus protected decoded pixels against the actual same-run baseline and effective mask.

    Evidence that has actually been checked

    All 30 original behavioral checks pass against the new entry. Four controlled work products were run with the full unchanged script passed directly to ESM interpreter stdin. A separate Pillow/numpy reader reopened the saved outputs and independently checked arrays and hashes.

    The campaign-copy demonstration preserves 1,210,664 protected pixels with zero protected changes. The known-source detail demonstration preserves 1,245,264 protected pixels; its fully permitted inner crop matches the known original samples. The planar-label demonstration preserves 1,499,154 protected pixels. A deliberately changed protected pixel is rejected at tolerance zero. A 3840 x 2560 file is exported and reopened, explicitly labelled as a resample rather than newly captured/native-generated detail.

    Honest fidelity and deliverables

    The demonstrations use an original fictional photographic fixture and controlled vector artwork. They do not establish real-customer product identity, automatic background removal or restoration of unknown print. Sharper-looking generated texture is not authentic packaging information. The workflow checks sources, count, exact copy, occlusion and visual seams separately from numerical comparisons.

    Deliver inspected PNGs, task-specific jobs/masks and reports, plus only genuinely created and verified native files if the host supports them. A PNG or JSON is not a layered PSD. Original bytes, protected normalized pixels, visual identity and user acceptance remain different claims.

    Included and required

    The one-time purchase provides an 18-file English text core with instructions, playbooks, example configs and the static helper. Image-generation models, fonts and native runtimes are not bundled. The helper requires approved Node.js 20.9+ and Sharp 0.35.5; generative work needs the buyer agent's separately available approved tool and credits. The protected Skill is loaded through Agensi, not promised as an offline source package. Generated artwork and task configs are distinct work products.

    Compatibility disclosure: Authenticated Agensi buyer resource loading, complete-script transfer to ESM interpreter stdin, Sharp/native dependency availability and finished-artifact retrieval remain unverified. Creator-local stdin tests are not end-to-end buyer acceptance. Twenty-four playbooks are guidance, not completed projects. Demonstrations use one original imagegen-created fictional photographic fixture, not real customer/SKU fidelity evidence. Known-source detail replacement uses an intentionally degraded proxy; no absent captured detail is recovered. Output is flattened 8-bit sRGB PNG plus task configs/reports. No built-in neural super-resolution, automatic cutout, OCR, feature registration, PSD exporter, curved-surface certification or print CMYK/PDF/X production is supplied. A 3840x2560 demonstration is explicitly resampled from 1536x1024. Generation tools/credits, fonts, stock and source identity rights remain separately provided. Follow the host image-tool rules; deterministic utilities are not a bypass.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for E-commerce.

    Authenticated Agensi buyer resource loading, complete-script transfer to ESM interpreter stdin, Sharp/native dependency availability and finished-artifact retrieval remain unverified. Creator-local stdin tests are not end-to-end buyer acceptance. Twenty-four playbooks are guidance, not completed projects. Demonstrations use one original imagegen-created fictional photographic fixture, not real customer/SKU fidelity evidence. Known-source detail replacement uses an intentionally degraded proxy; no absent captured detail is recovered. Output is flattened 8-bit sRGB PNG plus task configs/reports. No built-in neural super-resolution, automatic cutout, OCR, feature registration, PSD exporter, curved-surface certification or print CMYK/PDF/X production is supplied. A 3840x2560 demonstration is explicitly resampled from 1536x1024. Generation tools/credits, fonts, stock and source identity rights remain separately provided. Follow the host image-tool rules; deterministic utilities are not a bypass.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 14 days ago

    Software
    Node.js 20.9+Sharp 0.35.5
    Optional
    Image generation
    orediting tool
    Generation creditsFonts

    Sharp needs its native dependencies. Your agent must view images and pipe the script unchanged to Node.js stdin. Creative edits need your own tool, credits, fonts and input rights.

    Permissions

    Terminal / Shell

    File Scopes

    User-approved
    source
    designs
    matching
    detail
    photos
    exact-copy
    records
    and
    masks
    New
    task/revision
    directories
    for
    generated
    PNGs
    configs
    verification
    reports
    Read
    only
    explicitly
    approved
    dependency
    locations;
    no
    Skill
    code
    in
    buyer
    handoff
    ZIPs

    No network requests, directory scanning, external uploads or automatic dependency installation by the helper. It reads CODEX_PRIMARY_RUNTIME_NODE_MODULES only as an optional approved package path, not arbitrary environment data. Host creative tools may have separate service permissions and costs. Inputs, jobs and masks must not alias outputs; preserve source files and prior versions.

    Listed14 days ago

    What's inside

    Frequently Asked Questions

    The English workflow, 24 practical planning playbooks, example job configs and a single static helper for inspect/compose/warp/verify. It uses your agent and separately approved dependencies. It is not an online image service or a source download licence; access follows Agensi terms and service availability.

    No. The workflow distinguishes known-source replacement, resampling/sharpening and generated reconstruction. No neural super-resolution, automatic cutout, OCR or feature-registration model is included. A larger image cannot establish hidden or absent detail.

    You and the agent define scope and source roles. Where deterministic composition is permitted, explicit global allowed/protection masks constrain the final normalized pixels. The exported effective mask reflects the actual rounded combined permission; verification compares the saved output against its same-run baseline. Real identity and exact copy still need visual/source review.

    The agent can plan that scope and use an approved host tool. The script itself does not automatically detect subjects or remove old background inside transparent glass. Matching source identity, ownership, visibility and masks must be established, and host-required image-editing tools must still be used.

    The bundled compositor/warp export flattened 8-bit sRGB PNGs and reports. Task JSON, sources and masks support rerendering, not Photoshop layer editing. Native PSD/SVG or certified print files require a suitable separate tool and actual reopening/print checks; no built-in exporter or certification is promised.

    The creator example exports 3840 x 2560 by 2.5x resampling a 1536 x 1024 fixture. Those dimensions are checked, but they do not prove native 4K generation or new captured detail. A separate controlled detail example replaces a deliberately degraded proxy region with actual pixels from its known original.

    No. They are adaptable playbooks covering products, copy, collage, planar faces, edges, formats and source handling. Local demonstration uses one fictional photographic fixture in four work products, plus synthetic behavioral checks. No real-customer identity acceptance is claimed.

    Thirty original checks pass against the new single-file entry. Four unchanged-script stdin operations, independent saved-PNG comparisons, explicit source-detail replacement, a protected-pixel negative control and a large raster export are checked locally. Authenticated Agensi buyer resource loading, complete-script transfer to ESM interpreter stdin, Sharp/native dependency availability and finished-artifact retrieval remain unverified. Creator-local stdin tests are not end-to-end buyer acceptance. Twenty-four playbooks are guidance, not completed projects. Demonstrations use one original imagegen-created fictional photographic fixture, not real customer/SKU fidelity evidence. Known-source detail replacement uses an intentionally degraded proxy; no absent captured detail is recovered. Output is flattened 8-bit sRGB PNG plus task configs/reports. No built-in neural super-resolution, automatic cutout, OCR, feature registration, PSD exporter, curved-surface certification or print CMYK/PDF/X production is supplied. A 3840x2560 demonstration is explicitly resampled from 1536x1024. Generation tools/credits, fonts, stock and source identity rights remain separately provided. Follow the host image-tool rules; deterministic utilities are not a bypass.

    New in E-commerce