AI Landmark Cropping — Body-Aware Smart Crop from Any Anatomical Point, Across Any Pose
Autophoto’s landmark cropping uses AI to detect every major body point — head, eyes, nose, lips, chin, shoulder, knee, shoe, and more — and crops precisely between two landmarks you define. Set “From head / To knee” and every image in your batch is cropped at those exact anatomical boundaries, even when the landmark is not directly visible in the frame. Up to 5,000 images per batch.
Landmark cropping is a pipeline task. Define your crop rule once per deliverable — it applies automatically to every image in the batch, consistently, without manual intervention.
Last updated: April 2026
What consistent cropping actually costs at production volume
Fashion ecommerce has strict visual consistency requirements. Marketplaces, brand guidelines, and catalog standards often require every image to be cropped to the same frame. When a studio shoots 500 garments per day across multiple shot types and poses, applying those crop rules manually creates a serious bottleneck.
Manual crop in Photoshop
A retoucher crops one image manually in 1–3 minutes: open file, identify the correct body position, apply the crop, export. For 500 images per day that’s 8–25 hours of pure cropping work — before any other editing.
Inconsistency across retouchers
Different retouchers crop slightly differently. One crops from the eyebrows, another from the forehead. One includes the chin, another cuts it. Across a 5,000-image catalog, that variation adds up to a visually inconsistent storefront.
Pose variation breaks fixed-pixel crops
A standing model and a seated model shot to the same pixel dimensions produce completely different framing. Fixed-pixel batch crop tools cut the knee on the seated model and leave empty space below the standing model. Neither is correct.
11 body landmarks — each usable as a From point or a To point
Every landmark in Autophoto works bidirectionally. Any landmark can be the upper boundary (From) or the lower boundary (To) of your crop. You set the range — the AI detects where each landmark sits in each image and crops precisely at those two points.
Available crop landmarks
Common From → To crop combinations
Any landmark can be From and any can be To. These are common combinations — you define the range that suits your brand and marketplace requirements.
The crop works even when the landmark is not visible
When you set “From nose / To knee” and the model is photographed from the back, the nose is not visible in the frame. A conventional crop tool fails here, produces a wrong crop, or skips the image. Autophoto’s AI is trained on human body skeleton and pose structure — it estimates where that landmark’s position would be and crops precisely from that estimated position.
Crop set: Head → Knee. Model facing backward. The AI detects the back of the head, estimates full head position including top, and crops down to the estimated knee position based on body proportions. Output: correctly framed crop identical in position to a forward-facing model with the same rule.
Crop set: Head → Knee. Model facing backward. Face detection finds no face. Tool either errors out, crops from the top of the image frame, or applies a fallback generic crop. Result: inconsistent output requiring manual correction.
Crop set: Nose → Bottom. Model in side-angle pose. Nose is a partial profile. AI detects side-profile nose position from face skeleton geometry and crops correctly. Same pixel-position relationship to the frame as a front-facing model.
Crop set: Nose → Bottom. Model in side-angle pose. Face recognition may detect a partial face but cannot reliably place the nose at the correct position. Crop line drifts from image to image across different side-angle poses.
Crop set: Head → Shoes. Action pose — model walking, one leg forward. AI understands the dynamic body position, identifies both feet from body skeleton, and crops to the forward shoe sole. Consistent output across the action sequence.
Crop set: Head → Shoes. Action pose. Without full body skeleton understanding, the tool may miss the forward foot, crop to the wrong leg, or use the image boundary as the lower crop edge.
Multiple crop outputs from the same image — in one pipeline run
Because landmark cropping is a pipeline task inside a deliverable, different deliverables can apply different crop rules to the same image. One upload can produce multiple cropped versions simultaneously — each formatted for a different channel, marketplace, or use case.
Example: one model image, three deliverables
Upload one full-body on-model image. Configure three deliverables:
Different output formats and dimensions across deliverables carry no additional token charge.
Why this matters for catalog operations
Most fashion brands need multiple crop standards simultaneously: their own website uses a different frame than Amazon, which differs from Zalando, which differs from their own lookbook. With Autophoto, all of those outputs come from one batch upload. No re-uploading the same images with different settings. No running the same batch three times through different tools.
What landmark cropping is used for in real production workflows
Six of the most common studio workflows where landmark cropping replaces manual Photoshop work entirely.
Unrecognisable face crop
Many wholesale and B2B marketplaces require model faces to be cropped out. Nose → Bottom or Eyes → Bottom produces a consistent headless crop across your entire catalog. Works correctly on front, side, and back poses without manual masking or separate processing.
Platform-specific frame rules
Amazon requires a different frame to Zalando. Your own website uses a different crop to your wholesale catalog. Set the correct landmark pair per deliverable — each channel gets its exact crop automatically from the same upload.
Consistent upper portrait
Tops, blouses, knitwear, and jackets all need an upper body crop. Head → Bottom or Chin → Bottom applied consistently across every image in the category — regardless of whether the model is standing, turning, or at a three-quarter angle.
Bottom and legwear framing
For trousers, skirts, and dresses where the lower body is the focus, a Top → Shoes or One-piece → Shoes crop frames the garment correctly without including unnecessary upper body space.
Full-body for outerwear and dresses
Head → Shoes or Head → Human gives a clean full-body frame that includes the complete garment — particularly important for coats, maxi dresses, and full-length outerwear where showing the hem is essential.
Product detail crops
For close-up deliverables showing collar, cuff, or pocket details, landmark pairs like Chin → Top or Eyes → Top frame the garment detail at the top of the body without including unnecessary lower body.
Landmark cropping inside the pipeline — stacked with other tasks
Cropping in isolation solves one problem. Cropping as part of a pipeline — alongside background removal, shadow generation, and format export — solves the entire post-production workflow in one run.
Upload batch
Mixed poses, all angles
BG removal
2 tokens
Shadow
2 tokens
Landmark crop
1 token
Export
JPG + PNG + PSD
Download
Folder structure intact
Combines with AI shot-type detection
Autophoto’s AI detects whether each image is on-model, off-model, or on-mannequin before processing. Landmark cropping can be configured to apply only to on-model images — off-model flat-lays route to a different deliverable with a product-centred crop rule. All from one mixed batch upload.
Combines with meta rules
Apply different crop rules based on folder name or file name. Images in the “tops” folder get a Head → Bottom crop. Images in the “outerwear” folder get a Head → Shoes crop. Name patterns become routing logic — no manual file separation before uploading.
How Autophoto’s landmark cropping compares to other AI crop tools
Several tools offer body-based cropping for fashion images. The critical differences are in how precisely you can define the crop range and whether the tool handles poses where landmarks are not directly visible.
| Tool | How cropping works | From–To range | Occluded estimation | Back / side poses | Pipeline integration |
|---|---|---|---|---|---|
| crop.photo | AI face recognition selects a face or body marker as the single crop anchor point. You choose one marker. The tool crops from that marker downward to image edge or a set dimension. | Single point | Partial | Yes for face marker | Partial |
| Zyng AI | Body-aware cropping that detects and preserves key body parts while adjusting composition. You define parameters for which body area to focus on. Batch processing available. | Area-based | Not documented | Partial | No |
| autoRetouch | Cropping as a workflow component. Users can define crop parameters within a workflow sequence. Fashion-focused. ~100 image batch limit. | Not landmark-based | No | Partial | Partial |
| Cloudinary | Smart crop API for developers — detects faces and objects, crops to keep subject in frame for responsive images. Developer/CDN tool, not an ecommerce post-production platform. | No | No | Partial | No |
| ✦ Autophoto | 11 named body landmarks, each usable as From and To boundaries of a precise crop range. AI estimates landmark position from body skeleton and pose even when not visible. Works across front, back, side, and action poses. Multiple crop deliverables from one upload in one pipeline run. 5,000+ images per batch. | From–To range | Yes — all 11 | Yes — all poses | Full pipeline |
From–To landmark range — not just a crop anchor
crop.photo and others give you a single crop anchor point — one marker you crop from, with the lower boundary set by image edge or a fixed pixel dimension. Autophoto gives you two named landmarks: you define both where the crop starts and where it ends. “From nose, To bottom garment hem” is a precisely bounded crop that holds consistently regardless of model height, pose, or image aspect ratio.
Occluded landmark estimation across all 11 points
crop.photo supports back and side pose detection for face markers. Autophoto’s body skeleton model estimates position for all 11 landmarks — not just facial landmarks — across every pose including action shots. “From nose / To shoes” on a backward-facing running model produces a correct crop. No competitor documents this capability across the full landmark set.
Frequently asked questions about AI landmark cropping in Autophoto
AI landmark cropping uses body skeleton detection to identify named anatomical points — head, eyes, nose, chin, knee, shoe, and more — and crops an image precisely between two of those points. In Autophoto, you define a From landmark and a To landmark; the AI detects where each point sits in the image and crops at those exact boundaries, consistently across every image in the batch.
Body-aware cropping uses AI to detect specific body parts and apply a crop that respects anatomical boundaries — rather than a fixed pixel crop from the image edge or center. Regular auto-crop resizes from the center or edge of the image regardless of where the subject is. Body-aware cropping understands what is in the image and frames the subject correctly relative to their body structure.
Yes. Autophoto’s AI is trained on full body skeleton and pose structure. When a landmark is not directly visible — because the model faces away, is in a side pose, or is in an action stance — the AI estimates where that landmark’s position would be in three-dimensional space based on body proportions and pose geometry, and crops precisely from that estimated position.
Autophoto has 11 body landmarks: Head, Eyes, Nose, Lips, Chin, Top, One-piece, Hand, Bottom, Shoes, and Human. Every landmark can be used as either the upper (From) or lower (To) boundary of your crop. You define both points — for example, From Nose, To Shoes — and the AI crops between them consistently across your batch.
Landmark cropping costs 1 token per image. Output format (JPG, PNG, WebP, PSD, TIF) and output dimensions carry no additional token charge — you can produce multiple formats and sizes from the same image without extra cost.
Yes. Each pipeline deliverable has its own landmark crop rule. One upload can produce a full-body crop for your website, a nose-to-hem marketplace crop, and a head-to-knee portrait for social media — all from the same image, in the same batch run.
crop.photo uses a single face/body marker as the crop anchor point — one point from which the crop starts, with the lower edge at image boundary or a fixed dimension. Autophoto defines a From and a To landmark, creating a precisely bounded crop range with both an upper and lower anatomical boundary. Autophoto also estimates landmark position for all 11 landmarks when they are not visible, not just facial markers.
Yes. Setting From Nose / To Bottom or From Eyes / To Bottom consistently crops the face out of the frame while preserving the complete garment. The crop works correctly on front, side, and back poses — so a batch of mixed-angle model images all produce compliant headless crops without manual per-image adjustment.
Yes. Landmark cropping is a task inside your pipeline deliverable, alongside background removal, shadow generation, mannequin removal, and format export. All tasks run from one upload in one automated batch run. You do not need to crop separately and then remove backgrounds separately. The complete processed output is delivered in your original folder structure.
200 free tokens to start
Set up a landmark crop pipeline and run your first batch — front, back, and side poses together. Every new account starts with 200 free tokens, no card required.
Pair with
Background removal · Natural shadow · Ghost mannequin · Multi-format export
Compare
Autophoto vs crop.photo · Autophoto vs Zyng AI · Autophoto vs autoRetouch
High volume?
Processing 50,000+ model images monthly? Enterprise plans with custom token pricing available.